The Impact of Automatic Store Replenishment Systems on Retail

DISSERTATION
of the University of St. Gallen Graduate School of Business Administration, Economics, Law and Social Sciences (HSG) to obtain the title of Doctor of Business Administration

submitted by

Alfred Angerer
from Austria

Approved on the application of Prof. Dr. Daniel Corsten and Prof. Fritz Fahrni, PhD

Dissertation no. 3123

The University of St. Gallen, Graduate School of Business Administration, Economics, Law and Social Sciences (HSG) hereby consents to the printing of the presented dissertation, without hereby expressing any opinion on the views herein expressed. St. Gallen, November 17, 2005

The President

Prof. Ernst Mohr, PhD

dedicado a las dos mujeres ms importantes de mi vida: mi madre y Anne

VI

VII

Foreword
Fast moving consumer goods retailing is a highly competitive market. European retailers are continuously aiming to improve customer loyalty by offering good service. At the same time, they are struggling to reduce costs in order to stay competitive. One technology that promises to decrease the number of out-of-stocks while simultaneously reducing store handling costs is automatic store replenishment (ASR). At the heart of ASR systems lies software that automatically places an order to replenish stocks. Many European grocery retailers have started to implement such decision support systems.

Surprisingly, although several retailers have automated their order process in the last few years, there is almost no academic source examining this topic at the level of the store. It is worth noting that other technologies in retail, such as RFID (Radio Frequency Identification) and the introduction of the barcode, have received far greater attention from the public and from researchers. Furthermore, while the topic of extent and root-causes of retail out-of-stock has received substantial interest over the course of the last years, the question to what extent existing and new practices remedy OOS is largely unanswered. In particular, there is a debate whether ASR improve or worsen OOS. Therefore, Dr. Alfred Angerer has well chosen a topic of both managerial and academic relevance.

Although there are many success stories from practitioners describing the enormous advantages of introducing automatic store replenishment systems there has been limited empirical proof of this. To the best of my knowledge no conceptual framework exits that can help practitioners to choose an adequate automatic replenishment system. In order to develop such a model research on relationship between replenishment performance (e.g. OOS rate, inventory levels) and contextual variables (such as store and product characteristics) is required. Finally, it is not clear how retailers have to adapt its organization and processes to best support the chosen ASR system.

Dr. Angerer confidently identifies and covers several research gaps and manages to give answers to this research gaps by a skilful combination of quantitative and qualitative research methodologies. In a first part an exhaustive data set of a European retailer is examined. With this data analysis the performance of replenishment system before and after the introduction of ASR systems is compared.

VIII

Dr. Angerer is able to statistically prove and quantify the positive impact of such systems on inventory levels and out-of-stock rates. In the second part, several case studies illustrate how ASR systems are implemented in practice. The given recommendations on store processes help retailers to benefit most from automatic replenishment systems.

Overall, this thesis makes an important contribution to the field of retail operations in practice and theory. I personally wish Dr. Angerer's work wide attention in both academic and practitioner circles.

Prof. Dr. Daniel Corsten

IX

Acknowledgment
Rarely is a doctoral thesis the contribution of a single person. Many people supported and consulted me during my three years of research at the University of St.Gallen. Therefore, I would like to express my thanks to everyone who supported me in finalising this work. I am greatly indebted to my two advisors, Prof. Dr. Daniel Corsten and Prof. Dr. Fritz Fahrni. They guided me through the inevitable ups and downs that characterise such a long research project. I want to specially thank Daniel Corsten who supported my research from the very outset. Without his never ending striving for improvement, the present results would not have been gained. I also cordially thank my second adviser, Fritz Fahrni, who always helped me to look for the big picture in my work. Further, I am thankful to Prof. Dr. Frank Straube and Prof. Dr. Wolfgang Stlzle for their backup as directors of the KLOG. A special thanks goes to my colleagues Lars Dittmann and Christian Tellkamp, with whom I shared several "research camps". They decisively influenced my research. Jens Hamprecht deserves a special thank for his numerous advices as well as Dorothea Wagner does. Her extensive knowledge about the consumer goods industry was always a valuable contribution. My time a the University of St.Gallen would only have been half the fun without my colleagues. I want to thank Gunther Kucza, Marion Peyinghaus, Jrg Hofstetter, Jan Felde, Jan Frohn, Elias Halsband, Florian Hofer, Petra Seeger, Dirk Voelz and all other colleagues at the Kuehne-Institute for Logistics and the Institute for Technology Management for being such good colleagues and for all the good moments we shared. Throughout the last years, I received valuable contributions form researchers and students. Especially, I want to thank Johanna Smros and Michael Falck from the HUT for the enriching discussions and research projects I shared with them. I would also like to express my thanks to all the students whose bachelor and master thesis I coached. Their interviews provided a basic foundation for my research. Without the support from practitioners, this research would only have been a theoretical contribution. A very warm thank you to Roland S., who invested his time to provide me the data for the quantitative research. Further, I am very grateful to

Marianne S., Daniel B. and all the interview partners for the time they invested in my research project. Finally, I want to thank my mother and father, Toni, Lydia, Nic and Anne for their never-ending moral support. Despite the distance, I always felt their affection throughout my education and career. St. Gallen, November 2005 Alfred Angerer

XI

Content Overview
1. Introduction ................................................................................................................................... 1 1.1. 1.2. 1.3. 1.4. 1.5. 1.6. Logistics Contribution to Retail Excellence.............................................................................. 1 Excellence in Store Operations ............................................................................................... 3 New Technologies Enable Automatic Store Replenishment Systems..................................... 6 Research Deficit ...................................................................................................................... 8 Research Questions .............................................................................................................. 12 Thesis Structure .................................................................................................................... 14

2. Research Framework and Design ............................................................................................. 16 2.1. Research Framework ............................................................................................................ 16 2.2. Research Methodology ......................................................................................................... 18 2.3. Research Process ................................................................................................................. 23 3. Literature Research .................................................................................................................... 26 3.1. 3.2. 3.3. 3.4. 3.5. Inventory Management Perspective ...................................................................................... 26 Logistics and Operations Management Perspective ............................................................. 30 Business Information Systems Perspective .......................................................................... 41 Contingency Theory Perspective........................................................................................... 47 Literature Research Overview ............................................................................................... 51

4. Development of Models.............................................................................................................. 53 4.1. A Descriptive Model of Replenishment Systems ................................................................... 53 4.2. Classification of Automatic Replenishment Systems............................................................. 64 4.3. Explanatory Model................................................................................................................. 68 5. Quantitative Analysis.................................................................................................................. 85 5.1. 5.2. 5.3. 5.4. Sample and Methodology...................................................................................................... 85 Hypothesis Testing: Dataset1................................................................................................ 99 Dataset2: Pretest/Posttest................................................................................................... 122 Quantitative ResearchConclusions ................................................................................... 128

6. Field Research and Managerial Implications.......................................................................... 132 6.1. 6.2. 6.3. 6.4. 6.5. Research Sample................................................................................................................ 132 Replenishment Processes................................................................................................... 137 Organizational Changes and Personnel Issues .................................................................. 151 ASR Performance ............................................................................................................... 157 Lessons Learned and Recommendations for Management ................................................ 161

7. Conclusion ................................................................................................................................ 181 7.1. Theoretical Contributions..................................................................................................... 181 7.2. Contribution for Practitioners ............................................................................................... 183 7.3. Further Research Fields ...................................................................................................... 186 8. Appendix and References ........................................................................................................ 189 8.1. Statistical Appendix ............................................................................................................. 189 8.2. References .......................................................................................................................... 193 8.3. List of Interviews.................................................................................................................. 210

XII

XIII

Table of Contents
1. Introduction ........................................................................................................................1 1.1. 1.2. 1.3. 1.4. 1.5. 1.6. Logistics Contribution to Retail Excellence ...................................................................1 Excellence in Store Operations.....................................................................................3 New Technologies Enable Automatic Store Replenishment Systems ..........................6 Research Deficit............................................................................................................8 Research Questions....................................................................................................12 Thesis Structure ..........................................................................................................14

2. Research Framework and Design ..................................................................................16 2.1. Research Framework..................................................................................................16 2.2. Research Methodology ...............................................................................................18 2.3. Research Process.......................................................................................................23 3. Literature Research .........................................................................................................26 3.1. Inventory Management Perspective............................................................................26 3.1.1. Optimization in Inventory Management Research ...............................................27 3.1.2. Theoretical Sources on OOS ...............................................................................28 3.1.3. Contributions and Deficits of an Inventory Management Perspective..................29 3.2. Logistics and Operations Management Perspective...................................................30 3.2.1. Supply Chain Management and ECR ..................................................................31 3.2.2. Automatic Replenishment Programmes...............................................................33 3.2.3. Operations Management in Retail........................................................................39 3.2.4. Contributions and Deficits of a Logistics and Operations Management Perspective ..........................................................................................................40 3.3. Business Information Systems Perspective ................................................................41 3.3.1. Characteristics of ERP Systems ..........................................................................42 3.3.2. ERP Implementation and Selection .....................................................................42 3.3.3. ERP and Human Agency .....................................................................................44 3.3.4. Contributions and Deficits of a Business Information Systems Perspective ........46 3.4. Contingency Theory Perspective ................................................................................47 3.4.1. Contingency Theory at the Organizational Level .................................................47 3.4.2. Contingency Theory on Information Technology and Processes.........................49 3.4.3. Contributions and Deficits of a Contingency Perspective ....................................50 3.5. Literature Research Overview.....................................................................................51 4. Development of Models...................................................................................................53 4.1. A Descriptive Model of Replenishment Systems ........................................................53 4.1.1. Inventory Visibility ................................................................................................56 4.1.2. Replenishment Logic............................................................................................57 4.1.3. Order Restrictions ................................................................................................60 4.1.4. Forecasts .............................................................................................................61 4.2. Classification of Automatic Replenishment Systems ..................................................64 4.3. Explanatory Model ......................................................................................................68 4.3.1. Purpose and Structure of the Explanatory Model ................................................68 4.3.2. Hypothesis Development: Product Characteristics ..............................................71 4.3.3. Hypothesis Development: Store Characteristics..................................................77 4.3.4. Hypothesis Development: ASR Characteristics ...................................................81

XIV

5. Quantitative Analysis...................................................................................................... 85 5.1. Sample and Methodology........................................................................................... 85 5.1.1. Dataset1: Testing of Out-of-Stock Hypotheses ................................................... 85 5.1.2. Dataset2: Pretest/Posttest Analysis .................................................................... 93 5.2. Hypothesis Testing: Dataset1 .................................................................................... 99 5.2.1. Influence of Product Characteristics.................................................................... 99 5.2.2. Influence of Store Characteristics ..................................................................... 113 5.3. Dataset2: Pretest/Posttest........................................................................................ 122 5.4. Quantitative ResearchConclusions ........................................................................ 128 6. Field Research and Managerial Implications.............................................................. 132 6.1. Research Sample..................................................................................................... 132 6.1.1. Company Selection ........................................................................................... 132 6.1.2. Market Characteristics....................................................................................... 134 6.1.3. Supply Chain Structure ..................................................................................... 134 6.1.4. Chains and Store Formats ................................................................................ 136 6.1.5. Delivery Frequency and Order-to-Deliver Lead Times ...................................... 136 6.2. Replenishment Processes........................................................................................ 137 6.2.1. Inventory Visibility.............................................................................................. 138 6.2.2. Forecasts and Replenishment Logic ................................................................. 146 6.2.3. Order Restrictions ............................................................................................. 151 6.3. Organizational Changes and Personnel Issues ....................................................... 151 6.3.1. Structural Changes and Setup .......................................................................... 152 6.3.2. Personnel and Change Management................................................................ 154 6.4. ASR Performance .................................................................................................... 157 6.4.1. Performance Measurement ............................................................................... 157 6.4.2. Inventory Level Performance ............................................................................ 158 6.4.3. OOS Reduction and Overall Performance ........................................................ 159 6.5. Lessons Learned and Recommendations for Management..................................... 161 6.5.1. The Adequate Automation Level: Recommendations ....................................... 161 6.5.2. ASR Introduction ............................................................................................... 165 6.5.3. Technical and Organizational Requirements .................................................... 166 6.5.4. Store Operations: Recommended Action.......................................................... 170 6.5.5. Cost-Benefit Analyses ....................................................................................... 175 7. Conclusion..................................................................................................................... 181 7.1. Theoretical Contributions ......................................................................................... 181 7.2. Contribution for Practitioners.................................................................................... 183 7.3. Further Research Fields........................................................................................... 186 8. Appendix and References ............................................................................................ 189 8.1. Statistical Appendix .................................................................................................. 189 8.1.1. Calculation of the Inventory Level ..................................................................... 189 8.1.2. ANOVA Considerations and Prerequisites........................................................ 191 8.2. References ............................................................................................................... 193 8.3. List of Interviews....................................................................................................... 210 8.4. Curriculum Vitae....................................................................................................... 212

XV

List of Abbreviations and Acronyms

ANOVA ARP ASR ASRx CD CU CU/TU CRP CPFR EAN ECR EDI ERP DC DSD DSS HQ IS IT ITEM KLOG KPI MAD MAPE OOS OR OSA PC PDA POS QR SC SCM SKU TU VMI Analysis Of Variance Automatic Replenishment Programme Automatic Store Replenishment Automatic Store Replenishment System level x Cross-Docking Consumer Unit Consumer Unit per Trading Unit (=case pack size) Continuous Replenishment Planning Collaborative Planning Forecasting and Replenishment European Article Numbering Efficient Consumer Response Electronic Data Interchange Enterprise Resource Planning Distribution Centre Direct Store Delivery Decision Support System Headquarters Information System Information Technology Institute for Technology Management Kuehne-Institute for Logistics Key Performance Indicator Mean Absolute Deviation Mean Absolute Percent Error Out-Of-Stock Operations Research On-Shelf Availability Personal Computer Personal Digital Assistant (handhelds) Point Of Sales Quick Response Supply Chain Supply Chain Management Stock Keeping Unit Trading Unit Vendor Managed Inventory

XVI

XVII

Figures
Figure 1: Figure 2: Figure 3: Figure 4: Figure 5: Figure 6: Figure 7: The importance of logistics for different industries ..................................... 2 Percentage of logistics costs on total costs by industry (in %) ................... 3 Summary of OOS root causes.................................................................... 5 Thesis structure ........................................................................................ 15 Focus of research ..................................................................................... 16 Integrative research procedure................................................................. 19 Case study research as iterative process between theory and empiricism................................................................................................. 22 Figure 8: Research activities in this research project .............................................. 23 Figure 9: Spectrum of misfit resolution strategies.................................................... 44 Figure 10: Descriptive model of replenishment systems ........................................... 54 Figure 11: Exemplary time dependent course of the inventory stock level................ 56 Figure 12: Qualitative and quantitative forecasting techniques ................................. 62 Figure 13: Classification of automatic replenishment systems .................................. 65 Figure 14: Overview of hypotheses, product characteristics ..................................... 83 Figure 15: Overview of hypotheses, store characteristics ......................................... 84 Figure 16: Overview of hypotheses, ASR characteristics .......................................... 84 Figure 15: Distribution of the 84 products in Dataset1............................................... 88 Figure 16: Comparison of the replenishment systems by store................................. 92 Figure 17: Estimated OOS (order-related) rate by sales coefficient of variance...... 100 Figure 18: Estimated inventory range of coverage by sales coefficient of variance ................................................................................................. 102 Figure 19: Estimated OOS (order-related) rate by speed of turnover...................... 103 Figure 20: Estimated inventory range of coverage by speed of turnover ................ 104 Figure 21: Estimated OOS (order-related) rate by price .......................................... 105 Figure 22: Estimated inventory range of coverage by price..................................... 106 Figure 23: Estimated OOS (order-related) rate by CU/TU group............................. 107 Figure 24: Estimated Inventory range of coverage by case pack ............................ 108 Figure 25: Estimated OOS (order-related) rate by product size .............................. 109 Figure 26: Estimated inventory range of coverage by product size ......................... 110 Figure 27: Estimated OOS (order-related) rate by shelf life..................................... 111 Figure 28: Estimated inventory range of coverage by shelf life ............................... 112 Figure 29: Estimated OOS (order-related) rate by store.......................................... 114 Figure 30: Estimated inventory range of coverage by store .................................... 115 Figure 31: Estimated inventory coefficient of variance by store............................... 116 Figure 32: Relationship between shrinkage and OOS............................................. 117 Figure 33: Relationship of OOS and SKU density ................................................... 118

XVIII Figure 34: Relationship of OOS and number of personnel per m2 ...........................118 Figure 35: Relationship of OOS and number years of the store manager working in the store.................................................................................119 Figure 36: Relationship of OOS and the size of the backroom ................................120 Figure 37: Relationship of OOS and customer satisfaction......................................121 Figure 38: Mean inventory range of coverage in days of dairy products and Control1 ..................................................................................................123 Figure 39: Means of the repeated ANOVA on the coefficient of variance of the stock level, ASR3 group and Control1 .............................................125 Figure 40: Mean inventory range of coverage in days of non-food products and Control2.............................................................................126 Figure 41: Estimated means of the repeated ANOVA on the inventory range of coverage, ASR2 group............................................................................127 Figure 42: Estimated means of the repeated ANOVA on the inventory range of coverage, ASR2* group............................................................127 Figure 43: Mean inventory coefficient of variance in days of ASR2, ASR2* group and Control2 ...........................................................................................128 Figure 44: Supply chain structure of sample ............................................................135 Figure 45: Delivery frequency of sample ..................................................................137 Figure 46: Inventory storage places and product flow processes ............................138 Figure 47: Comparison of inventory records and real inventory in one store ...........142 Figure 48: Decision tree for practitioners .................................................................162 Figure 49: Cost of forecasting versus cost of inaccuracy .........................................168 Figure 50: Overview of store operations recommendations .....................................170 Figure 51: Comparison of inventory on shelf and total store inventory for a glue stick.................................................................................................172 Figure 52: Costs in relation to replenishment level ..................................................177 Figure 53: Theoretical contribution of thesis ............................................................181 Figure 54: Contribution for practitioners ...................................................................183 Figure 55: Relative inventory level curve without zero line ......................................189 Figure 56: Absolute inventory level curve after the correction .................................190

XIX

Tables
Table 1: Overview of research deficits................................................................... 12 Table 2: Overview of basic theoretical sources reviewed (excerpt)....................... 26 Table 3: Implementation of ARP-related items ...................................................... 35 Table 4: Effectiveness in achieving automatic replenishment-related goals.......... 36 Table 5: Information systems capabilities .............................................................. 37 Table 6: Summary of research streams perspectives............................................ 52 Table 7: Inventory notations................................................................................... 55 Table 8: Basic inventory decision rules.................................................................. 57 Table 9: Exemplary order restrictions .................................................................... 60 Table 10: Characteristics of automatic replenishment levels................................... 68 Table 11: Overview of hypotheses concerning product characteristics ................... 77 Table 12: Overview of hypotheses concerning store characteristics ....................... 81 Table 13: Overview of the utilization of the two datasets for hypothesis testing...... 85 Table 14: Overview of variables used in the analysis .............................................. 90 Table 15: Product characteristics of Dataset1 by replenishment system ................ 91 Table 16: OOS rates (order-related)of the sample .................................................. 92 Table 17: Inventory range of coverage of Dataset1................................................. 93 Table 18: Dataset for the pretest/posttest................................................................ 94 Table 19: Descriptive statistics of the dairy products (ASR3) and Control1 group (ASR0)........................................................................................... 95 Table 20: Descriptive statistics of the beauty and household group (ASR2) and Control2 (ASR0) ...................................................................................... 97 Table 21: Descriptive statistics of the non-food group (ASR2*) and Control2 (ASR0) ..................................................................................................... 97 Table 22: ASR level and sales coefficient of variance ANOVA on OOS (order-related)........................................................................................ 100 Table 23: ASR level and sales coefficient of variance ANOVA on inventory range of coverage.................................................................................. 101 Table 24: ASR level and speed of turnover ANOVA on OOS (order-related)........ 103 Table 25: ASR level and speed of turnover ANOVA on inventory range of coverage ................................................................................................ 104 Table 26: ASR level and price ANOVA on OOS (order-related)............................ 105 Table 27: ASR level and price ANOVA on inventory range of coverage ............... 106 Table 28: ASR level and CU/TU ANOVA on OOS (order-related)......................... 107 Table 29: ASR level and CU/TU on inventory range of coverage.......................... 108 Table 30: ASR level and product size ANOVA on OOS (order-related) ................ 109 Table 31: ASR level and product size ANOVA on inventory range of coverage.... 110

XX

Table 32: Regression of shelf life and shelf life squared on OOS ..........................112 Table 33: Correlation between OOS per week and store characteristics (Dataset1)...............................................................................................113 Table 34: ASR level and Store ANOVA on OOS (order-related) ...........................114 Table 35: ASR level and Store ANOVA on inventory range of coverage...............115 Table 36: ASR level and store ANOVA on inventory coefficient of variance..........116 Table 37: ASR level ANOVA on OOS (order-related) ............................................121 Table 38: ASR level ANOVA on inventory range of coverage ...............................122 Table 39: Performance of ASR3 group compared to the Control1 (ASR0)............124 Table 40: Repeated ANOVA on inventory range of coverage, ASR3 group ..........124 Table 41: Repeated ANOVA on the coefficient of variance of the stock level, ASR3 group ..................................................................................125 Table 42: Repeated ANOVA on mean inventory range of coverage, ASR2 group............................................................................................126 Table 43: Results overview: product characteristics hypotheses ...........................129 Table 44: Results: store characteristics hypotheses ..............................................130 Table 45: Results: ASR hypotheses.......................................................................130 Table 46: Overview of the results of the hypotheses tested...................................131 Table 47: Selected companies for the field research .............................................133 Table 48: Inventory range of coverage of European grocery retailers in days .......158 Table 49: Technical requirements and recommendations on operations and organization structure.............................................................................169 Table 50: Overview of possible benefits and costs following an ASR system introduction ................................................................................177 Table 51: Overview of further research opportunities ............................................186

XXI

Abstract
European fast moving consumer goods retailers face a mature market with low margins and high competition. To improve their situation, retailers are looking for technologies and concepts to increase consumer satisfaction while at the same time reducing costs. One technology that promises to increase the availability of the products on the shelf while simultaneously reducing store handling costs is automatic store replenishment (ASR). At the heart of ASR systems lies software that automatically places an order to replenish stocks of a certain product. A majority of European grocery retailers have implemented such decision support systems. Yet research in this area is practically non-existent. Therefore, this thesis aims to investigate the impact of this technology on retail, taking into account financial, organizational and personnel aspects. To answer this main research question, a quantitative and a qualitative methodology was chosen. First of all, based on theoretical sources and more than 50 interviews, a descriptive model and an ASR classification system is developed. Next, an explanatory model is developed with a view to enabling identification of the characteristics of products, stores and replenishment systems that influence the replenishment performance of retail stores. To be able to test the hypothesis derived from this explanatory model, exhaustive data from a grocery retailer is examined. The quantitative analysis clearly shows that even simple automatic replenishment systems are able to dramatically reduce the average shelf out-of-stock rate and at the same time lower inventory level. In addition, a major advantage of automatic systems over manual ones is that they show constant results, independently of product characteristics. Yet the analysis also shows that badly-parameterised automatic systems will fail to deliver the desired results. In order to better understand how ASR systems are best implemented in practice, four major grocery retailers are analysed in detail. These case studies illustrate the necessary technological and organizational changes and highlight the influence of ASR systems on the working behaviour of employees. Overall, this thesis makes contributions to both practice and theory. On the one hand, the results presented are a first stepping stone towards the creation of a basic theory of ASR systems. A descriptive model enables further researchers to make differentiated statements on the impact of ASR based on the classification developed. Another contribution is the explanatory model which tests existing and demonstrates new relationships hypothesised in inventory and operations management research. On the other hand, practitioners receive an overview of the

XXII

existent systems by which they may automate store replenishment. The determination of ASR benefits and necessary requirements help them to make a cost-benefit analysis. In addition, the several implications of the automation of their replenishment system for the organization and for human working patterns are illustrated. Practical recommendations on store processes help retailers to benefit most from automatic replenishment systems. And finally, a decision tree helps practitioners to identify the best-suited ASR system for each product category.

1. Introduction

1. Introduction
Grocery retailing is a highly competitive market (e.g. Keh and Park 1997). European retailers are continuously aiming to improve customer loyalty by offering good service. At the same time, they are struggling to reduce costs in order to stay competitive. The effort to achieve customer service excellence has only been partly successful, as the low average product shelf availability rates of 9295% (Gruen, Corsten et al. 2002; Roland Berger 2003b) and a sunk store loyalty underline. The major part of retailer costs are personnel costs, and in particular it is the operations in the store that require intensive staff dedication (Broekmeulen, van Donselaar et al. 2004a). The German retailer Globus has calculated that the logistics costs of the last 50 meters in the store, i.e. from the backroom to the shelf, are three times as expensive as the first 250 kilometres from the producer to the store gate (Shalla 2005). A technique that promises to reduce the out-of-stock (OOS) rate by simultaneously reducing the store handling costs are so-called automatic store replenishment (ASR) systems, the main research subject of this thesis. This chapter provides an introduction to the business challenges faced by retailers and the valuable role of logistics in retail, followed by a short introduction to ASR systems. Later, research deficits in the literature are identified and the research questions of this thesis are derived. Finally, an overview of the structure of this research study is given.

1.1. Logistics Contribution to Retail Excellence

The major market developments that make retail challenging started in the 1990s and still are prevalent today, namely high cost pressure, shorter innovation cycles, increasing consumer expectations and globalization (Baumgarten and Wolf 1993; Lee 2001). The common response of retailers has been a so-called quantity strategy: They introduced more product variants, invested in new channels of distribution, diversified store formats and expanded into new countries. However, the benefits harvested from such a strategy seem to have come to an end, as the market has become saturated. The fraction of private consumption that flows into food and nearfood retail has decreased continuously in the last two decades. In Germany, for example, it sank from 44.2% in 1990 to 29.3% in 2004 (Krber 2003), and this trend is typical for many developed countries. Nevertheless, a small group of retailers was able to defy this trend and outperformed the market. As a study by Accenture (2000) reports, approximately one-third of 63 examined retailers outperformed the other twothirds by far and showed a yearly revenue increase of at least 10% coupled with a

1. Introduction

higher-than-average increase in stock price. According to the study, this group had developed the right strategy by focusing their investments in areas where the most efficiency potentials were located. One of the areas with such potential is without doubt logistics, as effective and efficient logistics is the fundamental to successful retailing. Hans Joachim Krber (2003), CEO of Metro AG, describes logistics as "the physical accomplishment of the concern strategy." Figure 1 depicts the great importance of logistics for retail and various industry sectors under the aspects "differentiation" (i.e. logistics as a marketing tool) and "rationalisation" (i.e. logistics as a method of saving costs).
Retail/ consumer goods Chemical industry Machine building

Differentiation

Paper industry Plant construction Electronics

Automotive

Rationalisation
Figure 1: The importance of logistics for different industries
1

The importance of logistics for the retail sector is based on the nature of the products sold. Most consumer goods, for example daily food items, are relatively cheap and the consumer generally buys without lengthy quality or price comparisons. Nevertheless, the importance of logistics in other sectors is increasing as well, as Pfohl (2004) stresses.

Source: Kowalski (1992).

1. Introduction

A precise estimation of the logistics costs is rather difficult. Pfohl compared studies measuring the logistics costs as a percentage of turnover. The large differences in the results can often be explained by geographical differences between countries and their infrastructure levels. Yet even within a single country like Germany, there are several studies with significantly divergent figures. This is the result of the varying definition of logistics costs. One of the most cited studies is that by Baumgarten and Thoms (2002). They estimate the retailers' logistics costs at up to 27% of total costs (see Figure 2).

Automotive

Consumer goods

Retail

8.2%

7.6%

12.8%

12.2%

27.6%
2

26.7%

Figure 2: Percentage of logistics costs on total costs by industry (in %)

Even if other researchers have clearly lower estimations (e.g. Klaus 2003), there is a common agreement that there exists a large savings potential. Two studies from the year 1999 estimate the savings potentials at about 1225% (Baumgarten and Wolf 1993; European Logistics Association and A.T. Kearney 1999). In order to achieve these savings, new advanced logistics-technology is employed. But logistics should never be reduced to its cost-reducing effect, as logistics concepts can also be utilized to improve service and consequently increase sales (Angerer and Corsten 2004). The next section deals with one of the most important measures used to quantify customer-service levels: the on-shelf availability rate.3

1.2. Excellence in Store Operations

A high availability rate of products on the shelves is of utmost importance for retailers. All the efforts made to improve the supply chain are futile if, in the end, the consumer is unable to buy the product because it is not available on the shelf. There
2 3

Source: Baumgarten and Thoms (2002). The on-shelf availability rate is the percentage of products that are available for purchasing on the store's shelves at a particular moment in time.

1. Introduction

exist studies that show that out-of-stocks (OOS) in stores are the most frequently mentioned cause of frustration for dissatisfied customers in retail (Sterns, Unger et al. 1981). Interviews with practitioners confirm the importance of high shelf availability: "The three criteria that decide the success of a product are the right price, the right forms of advertisement and high on-shelf availability. (...) In particular, if there is a promotion, there is nothing more important than having the goods on the shelf!"4 Obviously, the impact of an OOS depends on the reaction of the customer: "The reaction of customers [on OOS] differs a great deal. If the customer buys a different brand, we are happy. If he or she does not buy anything at all, then we are not content. And if the customer buys the product in a competitor's store, that is a catastrophe! Seventy percent of customers change to the competition for good if they experience repeated OOS; and that is a complete catastrophe!"5 Furthermore, there is a strategic component to high shelf availability, as it ensures an advantage in increasingly competitive markets: "If we want to compete with new aggressive retailers such as LIDL which are planning to enter the Swiss market, we have to increase the turnover per square meter. For that, we need to increase the on-shelf availability (...) to make our stores more interesting for customers."6 The importance of a high availability is underlined by the research of Drze, Hoch et al. (1994) among others. They show that the total amount of money spent on any store visit is an elastic quantity and is highly dependent on product presentation and quantity on the shelf. Although the on-shelf availability rate plays such an important role in the business of retailers, it seems that only a minority of European grocery retailers systematically measures this important KPI (key performance indicator). A case study of 12 leading European grocery retailers has shown that only four companies have established a process for daily availability check (Smros, Angerer et al. 2004a). Only one retailer had implemented an electronic-based system for automatic checks. The magnitude of the OOS problem still appears not to have been identified by many retailers. They tend to derive the availability rate in their stores from the service level at their distribution centres (DCs). Their argument is that if the

4 5

Source: Arthur Mathys, Director Logistics, Denner, 04.08.2003. Source: Wolfgang Mhr, Director IT, Spar Switzerland, 16.02.2004. 6 Source: Wolfgang Mhr, Director IT, Spar Switzerland, 16.02.2004.

1. Introduction

DC can fulfil 99% of the store orders, then one can expect an on-shelf availability rate of 99%. This thought is not quite correct, as the work of Gruen, Corsten et al. (2002) demonstrates. Their meta-study proves that in the last few decades the OOS rate has not decreased. It seems to have remained rather stable at a level of about 8%. This high figure is rather surprising for manufacturers and retailers, as they expected a far better rate considering the progress made in technology and new logistic concepts introduced in the last few years. In order to tackle this problem, the study further examines the reasons for OOSs, as depicted in Figure 3.
Other cause 4% Retail HQ or manufacturer 14% Distribution centre 10% Store ordering 34%

Store forecasting 13%

Store shelving 25%

Figure 3: Summary of OOS root causes

Surprisingly, almost three-quarters of stock-outs are the direct result of retail store practices and shelf restocking processes. A very similar result was found in a KLOG project carried out with the European grocery retailer MYFOOD8. Sixty percent of the OOS situations at this retailer were caused by the ordering behaviour in the stores. In 10% of the cases the goods were in the store but not on the shelves. Here lies a possible answer for the ineffectiveness of existing ECR (Efficient Consumer Response) activities on the OOS rate. Many of these ECR activities concentrate on the smooth transportation of items up to the store gate. How replenishment orders are placed, how order quantities are determined and processes in the so-called last 50 metres in the store still remains an area for research. To see the financial implications of OOS incidents, Gruen, Corsten et al. (2002) conducted an estimate of the overall effect of OOS on sales that takes into account the response of consumers. The result is that on average, retailers might lose up to
7 8

Source: Gruen, Corsten et al. (2002). The name of this retailer and of three others that are examined in the case study section 6 have been made anonymous for confidentiality reasons.

1. Introduction

4% of their turnover due to SKUs (stock keeping units) being absent from the shelves.

1.3. New Technologies Enable Automatic Store Replenishment Systems

As half of all OOSs arise from incorrect ordering and forecasting processes, it is sensible to have a closer look at stores' replenishment processes and systems. Some decades ago, there was no alternative to manual store replenishment systems. A planner, for example the store manager, was responsible for deciding the two main parameters of replenishment systems, namely the amount to be ordered and the when to place the order. In order to do this, the planner had to check manually the quantity in stock. In the last decade, there has been an impressive diffusion of largescale information packages such as ERP (Enterprise Resource Planning) in organizations (Kallinikos 2004). In addition, identification technology (such as barcodes and scanners) and communication tools (such as EDI9) have become very cheap, their implementation and use is nearly routine (Kuk 2004). Today, these and other new technologies make it increasingly possible to automate the replenishment decision-making. The interviews conducted with practitioners as well as other surveys (Bearing Point 2003; Smros, Angerer et al. 2004a) clearly reveal a trend in retailing towards automating store replenishment processes: "The normal replenishment process has been until now consisted of store personnel deciding what quantity to order by looking on the shelf. Now, retailers want to let the systems take this decision."10 Semi-automatic systems merely support the planner in his decision, for example, by showing him electronically the inventory and order restrictions. Advanced automatic store replenishment systems are IT-based software systems that automatically decide when to order which quantity. Nevertheless, there are several differences in the complexity and performance of such ASR systems. The simplest systems just place an order as soon as an article is sold or when a certain minimum stock level is reached. No forecasts are made; the quantity to be ordered is calculated with a very simple algorithm (e.g. fill up to a certain level). This kind of automatic system is, for example, used by the Swiss retailers Mobile Zone, Marionnaud and Fust. One example of a complex, state-of-the art ASR system comes from the company SAF AG (Switzerland). The main advantage of their ASR software "SuperStore" is that it
9 10

Electronic Data Interchange. Source: Wolfgang Mhr, Director IT, Spar Switzerland, 16.02.2004.

1. Introduction

makes a separate forecast for every item in every store. This is in contrast to other software programs, which make their calculations at SKUs/stores clusters due to IT performance restrictions (Beringe 2002). Furthermore, such forecasts do not rely only on historic sales. Their sophisticated causal models also consider price, promotion, seasons, holiday and other events when predicting demand. The introduction of this product in the German over-the-counter chemist retailer dm-drogeriemarkt resulted in a 7080% reduction in OOS incidents and simultaneously reduced the inventory stock level by 1020% (Beringe 2002). A detailed classification of the various existing ASR systems is provided in section 4.1. Several technological developments were necessary to enable the implementation of such sophisticated replenishment systems: Electronic inventory systems Identification technologies (barcodes, scanners) Data warehousing capacities (for historical sales data) Electronic data interchange (EDI) IT computation power (for forecasts at SKU level) Enterprise Resource Planning (ERP) systems First, inventory management systems were introduced that made it possible to manage quantities of a product in the electronic systems. These electronic inventory systems profited markedly from identifying technologies such as the barcode. The order process was simplified by using fax and electronic connections (e.g. EDI) between companies. IT systems' storing capacity increased, making it possible to handle larger amounts of data. The storing of huge quantities of POS (point of sales) and inventory data became feasible with new data warehouses and storage mediums. Furthermore, not only was it internal data that was more easily accessible; thanks to larger communication bandwidths, it has became possible to access large quantities of external data as well. This new external data includes competitive information (e.g. the price of a competitor's products), market data (e.g. from marketing institutes) and collaborative data (e.g. collaborative forecasts with suppliers) (Beringe 2002). The increased IT-power performance has made it possible to calculate in fractions of a second increasingly complicated forecasts at SKU level. In a nutshell, IT-capabilities do not seem to be the decisive restriction anymore. This statement is underlined by a study by Sabath, Autry et al. (2001) which shows that the information system capabilities (such as timeliness of information or compatibility of the IT) of the surveyed manufacturers and retailers are on average on a rather high

1. Introduction

level.11 The authors conclude that these companies already have the basic requirements to operate automatic replenishment programmes. Yet, IT is only one step towards ASR systems; other questions concerning the organization and logistical processes arise. The introduction of ASR systems goes hand in hand with the implementation of large-scale ERP systems, and thus has radical implications for the organizations and processes of firms. For Kallinikos (2004, p. 8), the introduction of these packages marks "a distinctive stage in the history of computer-based information technology's influence in organizations." Their main achievement is the new possibility for integrating operations and information across functions, departments and modules. Therefore, the organizational and behavioural implications have to be considered. One example is the role of employees; the introduction of such automatic systems can result in a dramatic change of their working habits: "The changes resulting from the introduction of ASR systems are enormous. (...) It is a change of paradigm. Who has today the same job as 5 years ago? (...) Especially elderly employees have problems with the changes. Our planners have been doing their jobs for 25 years; one has to take this into account."12 The importance of ordering for the store employees can be seen on the following statement of a grocery employee, which altered Ren Descartes famous statement: "I order, therefore I am."

1.4. Research Deficit

What is the contribution of academic research in the field of ASR? Surprisingly, although several retailers have automated their order process in the last few years (Smros, Angerer et al. 2004a), there is almost no academic source examining this topic at the level of the store. It is worth noting that other technologies in retail, such as RFID (Radio Frequency Identification) and the introduction of the barcode, have received far greater attention from the public and from scholars.13 One explanation could be that ASR is a technology working in the background. If it works properly, consumers should notice it only indirectly, such as through higher availability in the stores. Yet this would not explain the interest received by other technologies, such as EDI, which also work in the background. To sum it up, the questions surrounding
11 12 13

See Table 5. Source: Stefan Gchter, DC-director, COOP, 16.02.2004. This statement can be illustrated by a look at the agenda of European ECR initiatives. There exist several working groups dealing with RFID and barcodes, but none that deals with automatic replenishment.

1. Introduction

ASR systems arising for practitioners and researchers can only be partially answered through a review of the literature. The few sources related to this topic are presented in the following.

General Inventory Management Literature The contributions that come from general inventory management literature are rather basic. Existing academic sources of inventory modelling sources seek to answer the two primary questions that arise when dealing with replenishment systems, namely when should which quantity be ordered (Wagner 2002). Many papers in the operations research (OR) field concentrate on the modelling of replenishment systems, and try to identify an optimum under certain conditions (see Groote 1994; Silver, Pyke et al. 1998; Bassok 1999; Gudehus 2001). Algorithms calculate minimal inventory levels by choosing the right order quantity and order point so that certain a priori set objectives are fulfilled (e.g. a certain percentage of service level, a maximum out-of-stock rate, etc.). In general, much of the inventory management literature remains very theoretical. The implementation of these models in daily business is rather difficult (Wagner 2002). Many simplifications are made. When calculating the optimum, the specific situation of retailers at store level is not taken into account. The critical costs of retailers at store level are not inventory holding costs, but handling costs, which can be between 35 times as large as the former (Broekmeulen, van Donselaar et al. 2004b). Yet, store handling costs are seldom taken into account in these mathematical models (cf. van Donselaar, van Woensel et al. 2004).

ASR Related Literature One of the few sources dealing directly with ASR systems is a dissertation published by Norman Gtz (1999). His main achievement has been to develop software that enables the automation of order placement. This program uses existing forecasting heuristics and combines them into a new one. The benefits of Gtz's program are shown with a simulation based on real data from two stores of a German drug retailer. The theoretical benefit compared to the old system is an average cost reduction of 14.5%, mostly from a reduction in the inventory holding costs. Gtz's simulation shows a strong effect of the automation on the OOS rate: Out-of-stocks are reduced by almost 80%. The time savings for the stores are calculated at about 5 hours per ordering day. The remarkable contribution of this thesis is that for the first

10

1. Introduction

time the benefits of such systems were calculated, at least in theory.14 Gtz stresses that one of the main advantages of the system is that retailers have the power to realize the described benefits on their own, independently from the rest of the market. Nevertheless, his work is only the very beginning of the research on this topic. The IT systems described in Gtz's work are no longer state-of-the art.15 The overall focus of his work is rather mathematical; the main goal is the development of an optimal forecasting heuristic. Consequently, many aspects in the context of ARP systems such as optimal implementation and organizational influences are not considered.16

Supply Chain Management As valuable as Gtz and other contributions from inventory management research are, they show the limits of focusing strongly on mathematical or IT aspects when dealing with ASR systems. An approach to the topic from a more abstract level could be helpful, as the implementation of ASR systems is a fundamental change in the way the flow of materials is triggered in a supply chain. Therefore, SCM research, which deals with the importance of having demand-based replenishment systems (pull systems), is examined in this thesis.17 One effect of such a pull-supply chain is a major increase in efficiency and performance (Fiorito, May et al. 1995; Cottrill 1997; Closs, Roath et al. 1998). Because the competition between grocery retailers is so fierce, Bell, Davies et al. (1997) regard pull-supply chains as a necessity for every retailer. A practical implementation of the idea of a demand driven supply chain are automatic replenishment programmes (ARP). Common ARPs are Vendor Managed Inventory (VMI), Continuous Replenishment Planning (CRP), Quick Response (QR) and Collaborative Planning, Forecasting and Replenishment (CPFR).18 Overall, the sources on this topic (e.g. Daugherty, Myers et al. 1999; Ellinger, Taylor et al. 1999; Myers, Daugherty et al. 2000; Sabath, Autry et al. 2001) show the major benefits of such programmes, because OOS and handling costs are reduced while the inventory turn increases.

14

Gtz (1999) used real data for his analysis. Yet, he did not prove that such systems would also work under real life conditions. 15 For example, in order to save computing power, the products are clustered into 4 groups that have a common forecast function. Today's systems have evolved rapidly in the last few years so that this constraint is no longer relevant for today's ERP systems. 16 Only once does Gtz acknowledge that performance could depend on the satisfaction of employees with the new software and their commitment to it (Gtz 1999, p. 186).
17

A pull system is driven by demand at the lowest point of the chain (Christopher 1998). In the context of this thesis this would be the shopper in the store. 18 See for an explanation of these concepts Christopher 1998, chapter 7; Seifert 2002; Alicke 2003, pp.168-169

1. Introduction

11

Although the research methodology and findings of these researchers concerning the performance and context influences of ARPs are remarkable, the transfer of their results to the ASR systems context is limited. The research deficit in this field is that the theoretical sources have concentrated up to now on pull systems from the manufacturer to the retailer's distribution centres. In the context of ASR systems, it is also necessary to consider the replenishing of stores. In ASR systems the demand is being driven by the shopper in the store.19 Myers, Daugherty et al. (2000) recommend focusing future research on replenishment automation in a single industry. Sabath, Autry et al. (2001, p. 103) further state that "the issue of information systems capabilities is vital as well and deserves further study." Furthermore, the authors point out the importance of making additional investigations concerning organizational structure for developing decision guidelines. All these research recommendations are considered in the conception of this thesis.

Overview of Research Deficits To sum up, more research on this topic is required because the subject of ASR has not received in theory the attention it deserves considering its importance for practitioners. The first deficit is that there is no academic source describing the different systems in use; thus, a descriptive model and a classification has to be developed. Furthermore, an examination of retailers has revealed that the introduction of ASR was often part of changes taking places in the entire ERP system, therefore significant financial and managerial inputs are necessary (Keh and Park 1997). Although there are many success stories from practitioners describing the enormous advantages of introducing automatic store replenishment systems (see, for example, Beringe 2002; Anderson 2004; Hopp and Arminger 2005) there has been only limited examination of such statements from academic sources. Consequently, it is not surprising that some of the retailers interviewed are sceptical about the sense of such systems. Even if practitioners are convinced as to the utility of ASR, they remain insecure on the question of which system to choose for different types of products. There is no conceptual framework available at the moment which would help practitioners to choose an adequate replenishment system. In order to be able to develop such a model more needs to be known about the relationship between replenishment performance (e.g. OOS rate, inventory levels) and contextual

19

For a detailed explanation of the limitations of ARP see section 3.2.2.

12

1. Introduction

variables (such as store and product characteristics). Finally, it is not clear how retailers have to adapt its organization and processes to best support the chosen ASR system. The research deficits in the context of ASR systems are summarized in Table 1. Research Deficits Concerning ASR Systems Missing: descriptive model and classification qualitative study on benefits knowledge about relationship between store and product characteristics and the performance of replenishment systems method determining necessary replenishment system for each product category knowledge on the change retailers' organization and processes
Table 1: Overview of research deficits

1.5. Research Questions

In the last sections it was demonstrated that numerous unknowns exist in the context of ASR systems. Therefore, this research aims to answer following main question: Q: Under which conditions can retailers benefit from automatic store replenishment systems? Practitioners (cf. Bearing Point 2003) often speak of automatic replenishment systems without taking into account that these systems can vary from very simple heuristic- based decision systems to highly sophisticated ones with self-optimization and complex forecasting models. This thesis aims to help practitioners choose the right system for their business. This implies a need for identification as well as categorization of the automatic replenishment systems available. For that, this thesis first develops a descriptive model from which a classification is derived. Consequently, the first sub-question, which provides support in answering the main research question is: Q1: How can automatic store replenishment systems be classified?

1. Introduction

13

The kind and magnitude of benefits discussed in theory and practice is very broad; therefore, they are addressed in this thesis in particular. For three of the possible benefits (fewer OOS incidents, lower inventory levels and lower inventory variability) quantitative examinations are carried out. The second sub-question in this thesis is: Q2: What benefits can a company expect from the implementation of replenishment systems? Practitioners want to know if the replenishment systems they are using are best suited for them. To be able to choose the right systems for a given business and product category, it is necessary to perceive the interrelations between the elements of such systems and to understand the influence of environmental setting on performance. Therefore, knowledge about the influence of store and product characteristics on the ASR system outcome is necessary. Consequently, the next sub-question is: Q3: How is the performance of ASR systems influenced by product and store characteristics? A new automatic replenishment system does not only influence the performance of replenishment it can also influence the entire distribution system, delivery frequencies and employee working behaviour. Consequently, the choice of a new replenishment system with all its implications for an organization is a strategic decision and will be one focus of this thesis. Practitioners need a methodology for choosing the right system. For this reason, the next sub-question is: Q4: Which ASR system is recommended given the characteristics of a certain product and retailer? It can be assumed that some companies will not have a system that adequately meets their needs. Consequently, the recommendation will be to implement another type of ASR system. For an automatic replenishment system to reach its full potential, next to technical requisites it is necessary to adapt the retailers' organization and internal processes. Therefore, the last question that arises and which will be examined is: Q5: Which intra-organizational aspects of a company have to be changed to adapt a new ASR system?

14

1. Introduction

1.6. Thesis Structure

This thesis is structured as follows: Chapter 1 highlights the major contribution of logistics for retailers and describes current challenges in retail. This chapter ends with the derivation of the principal research questions from the research deficits. The methodology for addressing these research questions is depicted in Chapter 2. This chapter sets out the research framework, the methodologies used to address the research questions and illustrates the activities undertaken by the author to accompany and guide the research process. Chapter 3 highlights the contribution of theoretical sources for this thesis. Four research perspectives are examined, namely inventory management, logistics and operations management, business information systems and contingency theory. The theoretical findings from this chapter, together with evidence from interviews, are the foundation of Chapter 4. In this chapter, hypotheses and models concerning automatic store replenishment systems are developed. First, a descriptive model and a classification structure for ASR systems are presented (Q1). Second, an explanatory model is developed that explores the correlation between the performance of retailers and the store, product and ASR system characteristics. The created hypotheses are tested in Chapter 5, in a quantitative analysis with the help of real inventory and sales data from a grocery retailer. One part of the statistical analysis shows how the OOS rate and the stock levels of the automated goods changed compared to a control group (Q2). Another part of the chapter illustrates the correlation between store/product characteristics and ASR performance (Q3). A major finding described in this chapter is that ASR systems are indeed beneficial for retailers, yet their contribution depends strongly on how they are implemented. Therefore, in Chapter 6 there is a description of how four European retailers have implemented sophisticated ASR systems. The case studies show the required organization and processes for successful ASR implementation (Q5). This chapter ends with several action recommendations for mangers as regards store operations and choice of system (Q4). Finally, Chapter 7 summarizes contributions to theory and practice. Figure 4 illustrates the structure of this thesis.

1. Introduction

15

1. Introduction 1. Introduction The challenge of store replenishment for retailers The challenge of store replenishment for retailers Research deficits and research questions Research deficits and research questions

2. Research Framework and Design 2. Research Framework and Design How can the research questions be answered? How can the research questions be answered? Result: Research framework and methodology Result: Research framework and methodology defines main focus of 3. Literature Research 3. Literature Research Four theoretical perspectives: inventory management, logistics/operations Four theoretical perspectives: inventory management, logistics/operations management, business information systems and contingency theory management, business information systems and contingency theory Result: Theoretical foundation of thesis Result: Theoretical foundation of thesis combined with interview results is the basis of 4. Development of Models 4. Development of Models Descriptive model and classification of ASR systems Descriptive model and classification of ASR systems Explanatory model Explanatory model Result: Models and hypotheses to describe and explain different ASR systems Result: Models and hypotheses to describe and explain different ASR systems developed hypotheses are tested in 5. Quantitative Analysis 5. Quantitative Analysis Sample and statistical methodology Sample and statistical methodology Testing of hypotheses Testing of hypotheses Result: Proof of the benefits of ASR systems Result: Proof of the benefits of ASR systems how to introduce ASR systems in practice is shown in 6. Field Research and Managerial Implications 6. Field Research and Managerial Implications Case studies from four European retailers Case studies from four European retailers Result: Required changes for successful ASR implementation Result: Required changes for successful ASR implementation Decision tree: adequate ASR system Decision tree: adequate ASR system Recommended store operations Recommended store operations

Q1 Q1

Q2 Q3 Q2 Q3

Q4 Q5 Q4 Q5

7. Conclusion 7. Conclusion Contribution to practice and theory Contribution to practice and theory Research limitations Research limitations 8. References & Appendix 8. References & Appendix
Figure 4: Thesis structure

16

2. Research Framework and Design

2. Research Framework and Design

The outcome of any research is strongly affected by the choice of the research methods and strategies. As Scandura and Williams state (2000, p. 1249) "[a]ny research method chosen will have inherent flaws, and the choice of that method will limit the conclusions." This means that design choices about instrumentation, data analysis and construct validation may affect the types of conclusions that are drawn (Sackett and Larson 1990). Therefore, this chapter gives an overview of the research methodology chosen for this thesis. First, a research framework narrows the research field down. Second, the qualitative and quantitative research methods are depicted before, finally, the research activities that led to this thesis are presented.

2.1. Research Framework

The research framework mainly focuses on the role of automatic replenishment systems for European grocery retailers. The main research field of this thesis is not the distribution system to the store itself, but the ordering logic that lies behind it (see also Figure 5). This means that the focus of the thesis is the logic of the system that decides at what time and in what quantity the goods are replenished in a store. The routes taken by the goods or their mode of transportation are only secondary in this context. Consequently, whenever in this thesis the term "supplier" is used, there is no differentiation between products that have come from a retailer's own distribution centre or directly from the manufacturer. The shop-floor logistics (i.e. how the goods are transported from the backroom to the shelves) are a relevant part of the research framework, especially when talking about changes in process due to ASR implementation.
Focus of thesis

Flow of information
(order-, inventory level-, POS-data ...)

PrePresupplier supplier

Supplier Supplier

DC retailer DC retailer Direct supplier Direct supplier

Store ordering Store ordering processes processes

(major European (major European grocery retailers) grocery retailers)

Customer Customer

Flow of material
(replenishment of products)

Figure 5: Focus of research

2. Research Framework and Design

17

The focus on Europe does not mean that such systems are not interesting for other retailers from other world regions as well. The OOS studies mentioned in the introduction show a similar level of operational problems all over the world (Gruen, Corsten et al. 2002; Roland Berger 2003b), and replenishment systems play an important role for North American retailers as well. A Bearing Point technology study (2003) shows that although 61% of the North American retailers studied use automatic replenishment systems, only 38% have a well-documented and established inventory management strategy and thus a clear view of how such systems are best implemented within the organization. As there are no fundamental differences in the logistics or store operations between these two regions, one can presume that the findings will be significant for American retailers as well.20 The same holds true for retailers from other regions so that the result of the European research will be most probably globally relevant. The grocery retailers group is among the most interesting ones in retail, as the leading grocery retailers have in the last decades broadened their categories to products that are not related to the food sector at all. Guptill and Wilkens (2002) describe a grocery store as a retail store with at least 1,500 different food items and/or $2 million in annual sales that sells dry grocery, canned goods and non-food items plus some perishable items. A grocer is, according to the Oxford English Dictionary, a "trader who deals in spices, dried fruits, sugar and, in general, all articles of domestic consumption except those that are considered the distinctive wares of some other class of tradesmen."21 This last definition does not seem to be precise enough nowadays. The leading retailers sell through their distribution channels almost all kinds of consumer goods that were some decades ago only available at certain speciality stores (e.g. apparel, computers, household goods, garden articles, etc.). The term modern grocery22 is sometimes used to separate the core business of groceries (food and near food) from more exotic products (e.g. financial services). This large variety of articles force retailers to have separate logistics strategies for their items, depending on various factors such as speed of turnover, value, demand volatility, etc. Some retailers have adopted different strategies and channels to distribute their products. There exists a large variance of stores, ranging from rather small supermarkets (200 m2) to huge hypermarkets (10,000 m2 and larger). This breadth of product range and store size among grocery
20 21 22

This is a so-called analytical generalization as described by Punch (1998). Source: definition found on the site http://www.oed.com (accessed 01.09.2005). The term modern grocery is taken from the definition of Planet Retail: modern grocery distribution includes both grocery and non-food sales from modern grocery distribution formats. It excludes sales from independent specialist formats and wet markets. Source: http://www.planetretail.org (accessed 01.09.2005).

18

2. Research Framework and Design

retailers will facilitate the generalisation towards other retailers with similar product categories or distribution channels. Another advantage of this group is the high variance that exists within it: some grocery retailers still rely on completely manual systems, while others have implemented very sophisticated systems with complex forecasting algorithms (Smros, Angerer et al. 2004a). Grocery retailers have to fulfil consumers' wishes immediately, as supermarket visitors are not normally willing to wait for the delivery of their goods. This business has further a strong demand volatility as many products are easily substituted, making SKU-level forecasts difficult. Therefore, all the grocery retailers have adopted a make-to-stock strategy.23 Another benefit of studying grocery retailers is that they are rather well organized in initiatives such as ECR-Europe. There are several project groups analysing their supply chains and the benefit of collaboration, standardization, electronic data interchange, identification and more so that additional information sources are more easily available for researchers. And last but not least, the choice of grocery retailers follows the research tradition of the institutes ITEM and KLOG at the University of St.Gallen, where this subgroup of retailers has been examined for several years.

2.2. Research Methodology

The long tradition of the University of St.Gallen to focus on topics that are relevant to business practice is fully continued by this work. One of the main advocates of this statement is without doubt Hans Ulrich. For him, business science is understood as a leading and managing-science and has thus the central objective of giving practitioners the ability to act and to make decisions in a scientific way (Ulrich 1981). The starting point for each research project in business science is an analysis of existing practical problems. First of all, interesting situations, correlations and contexts are observed from a practical point of view and then are conceptualised (Ulrich 1981). The concepts that are developed are tested in practice again and again and become gradually more refined. This iterative learning process will finally generate at the end of the research process theoretical and practical solutions to the identified problems that can again be tested in practice (Kromrey 2002). Ulrich's design of a 7-step research process was the base for the structural approach of this thesis (see Figure 6). The main characteristics of this procedure are the iterative approach and the deep contact of the researcher with the practice.

23

For an explanation of make-to-stock see Alicke (2003, p. 50).

2. Research Framework and Design

19

Step 1: Identifying and structuring problems and their potential solutions that Step 1: Identifying and structuring problems and their potential solutions that are relevant to business reality are relevant to business reality Empirical social sciences

Practice

Step 2: Identifying and interpreting theory and hypotheses of the empirical Step 2: Identifying and interpreting theory and hypotheses of the empirical social sciences relevant for the targeted problem social sciences relevant for the targeted problem

Formal sciences

Step 3: Identifying and specifying formal scientific procedures that are relevant Step 3: Identifying and specifying formal scientific procedures that are relevant for the targeted problem for the targeted problem

Step 4: Identifying and assessing the relevant application context Step 4: Identifying and assessing the relevant application context

Practice

Step 5: Deriving rules and models Step 5: Deriving rules and models

Step 6: Testing rules and models in the application context Step 6: Testing rules and models in the application context

Practice

Step 7: Documenting research results and consulting of practitioners Step 7: Documenting research results and consulting of practitioners

Figure 6: Integrative research procedure

24

In the first part of the research process (steps 1 to 4), a mix of qualitative research methodologies is used. These research steps are documented in this thesis in chapter 2 and 3. Campbell and Fiske (1959) have stressed the importance of using several methodologies to overcome the main deficiency of qualitative research: the limited generalisation. Denzin (1978; 1989) in turn develops the term "multiple triangulation" that applies when researchers combine multiple observers, theoretical perspectives, sources of data and methodologies in one investigation. He further states that "all the advantages that derive from triangulating single forms are combined into a research perspective that surpasses any single-method approach" (Denzin 1978, p. 304). The term triangulation is a metaphor from navigation which "use[s] multiple reference points to locate an object's exact position" (Smith 1975, p. 273). The triangulation of data collection settings affects the external validity of the results (McGrath 1982). Several researchers from social and business sciences, such as Webb, Campbell et al. (1966), Smith (1975) and Jick (1979), recommend triangulation. The research process used in this thesis adopts this methodology by simultaneously combining different research methods and data collection forms.

24

Source: adapted and translated from Ulrich (2001, p. 222).

20

2. Research Framework and Design

Mc Grath (1982) defines eight research strategies used in management research. The one strategy corresponding closest to the methodology used in this thesis is the field study. Field studies investigate behaviour in its natural setting. The data is collected by the researchers themselves on site. Scandura and Williams (2000, p. 1251) give the advantages and risks of this method: "This strategy maximizes realism of context, but it can be low on precision of measurement and control of behavioural variables (there is lack of experimental control). It can also be low on generalizability to the population, with the study population not representative of the target population." Despite these drawbacks, this research strategy is extremely popular. In a review of 385 papers of the journals Academy of Management Journal, Administrative Science Quarterly and Journal of Management from the years 199597, Scandura and Williams (2000) found out that 67.5% of all papers had chosen field studies as the research strategy. Ten years previously the percentage was only 54.1%, therefore a significant increase had taken place. An example of this strategy in this thesis is the KLOG OOS-project conducted for the grocery retailer MYFOOD. This was not an experiment, as no variable was manipulated, and the researchers acted merely as observers. The data collection and its analysis comprise qualitative and quantitative research strategies. Friedli, Billinger et al. (2005) state that the three most frequently used methods of qualitative research are action research, grounded theory and case study. The main methodology used in this thesis is the last. The basic idea behind case studies is to investigate only a small number of cases, sometimes even only one, yet these in a great detail. As Punch (1998, p. 150) states, "the general objective is to develop as full an understanding of the case as possible." According to Yin (1988, p. 23), a case is an empirical inquiry that: " Investigates a contemporary phenomenon within its real-life context; when the boundaries between phenomenon and context are not clearly evident; and in which multiple sources of evidence are used." As demonstrated in section 1.4, there are almost no existing academic sources that investigate the topic of this thesis. The focus on the case study method seems therefore to be most appropriate for this research project, as this methodology is recommended for researching topics in new areas (Eisenhardt 1989). The external validity of qualitative studies is sometimes considered to be limited. It is true that case study research is not able to give statistical generalisation, hence the external validity

2. Research Framework and Design

21

comes from analytical generalisation (Gassmann 1999). In case study research, however, generalisation is not always the goal. Some cases might be so important or interesting that they deserve a study in their own right (Punch 1998). To achieve generalisation, it is necessary to conduct the research at a sufficient level of abstraction: "The more abstract the concept, the more generalisable it is. Developing abstract concepts and propositions raises the analysis above simple descriptions, and in this way a case study can contribute potentially generalisable findings" (Punch 1998, p. 155). The main advantage of case studies is that they are most appropriated for describing and understanding complex social systems (Marshall and Rossmann 1995). Another argument for the case study approach is that the phenomenon observed in this thesis happens in the presence of but cannot be influenced by the researcher (Yin 1988). The main reason for choosing the case study method is that case studies have a holistic focus and aim to preserve and understand the wholeness and unity of a case (Punch 1998). More than 50 interviews with practitioners were conducted, with the aim of reaching an in-depth understanding of retailers' logistics (see the interviews listed in section 1.1). These interviews are conducted orally, as the actuality and explorative character of this research phase would not support a written interview technique (Lamnek 1993). The intimate connection with empirical reality that is demanded by Glaser and Strauss (1967) in order to develop a sound, relevant and testable theory is guaranteed with this approach. There are, nevertheless, also elements from another two popular qualitative methods. The main idea from grounded theory is that qualitative research has the goal of generating new theories (Punch 1998; Friedli, Billinger et al. 2005). And from action research the idea is adopted that the starting point for a research project is not a theory but a problem in practice (Coughlan and Coghlan 2002). The aim of the action researcher is to attempt to change the examined system and obtain a desired new status (Ulrich 1981; Coughlan and Coghlan 2002). The author was not personally involved in this change; nevertheless the last chapter of the thesis gives recommendations to managers as regards technical and organizational changes that a retailer may undergo to reach a new level of automation. In order to derive and test rules and models (steps 5 and 6), four case studies of retailers using advanced ASR systems are described (chapters 4 and 5 in this thesis). This is the part of the research project with the greatest interaction with

22

2. Research Framework and Design

practice; several iterative loops between theory generation and theory verification are carried out. On the one hand, the information gained from the interaction with the chosen companies is used to verify and refine the descriptive model developed in sections 4.1 and 4.2. On the other hand, the information gathered is the main source for developing the explanatory model. Closure is achieved when the differences between collected data and developed theory is small (Bansal and Roth 2000). A good illustration of this iterative process can be seen in Figure 7.

Literature Literature review review

Differentiation, Differentiation, abstraction, change abstraction, change of perspective of perspective

Critical reflection Critical reflection of gained picture of gained picture of reality of reality

Problems of Problems of practice practice

Own Own constructs constructs

Theoretical Theoretical comprehension comprehension

Iterative learning process

Data collection Data collection

Phenomenon Phenomenon of practice of practice

Questions to Questions to reality reality

THEORY

EMPIRICISM

Figure 7: Case study research as iterative process between theory and empiricism

25

For the creation and verification of the models, the author does not rely only on qualitative data. As stated by Yin (1988) and Eisenhardt (1989), the data collection method and the case study process can also include more qualitative elements. While experiments have great internal validity because of their precise control of variables, an additional external validity can be achieved by accompanying surveys (Scandura and Williams 2000). Therefore, next to several in-depth interviews (qualitative data), archival documents review, participant observation and analysis of the data gained from the ERP system of one retailer is included. This process allows the researcher to gain an even broader understanding of the company, or as Jick (1979, p. 603) states: "a more complete, holistic and contextual portrayal of the unit(s) under study." In addition, the quantitative analysis is a critical part of the testing of the models and hypotheses created in section 4.3. Some researchers criticise the simultaneous use of quantitative and qualitative methods, as every method has different aims and purposes (Dey 1993). The traditional distinction between two schools of social sciences, one oriented towards

25

Source: translated from Gassmann (1999).

2. Research Framework and Design

23

the qualitative development of theories, the other directed at the quantitative testing of theories, is criticised by Baumard and Ibert (2001). The authors point out the importance of not being dogmatic, as both forms of data are useful for constructing and testing theories. Or, as Glaser and Strauss (1967, p. 17) formulate: "There is no fundamental clash between the purpose and capacities of qualitative and quantitative methods or data. (...) Each form of data is useful for both verification and generation of theory." The final step in this research project (step 7) is the developing of recommendations for managers (chapter 6 of this thesis) and the documentation of the findings in form of the presenting thesis.

2.3. Research Process

The research process was guided by several activities which can be seen in Figure 8.
Summer 2005 (7) Synthesis of the results: writing of thesis (7) Synthesis of the results: writing of thesis (6) Field research: qualitative and quantitative analysis (6) Field research: qualitative and quantitative analysis

(2) Activities (3) Supervision (4) European (2) Activities (3) Supervision (4) European of Bachelor with grocery of Bachelor with grocery and Master ECR-Europe research and Master ECR-Europe research theses study theses study

(5) Consulting (5) Consulting projects projects

October 2002

(1) Literature review (1) Literature review

Figure 8: Research activities in this research project

The investigation began, as expected, with a thorough review of the available literature and other information sources (Activity 1). Due to the lack of academic papers dealing with ASR systems, the research started with more general inventory theory and logistics literature. Later, the research concentrated on papers targeting typical retail problems, such as data accuracy, inventory visibility and other logistics

24

2. Research Framework and Design

challenges. A new perspective came from the study of papers dealing with business IT and contingency theory. The involvement in the ECR Europe community was critical to obtain first hand information from practitioners (Activity 2). In several dialogues and meetings, the author deepened his understanding of business practices and current challenges in the retail and consumer goods industry. The involvement of the author in the ECR community was: Personal involvement in the ECR-Switzerland working group "On-shelf availability" Participation in the quarterly ECR-Switzerland working meetings and the official ECR Europe Conferences in Berlin (2003), Brussels (2004) and Paris (2005). Participation in various activities of the ECR Europe Academic Partnership, including the publishing of the ECR Journal and the organization of the annual ECR Student Award. The supervision of bachelor and master theses was an additional important source of first hand information from practitioners in the industry (Activity 3). All supervised works focused on actual problems of enterprises in the consumer goods sector and gave a broad overview of the performance of logistics within retailers: CPFR in the sports industry. Simon Steiner (2003) CPFR in the fashion and apparel sector. Marion Brgger (2003) CPFR in the German grocery retail market. Juerg Neuenschwander (2004) CPFR in the consumer electronics industry. Thomas Khl (2004) Chances and limits of CPFR in the Swiss grocery retail. Diego Rtsch (2004) Limits and possibilities of CPFR in the Swiss grocery retail. Dominic Loher (2004) Supply Chain Management in the fashion industry: critical factors. Benjamin Brechbhler (2004) The operational control of stores with inventory management systems. Remo Maggi (2004) Efficient Consumer Response. Impacts on consumer and welfare. Gabriella Todt (2005) The supervision of this work resulted in over 45 documented interviews with directors and managers in charge of logistics and supply chain management, category management, or information technology. The chosen companies were mainly retailers, followed by logistics and technology services providers. This more

2. Research Framework and Design

25

explorative approach led to a deep understanding of the current logistics processes within retailers. The result of this inductive qualitative approach was the creation of the descriptive and parts of the explanatory model (see chapter 4.3). The next information source for the creation of the models and hypotheses was the pan-European study "Logistics processes of European grocery retailers" (Activity 4). It was launched in autumn 2003 and lasted for about a year. The aim of the study was to investigate the current logistics processes and performance of leading European grocery retailers. The study was conducted as a collaborative project involving researchers representing five different European universities. The consulting projects conducted together with industrial partners were another important source of information (Activity 5). The most valuable project for the research into on-shelf availability was conducted with the grocery retailer MYFOOD. The data gained through this project is the main data source for the quantitative analysis (see chapter 5). In addition, close contact with employees from MYFOOD gave helpful insights into retailers' replenishment operations. The last activity before finishing the thesis (Activity 7) was the creation of field research of European retailers that have ASR systems in use (Activity 6). First, in short interviews (about 15 minutes), some 30 German, Swiss and Austrian retailers from very different consumer goods sectors were asked about their replenishment practices. The information gained in this survey was used to choose the four companies for the case study and also to design a semi-structured interview for the analysis. The selected companies and the outcome can be seen in chapter 6.

26

3. Literature Research

3. Literature Research
There exists almost no academic source that deals directly with automatic store replenishment systems at retail store level. Yet, even if theory cannot provide direct answers to the research questions, it is fruitful to consider the presented practical problem from different theoretical perspectives in order to obtain new impulses for this research project. Therefore, four theoretical research sources are addressed below in detail to verify, what contribution further theoretical streams can offer. The literature research focuses on sources dealing with inventory management, logistics and operations management, business IT and contingency theory (see Table 2).

Perspective Inventory Management

Contribution to the Thesis/ Research Question

Definition of decisions an ASR system has to take (descriptive model) Existing replenishment logics and strategies for replenishment systems (classification) Performance of different replenishment logics OOS research: reasons and interrelations Forecasting methods and typologies Benefits of well specified and organized logistics systems and programmes Best practice in store operations Coordination of structures, processes and systems to increase the efficiency of the SC Role of Enterprise Resource Planning (ERP) systems Organizational and human agency changes due to new technology

Q1 Q3

Logistics and Operations Management Business IT Theories Contingency Theory

Q2 Q5 Q4 Q5 Q4 Q5

Importance of the choice of the right ASR system for a given environment Organizational perspective: influence of the contextual factors within an organization on performance of ASR systems

Table 2: Overview of basic theoretical sources reviewed (excerpt)

3.1. Inventory Management Perspective

The first perspective that is used to examine the research problems focuses on the challenge of right inventory management. Existing academic sources in inventory modelling seek to answer the two primary questions that arise when dealing with a replenishment system, namely when should which quantity be ordered (Wagner 2002). Many sources in the operations research (OR) field concentrate on the modelling of replenishment systems, and try to identify an optimum under certain restrictions (see Groote 1994; Silver, Pyke et al. 1998; Bassok 1999; Gudehus 2001). They seek to obtain a minimal inventory level by choosing the right order quantity and order point so that certain a priori set objectives are fulfilled (e.g. a certain percentage

3. Literature Research

27

of service level, a maximum rate of out-of-stock, etc.). This approach has been widely used in practice for many years (cf. Chang 1967). More sophisticated models seek to optimize a specific utility function. The challenge here is to identify and quantify all the relevant logistics costs that should be included in this function. Inventory costs include factors such as the cost of carrying stock, order costs, safety stock costs, transport costs and an estimate of the out-of-stocks costs. For example, Galliher, Morse et al. (1959) and Dalrymple (1964) explicitly consider OOS-costs in their inventory control systems.26 The financial importance of optimal inventory control for companies is stressed by many authors such as Vollmann, Berry et al. (1992) and Dubelaar, Chow et al. (2001). Some sources even investigate the importance of the inventory for entire economies. In the United States, for example, the inventory value as a percentage of the GDP was in 1993 as high as 17.7% (Silver, Pyke et al. 1998). In the following sections, first theoretical inventory management sources dealing with mathematical models are highlighted. Papers and studies that in particular examine OOS situations in retail form the second part of the inventory management perspective.

3.1.1. Optimization in Inventory Management Research Inventory control systems have the aim of balancing demand and supply, reducing overall inventory costs and assuring an adequate service level (Wegener 2002). There are countless papers dealing with the modelling of sophisticated mathematical solutions for such inventory management systems (see e.g. Chang 1967; Inderfurth and Minner 1998; Ketzenberg, Metters et al. 2000).27 Wagner (2002) criticises the rather theoretical approach of these papers; their implementation in real life applications is problematic. Several other authors concentrate on retailers and their specific inventory management challenges. An identified central strategic issue that influences the success of a retailer is the setting of the right service level (Balachander and Farquhar 1994; Gudehus 2001; Zinn, Mentzer et al. 2002). Other papers deal directly with the optimal ordering policy for retailers from a marketing point of view, i.e. they look for the optimal assortment and shelf combination to maximize sales. In the 1960's and 70's, several experiments were conducted to measure the effect of shelf facings and inventory quantity on sales (e.g. Kotzan and Evanson 1969;

26 27

A detailed discussion of this cost function can be found in section 4.1.2. For an overview of the history of OR research on inventory control, see Wagner (2002).

28

3. Literature Research

Krueckenberg 1969; Cox 1970; Curhan 1972). In the following decades, other models sought to calculate an optimum in order to minimize inventory, shelf space costs and backorder (c.f. Corstjens and Doyle 1981; Cachon 2001). In these models, the handling costs are normally ignored. An exception is the paper by Broekmeulen, van Donselaar et al. (2004a). These last researchers state that the inventory carrying costs are low compared to the handling costs, therefore they suggest a replenishment logic that takes shelf space and package restrictions more strongly into account.

3.1.2. Theoretical Sources on OOS A group of inventory management papers reviewed deals directly with OOSs. These papers can be split into two sub-groups. The first group empirically studies the extent and causes of OOSs. The first OOS study, conducted nearly 40 years ago, reports an average OOS rate of 12.2% (Progressive Grocer 1968). More recent studies report an OOS rate between 7% and 10% (Andersen Consulting 1996; Gruen, Corsten et al. 2002; Roland Berger 2003b; Stlzle and Placzek 2004). The second sub-group takes a marketing and behavioural perspective and studies the reaction of consumers towards out-of-stock that crucially influence retailers' sales (e.g. Emmelhainz, Stock et al. 1991; Drze, Hoch et al. 1994; Zinn and Liu 2001; Sloot, Verhoef et al. 2002). Retailers try to increase their sales with two groups of market-driven tactics. On the one hand, there are "out-of-stores" tactics, which try to bring more consumers into the stores. Avoiding OOSs is, by contrast, an "in-store tactic." The latter tactics generally attempt to extract more surplus from shoppers once they are in the store. An attractive, full shelf attracts the attention of the consumer, making a purchase more probable. As shelf space is expensive28, retailers have to decide whether to place another facing of a certain product (to increase its visibility and/or reduce the OOS probability) or to place an additional SKU (Drze, Hoch et al. 1994). Several studies prove the value of such store specific micromerchandising; consumer decision-making can be strongly influenced. Only one third of purchases are specifically planned in advance of a shopping trip (Dagnoli 1987). Many buying decisions are made on a low level of involvement and very quickly (Hoyer 1984). In addition, the average shopper shops in 34 supermarkets each week (Coca-Cola Retailing Research Council 1994). With these facts in mind it is easy to see the magnitude of the impact OOS still has today for grocery retailers'

28

In the USA store occupancy costs range between $20 per square foot for dry grocery and $70 per square foot for frozen goods (Drze, Hoch et al. 1994).

3. Literature Research

29

sales. The knowledge of consumer behaviour is necessary to calculate the losses connected with OOSs.

3.1.3. Contributions and Deficits of an Inventory Management Perspective The inventory management perspective has a significant contribution to this thesis: Basis of descriptive model Foundation of relationships in explanatory model OOS magnitude and impact The descriptive and explanatory models developed in chapter 4 are based on contributions from inventory management sources.29 In particular, the sources that deal with challenges of replenishment in retail are of major value for this thesis. The papers and studies dealing with OOS highlight the magnitude of this problem in practice, illustrate the underlying reasons for low product availability and demonstrate possible solutions. Nevertheless, there are also some limits to the potential contribution of the theoretical inventory approach to this research field: Simplified view of reality Context parameters are regarded as given Missing organizational and contextual aspects First, all mathematical inventory models assume a very simplified view of reality so that it is often very difficult to apply such systems a given real-life situation. Second, it is often assumed that the parameters in such systems are given. What is overlooked is that many parameters can be changed by the organization (e.g. delivery frequency, case pack size) so that the strategic dimension of such decisions are not examined. And third, these OR papers normally neglect the organizational and contextual aspects of replenishment systems, which are nonetheless critical for the performance of such systems. An exception to the last statement is the paper by Zomerdijk and Vries (2002), as the authors place their research focus on environmental influences of inventory control systems. The basic message of the authors is that, beside the traditional points of attention such as order quantities and replenishment strategies, it is critical to take care of contextual and organizational factors. The authors identify four significant aspects in the organizational context of inventory management: task allocation, decision-making and communication processes as well as the behaviour

29

For an overview of authors see Table 6.

30

3. Literature Research

of personnel. The notion of examining the contextual focus of replenishment systems is a fundamental concept that is incorporated into the research questions (see section 1.5). Overall, a purely mathematical approach to inventory management is not appropriate, as Wagner (2002) states. He sees a major problem in the fact that classical inventory research is blind to all the "dirty data" issues that challenge companies (i.e. the data in the system is not accurate enough to be used for control and calculations). As Wagner further states, inventory modelling research is far removed from the entrenched software that now drives supply chain systems. Therefore, it is necessary to implement a comprehensive discussion of ERP and inventory holding systems, as will be accomplished in section 3.3.

3.2. Logistics and Operations Management Perspective

Academic sources dealing with logistics and operations management form the second perspective addressed in this thesis. Logistics research can be defined as the systematic and objective search for and analysis of information relevant to the identification and solution of any problem in the field of logistics (Chow and Henriksson 1993). The basic assumption behind logistics research is that a particular course of action will be correlated with logistics performance (Chow, Heaver et al. 1994). The problem starts with the definition of performance, sometimes hard measures are meant (such as delivery time or net income), and sometimes more soft measures are in the focus (such as consumer happiness ratings or flexibility). Both perspectives have their strengths and weaknesses. A comprehensive overview of the literature on this topic is made by Chow, Heaver et al. (1994). Their main criticizing point is that none of these studies examines logistics performance in the context of supply chains. Yet this statement has to be revised today in the light of the comprehensive literature on this topic (cf. Stlzle, Heusler et al. 2001; Karrer, Placzek et al. 2004; Stlzle 2004). The logistics research cannot be completely separated from another research stream that is of great relevance for this thesis: operations management. Operations management is an area of business that is concerned with the production of goods and services, and involves the responsibility of ensuring that business operations are efficient and effective.30

30

Definition by Wikipedia (URL: http://www.wikipedia.org, accessed 1.09.2005).

3. Literature Research

31

In the following, three research streams are looked at: sources dealing with Supply Chain Management (SCM) and Efficient Consumer Response (ECR), with Automatic Replenishment Programmes (ARP) and sources examining retail operations.

3.2.1. Supply Chain Management and ECR Supply Chain Management describes an approach between interdependent firms to collaboratively manage material and information flows (Corsten 2002). One of the basic principles of SCM is the management of connected activities that involve several units in a supply chain. The term "units" can refer to departments within an organization, but SCM also deals with "the management of upstream and downstream relationships with suppliers and customers" (Christopher 1998) that are outside the company. One of the goals of SCM is to reduce the inefficiencies, e.g. the bullwhip effect (Lee, Padmanabhan et al. 1997a), observed in many supply chains. The term SCM is understood here in broader terms; the explicit distinction between logistics management and supply chain management that is sometimes found in the American literature is not made. Many SCM papers focus on the ideal design of supply chains (Lee and Billington 1992; Anupindi 1999; Gudehus 2001; Gudehus 2004; Pfohl 2004), how to control them (Lee and Billington 1993), how to cope with the uncertainty of demand (Fisher, Hammond et al. 1994; Feitzinger and Lee 1997) and the importance of sharing information between firms (Lee, Padmanabhan et al. 1997b; Cachon and Fisher 2000; Chen, Drezner et al. 2000). Hines, Holweg et al. (2000), for instance, have built a very colourful model of supply chains, comparing supply chain systems with waves breaking on a beach. The book of Womack and Jones (1996) brought new momentum to logistics and operations management by describing the authors' notion of a lean supply chain. One field of SCM deals with the creation of pull-based systems and is therefore closely related to the ASR topic. Several authors stress the need for push systems for obtaining efficient supply chains (Fiorito, May et al. 1995; Cottrill 1997). As Bell, Davies et al. (1997) state, one of the main opportunities and at the same time challenges for food retailers nowadays is the transformation of supply chains from a push to a pull strategy. A pull system is driven by demand at the lowest point of the chain (Christopher 1998). In the framework of this thesis, the lowest point is the shopper. The simulation of Closs, Roath et al. (1998) demonstrates the benefits of a response-based system. Their pull-replenishment model is able to reduce uncertainty by sharing information between supply chain partners. The created automatic

32

3. Literature Research

replenishment system reduces (at least in theory) system-wide inventory by 25%, delivers 23 percentage points better service level and gives more robust results across a wide range of environments than traditional retail order-based systems. An additional finding of major interest is that most of the inventory reduction happens at retailer level. As a study of 12 major European groceries shows (Smros, Angerer et al. 2004a), this is the part of the supply chain with the most need for improvement in the stock situation. The new pull-replenishment system described by Closs, Roath et al. (1998) creates benefits for all partners in the supply chain. A prerequisite for this system, apart from investment in information technology, is the integrated interorganizational management of relationships. This means a change in the mindset of the buyers and sellers, as the risks and the rewards have to be shared in a cooperative approach. The first hurdle in such a cooperative attitude is the initial cost associated with the IT introduction. A drawback of Closs, Roath et al.'s (1998) approach is that their model stops at the retailer DCs, and does not take into account retailers' stores. Yet, the aim of a responsive supply chain cannot be solved by just the implementation of such technical systems. How retailer's processes and structures have to change to adapt such a demand-based supply chain is, for example, partly described in the case study by Fisher, Hammond et al. (1994) on the apparel retailer Obermeyer. Again the contextual situation plays a major role when designing the supply chain and replenishment systems and processes. Lee (2001) stresses that it is impossible for the human mind to optimize all the operations necessary for such demand-based systems. Lee therefore promotes the utilization of computer-based tools in retail. He has created a list with characteristics and requirements necessary for such systems to perform, such as the capability of doing precise demand forecasts based on marketing data and the capability of self-learning to adapt to new environments. Finally, there are some very useful articles by practitioners highlighting the importance of processes and organization for the performance of the entire supply chain, and how changes in these two factors can influence the performance of the entire company (Duffy 2004). Processes are in many of today's companies supported by Enterprise Resource Systems (ERPs).31 An interesting study that examines the influence of ERP systems on SCM is the work of Akkermans, Bogerd et al. (1999). The authors argue that once an ERP system is installed, this system can act as the backbone supporting business developments in many areas, among them SCM. However, the original ERP systems were never designed to specifically support
31

See also section 3.3 for more details on ERP systems.

3. Literature Research

33

SCM, and companies may find that such a system becomes a disadvantage, as Akkermans, Bogerd et al. (1999, p. 20) state: "Their architectural advantage of being fully integrated for one firm becomes a strategic disadvantage in this new business environment, where modular, open and flexible IT solutions are required." Since this Delphi study was conducted, several of the ERP software vendors have adapted to the trend towards inter-company SCM and have enhanced their ERP systems.32 Time will show how successful these new systems are in supporting SCM processes across multiple enterprises. Efficient Consumer Response (ECR) started originally as a supply chain initiative for the fast-moving consumer goods industry (Corsten 2002). Its main purpose is the collaboration between suppliers and retailers (Kurt Salmon Associates 1993). Since then ECR has developed and now includes marketing elements such as category management alongside SC elements (ECR Europe 2000). There are several publications of the ECR community that are cited in this thesis, dealing mainly with store operations and OOS. Yet the core element of ECR, collaboration, is not within the scope of this work.

3.2.2. Automatic Replenishment Programmes Automatic Replenishment Programmes (ARP) are a practical implementation of the idea of pull-supply chains. In these programmes, the vendor is responsible for the replenishing of the retailer's stock. To be able to do this, the vendor receives data such as inventory level information and POS data from the retailer. Replenishment is then pulled by the actual sales; thus no orders have to be actively placed by the retailer. The first successful implementation of such a system was between Wal-Mart and Procter & Gamble in the late 1980's (Kuk 2004). Common ARPs are Vendor Managed Inventory (VMI), Continuous Replenishment Planning (CRP), Quick Response (QR) and Collaborative Planning, Forecasting and Replenishment (CPFR). According to Davis (1993), the benefits of ARP concepts such as VMI are generally accepted, namely the possibility to reduce OOS rates and inventory levels. An interesting ARP paper related closely to the ASR topic is the contribution of Achabal, McIntyreet al. (2000). In this paper a decision support system (DSS) is described that was used by an apparel manufacturer to forecast the demand of
32

For example, the company SAP has introduced a software package called "mySAP SCM."

34

3. Literature Research

selected VMI retailers. It is worth noting that the system described uses a forecasting model to optimize the supplier retailer deliveries to the retail DC. Yet, the system does not take into account the replenishment at the store level. Nevertheless, the contribution of this paper is that for the first time a DSS in use is described that combines inventory replenishment decisions with advanced causal forecasting models derived from marketing theory. Their multiplicative forecasting model takes into account effects from price elasticity, promotional influences and seasonal effects and bases on pooled, weekly product data. The DSS system was tested for 30 retailers in a three-year period. The result was a dramatic OOS rate reduction of 70%. In addition, the service level at the DC increased from about 73% to 92%. A minor drawback was a slight fall in the inventory turnover. Further benefits for the vendor were better production planning, as retailers' orders were more stable and reliable with the new system. This is mostly attributed to the fact that an inflation of orders, resulting from "shortage gaming" (cf. Lee, Padmanabhan et al. 1997b), is avoided. In addition, the retailers had the usual benefits of VMI systems: they had lower costs and expended less effort on order placement and they profited from a cost-effective and more precise sales forecasting tool. The increase in forecasting accuracy is attributed to the utilization of several vendors' data and to the fact that more time is spent on the customization of the models. The main achievement of Achabal, McIntyre et al. (2000) is to show in practice that such DSS systems can dramatically improve inventory efficiency and are profitable for vendors and retailers. Yet, their paper concentrates mainly on the development of an efficient forecasting model and less on the automatic replenishment or on the contextual prerequisites of such systems. A deficit of their contribution is the missing description of new processes necessary for the effectiveness of automatic replenishment systems. A major empirical contribution to the investigation of ARPs comes from an American research group that based their findings on a survey of 75 American manufacturers and 23 retailers (Ellinger, Taylor et al. 1999; Stank, Daugherty et al. 1999). The authors see the main problem in modern retailing as the increased number of SKUs that have to be handled. The retailer's aim is to cope with this complexity without increasing the number of OOSs or inventory levels. In their first papers (Daugherty, Myers et al. 1999; Ellinger, Taylor et al. 1999), the research group studies the appearance and effects of such systems. First, the implementation of ARP-related elements in companies is measured. These are elements that are considered by the authors necessary for the implementation of ARP. As can be seen in Table 3, it is not only technological capabilities that are mentioned but also operations and organizational elements are to a certain degree a prerequisite for ARP. This table has

3. Literature Research

35

a major relevance for this work, as in section 6.5.3 the technological and organizational requirements for ASR systems are examined.
Item
Bar coding at retail level Electronic data interchange Automatic replenishment of basic goods Automatic forecasting for staple goods Pre-season planning with trading partners Cross-functional teams Automatic forecasting for fashion/seasonal goods Bar coding on container labels Direct store delivery of products Joint planning or replenishment of products Electronic payment Electronic point-of-sale scanners Advanced ship notices Cross-docking operations Joint forecasting Vendor-marked merchandise Activity-based costing

Mean*
5.66 5.46 5.45 5.33 5.26 5.26 5.22 5.20 4.91 4.79 4.74 4.66 4.53 4.46 4.21 3.93 3.77

Standard Deviation
2.31 1.77 1.76 1.90 1.93 1.65 2.45 1.86 2.42 1.84 1.91 2.74 2.18 2.17 1.89 2.56 2.19

*7-point scale: 1=not implemented 4=somehow implemented 7=fully implemented

Table 3: Implementation of ARP-related items

33

Ellinger, Taylor et al. (1999) come to the conclusion that for a basic involvement in ARP programmes, the most important elements are bar coding at retail level and electronic data interchange. With these elements in place, a basic level of automation is possible. For more sophisticated replenishment programmes, advanced elements such as joint planning and electronic payments, are necessary. The second question Ellinger, Taylor et al. (1999) looked at was the effectiveness of such programmes. Effectiveness was judged with the help of several scales as can be seen in Table 4. The authors showed that the group having a high ARP implementation level (more than 20% of total dollar sales involved automatic replenishment) has significantly faster inventory turns and at the same time reduced product handling, inventory holdings and costs. A reduction of OOSs could not be proved. However, higher profitability and overall relationship performance for the high ARP involvement group could be shown. These are important findings, as in this thesis the effect of ASR systems will be measured on three scales: inventory level reduction, inventory level variance and OOS.

33

Source: Ellinger, Taylor et al. (1999, p. 29).

36

3. Literature Research

Goal
Improved/increased customer service Fewer stock-outs Improved reliability of deliveries Reduction of discounting Reduced return and refusals Faster inventory turns Reduced over-stocks Reduced product damage Reduced inventory holdings Reduced handling Reduced costs

High ARP involvement

5.61 5.42 5.42 5.35 5.12 5.35 4.95 4.81 5.28 4.93 5.08

Low ARP involvement

5.40 5.33 5.19 4.86 5.03 4.44 4.92 5.06 4.17 4.23 4.06

*7-point scale: 1=not at all effective 7=extremely affective. Bold: significant difference between groups (p<.05)

Table 4: Effectiveness in achieving automatic replenishment-related goals

34

In a third paper Myers, Daugherty et al. (2000) test a series of hypotheses related to the antecedents and outcomes of automatic replenishment programmes. In their model, the authors postulate that the effectiveness (cost and service quality) of ARP systems will depend on three factors: Transaction-specific factors (factors influencing transaction costs, such as the standardization level of the ARP) Strategic factors (the market and profit orientation of the firm) Organizational factors (characteristic factors of the firm, such as the firm size) This model could only be partly supported by the data. Market-oriented companies had indeed as expected a greater service effectiveness at the cost of lower cost effectiveness. Service effectiveness was positively influenced by the firm size, but astonishingly, larger firms had lower cost effectiveness. A similar finding was reported by Kuk (2004). In his analysis of 25 electronics industry companies, the members of small rather than large organizations expected and perceived higher returns from ARPs. Another surprise was that decentralized companies did not have lower service effectiveness as expected. Not surprising was the fact that with increasing managerial commitment also the cost and service effectiveness rises. Further, Myers, Daugherty et al. (2000) expected a positive correlation between companies markets' competitiveness and their service effectiveness. According to the
34

Source: adapted from Ellinger, Taylor et al. (1999, p. 33). The results depicted in this table vary slightly from paper to paper of the American research group, though they come from the same empirical base. The authors do not explain the reasons for these differences.

3. Literature Research

37

data, the opposite was true, in highly competitive markets the ARP-related service effectiveness went down. This effect is explained by the fact that the fast-changing demand for goods has made accurate replenishment very difficult for ARPs. The authors also looked at the implications of ARP effectiveness for overall firm performance. They could prove a significant positive relation between the effectiveness of ARP and strategic performance (creation of barriers to competitors). The results of this paper deliver important insights for this thesis. Myers' and Daugherty's paper pinpoints that many of the factors influencing ARP success lie in organizational structure and human agency. Firm size, centralization of replenishment decisions and commitment to ASR seem to play an important role. But also the environment has to be taken into account, as the degree of market competitiveness plays a role. The last work analysed that is based on this American survey is by Sabath, Autry et al. (2001). In this paper, the authors first of all analyse the information systems capabilities of the companies in the sample using nine items. As can seen in Table 5, most IT systems are able to provide basic information, yet their capability for internal and external connectivity as well as speed and exception management are perceived to be rather low.
Overall Mean (n=97)
5.70 5.33 5.16 5.07 5.01 4.88 4.65 4.59 4.58

Information Systems Capabilities

Daily download of information Accuracy of information Timeliness of information Formatted to facilitate usage Availability of information Internal connectivity/compatibility Formatted on exception basis Real-time information External connectivity/compatibility

Centralized Group
5.49 5.19 4.97 5.11 4.58 5.31 4.41 4.19 4.29

Decentralized Group
5.88 5.45 5.33 5.03 5.39 4.49 4.87 4.95 4.85

*7-point scale: 1=not at all effective; 7=extremely affective; Bold: significant difference between groups at .01

Table 5: Information systems capabilities

35

Another question posed by the authors is whether there is a significant difference between the IT capabilities of decentralized companies compared to centralized ones. This is not a trivial question, as the theories on this topic are ambiguous. For example, the coordination theory states that to decide whether new communication technology will lead to more or less centralization, one has to look at which cost types are reduced (Gurbaxani and Whang 1991). If the cost to collect the information
35

Source: adapted from Sabath, Autry et al. (2001, p. 100).

38

3. Literature Research

required to take a decision falls, then the result will be greater centralization of the organization. On the other hand, if the agency costs sink (i.e. the costs of the controlling of the single independent units, in the case of this thesis, the stores), the utilization of new coordination technology will lead to greater decentralization.36 According to the survey of Sabath, Autry et al. (2001), the decentralized group tends to have higher IT capabilities. This could be a sign that the agency costs are more effectively reduced by the implementation of such IT technologies. An exception here is the internal compatibility that is stronger in the centrally-organized group. This phenomenon is explained by the authors on the basis of coordination theory. In centralized organizations it is necessary to have the information in the upper levels of the organization, therefore internal connectivity and compatibility is most important. Further, Sabath, Autry et al. (2001) examine the influence of the degree of centralization on the performance of ASR systems. Their basic result is that, in all examined areas, a decentralized organization will tend to achieve better results. Decentralized firms with ASR systems in use tend to perform better in relation to OOSs, reduction of overstocks, returns, handling costs and product damages. This strong argument for a decentralized system is contrary to the trend of power centralization which has been observed within European grocery firms (Smros, Angerer et al. 2004a). This American research group explains the sometimes ambiguous results of their studies by the relatively small number of companies in the sample and by the relative short period these automatic programmes have been in place. In addition, three more problematic issues can be seen. First of all, the way the effectiveness was measured could be misleading. The benefits of the systems are measured by a self-estimation of the interviewees on a 7-step Likert scale.37 Consequently, it is possible that companies that introduce sophisticated replenishment systems will also have a clearer view of the costs and performance of their supply chains. Problems that were hidden before are now visible; the benefits from ASR systems are therefore not adequately estimated. Second, it could be problematic that the researchers do not differentiate between the effects of ARP for retailers and manufacturers. It is a common statement from practitioners and researchers that, for example, VMI

36 37

For an overview on agency costs and theory, see Kieser (2002, p. 209). In the experience of the author, with this approach there is the risk of obtaining biased results. In the European survey by Smros, Angerer et al. (2004a) the authors asked a group of retailers about their happiness with their OOS rate. It is remarkable that the retailers who did not have implemented systematic methods to measure their OOS rate tended to be more satisfied with their on-shelf availability than the other group.

3. Literature Research

39

programmes create different costs and benefits for retailers and manufacturers.38 And third, although the replenishment programmes are called "automatic," this term is not used in the significance of this thesis (i.e. "not manual; executed by a machine"). The term "automatic" means in this context that the replenishment is managed by the vendor and not by the retailer. But whether the replenishment itself really happens automatically, i.e. the vendors' IT system takes the replenishment decisions, is not differentiated in these papers. Thus the contribution of these papers on the implications of having an IT system take over the replenishment decisions is limited.

3.2.3. Operations Management in Retail The sources studied from the operations management field have a focus on operations in the retail environment. In the following, some challenges found in the literature concerning store operations and replenishment systems are highlighted. One weighty issue when dealing with automatic replenishment systems is the data accuracy. As noted by both practitioners and researchers, the performance of automatic replenishment systems dramatically depends on the accuracy of the data. There are very few papers that investigate this subject (Raman, DeHoratius et al. 2001b; Wagner 2002; Fleisch and Tellkamp 2004; Tellkamp, Angerer et al. 2004), although it is highly relevant for retailers. In a case study with a US retailer, Raman (2000) found that inventory was inaccurate for over 70% of stock-keeping units (SKUs) in the store. These figures are based on physical inventory counts at six stores. Each store had on average 9,000 SKUs. Physical inventory was below book inventory for 42% of SKUs and above for 29% of SKUs. The total difference was 61,000 units or a mean of 6.8 units per SKU. This compares to an average inventory of 150,000 units per store. One of the major challenges of accurate inventories is shrinkage. A German study examining 104 German retailers estimates the inventory difference at 4.1 billion euro in the year 2004 (Horst 2005), and half of these losses are due to theft. The inventory differences add up to 1.1% of the gross product value. According to another US survey, shrinkage in the retail industry amounted to 1.7% of sales in 2002 (Hollinger and Davis 2002). The importance of this topic for practice is seen in the high sums invested in anti-theft measures, for instance, Germany's companies spent 900 million euro (Horst 2005).

38

See for a list of benefits and disadvantages, e.g. Alicke (2003, p. 171).

40

3. Literature Research

The replenishment of the shelves is one of the most time-consuming store operations found. The German grocery retailer Globus has calculated that 47% of its internal logistics costs are caused by in-store shelf replenishment. In comparison, the costs for the transport of the goods to the store (and back in the case of returns) represents only 14% of total costs (Shalla 2005). The replenishment effort is influenced by the size of the shelf, the available space in the backroom, the case pack size (CU/TU) and the store replenishment frequency. Retailers, such as Globus and Manor (Switzerland), have therefore started to design stores without a backroom in hopes of reducing this costly and error-prone process. Most of the European grocery retailers examined are centrally organized. Nevertheless, the role of the store manager in the performance of the store is still important. This person is responsible for several employees, has many freedoms regarding the processing of store operations and has decision rights over operating expenses such as labour costs. One of the ways discussed in the literature to oversee and motivate store managers is through appropriate incentives (see e.g. DeHoratius and Raman 2005). The role of store managers during and after introduction of an ASR system was found during the research for this thesis to be a critical element. The store manager's influence on the success of the replenishment system change was stressed by all interviewees. Overall, the review of the operations literature was very fruitful. The literature helped to identify the relevant fields for the field research in this paper. This was possible as, contrary to the literature about mathematical inventory management, the operations management literature examines topics very close and relevant to retailers' daily challenges.

3.2.4. Contributions and Deficits of a Logistics and Operations Management Perspective The logistics and operations management research explores a very wide range of topics. This is due to the integrating function that logistics has in today's enterprises (Pfohl 2004). As the implementation of ASR impacts so many areas of a retailer's business, logistics' holistic approach is a valuable knowledge source. In concrete terms, the contributions of these theories are: Expected benefits of ASR system Technologies and processes necessary for ASR implementation Conception of store operations

3. Literature Research

41

First of all, this thesis benefits from the ARP literature and their methodologies for measuring the performance of replenishment systems (research sub-question Q2). Second, the hypotheses found in theoretical sources regarding the correlation between IT capabilities, context of the organization and the outcome of automatic replenishment system implementation are extremely helpful in creating the models in the next chapter and in determining the IT requirements for each ASR level (see Table 49). Finally, the examination of store operations literature brought a major contribution for this thesis. As the field research in chapter 6 will show, the implementation of an ASR system fundamentally changes some of the processes in the stores as well as the organization structure. The best practices identified in the literature are compared to the processes of the retailers in the sample. Therefore, this theoretical perspective will be most helpful for the answering of research subquestion Q5, recommendations on how to re-engineer a company to adapt to a new ASR level. Yet, it has to be said that the contribution of the research streams ECR and SCM for this thesis is limited. The focus of this research project is on the replenishment of goods to the point of sales of the retailers. For this research context, the distribution systems along the supply chain, from the suppliers up to the retailers, play a secondary role. Concepts that are mainly centred on collaboration and which aim to reach a global inventory optimization across the supply chain are not considered in this thesis. The only contribution of suppliers for the success of advanced ASR systems is to ensure high inventory visibility. This is realized by having an identification system (e.g. barcodes) at item level and by the use of EDI messages such as dispatch advice. Yet for future research, the role of the supplier might increase when automatic replenishment systems are implemented across several supply chain tiers.

3.3. Business Information Systems Perspective

A stream of the business information systems research focuses in particular on the impact of new information technologies on companies and the human agency. This research branch has a longstanding tradition. It began as a stream of social research on the influence of human behaviour and technology in general (e.g. Mumford 1934; Noble 1984), focusing later on the impact of information technology (e.g. Zuboff 1988; Kling 1996). In this last section of the theoretical foundation, the role of Enterprise Resource Planning (ERP) systems is examined. First of all, the characteristics of ERP systems are analysed. Second, an overview of literature

42

3. Literature Research

regarding successful implementation is given before finally, the influence of ERP packages on human agency is depicted. 3.3.1. Characteristics of ERP Systems Watson and Schneider (1999) see ERP as a generic term for an integrated enterprise computing system. In order to integrate the different functions of a company, a common database is necessary (Gronau 2004). Today's ERP systems have developed from systems used in production to determine the necessary material quantities, so-called material requirements planning systems. By now these systems are widespread in all industry sectors and have expanded their functionality to include also human resources and finance elements. A recent paper on the influence of ERP on companies is the work of Kallinikos (2004). For him, the widespread introduction of ERP packages in organizations is radically changing human agency at work. Kallinikos (2004, p. 9) states that a main characteristic of an ERP system is "the reconstruction of the very micro-ecology of organizational tasks to which any single transaction belongs." This means that in a company with advanced ERP systems isolated tasks no longer exist. Even the smallest transaction is recorded and thus must be accounted for. This has two effects. First, the effects of one's actions become visible to others and to the individual undertaking the activity. For instance, the connection of the SAP R/3 modules "Financial Accounting" and "Warehouse Management" helps in recognizing the financial effects of one choice on the rest of the organization (Bancroft, Seip et al. 1996; Soh, Kien et al. 2000). This interconnection is the reason why ERP systems have had a more profound effect on human agency in organizations than other expert systems (Kallinikos 2004). ERP systems are not only a structured rendering of operations; they create a manageable organizational reality. The second effect of ERP systems is that the order in which transactions are correctly executed is stipulated by the system and thus makes the way tasks are executed by employees homogeneous.

3.3.2. ERP Implementation and Selection Integrated ERP systems are regarded by some authors to be of strategic importance for the competitiveness of companies. Therefore, some years ago, Vossen (2000) complained that German retailers were lagging behind the competition. He estimated lost turnover due to antiquated systems at 3.5% and stressed the necessity for a change of the IT system. Researchers have examined factors influencing the

3. Literature Research

43

successful selection and implementation of ERP packages (Kumar, Maheshwari et al. 2001; Sarpola 2003) and the impact of organizational and cultural factors (Krumbholz, Galliers et al. 2000; Soh, Kien et al. 2000). More practical managerial guidelines (e.g. O'Leary 2000; Ptak and Schragenheim 2000; Merkel 2003) became more popular in dealing with ERP choice and usage, as the implementation of an ERP system can be extremely costly. The list of failed and abandoned ERP introductions is long (Markus and Cornelius 2000), making the necessity for such guidelines clear. A study from the year 1999 shows that, on average, ERP projects are 178% over budget, take 2.5 times as long as intended and deliver only 30% of the promised benefits (Densley 1999). This goes hand in hand with the existing complaints of many practitioners that IT expenditure has failed to show significant productivity gains. In the literature, this phenomenon is known under the term "productivity paradox" (Stratopoulos and Dehning 2000). It was first described in 1987 by Steven Roach, the chief economist at Morgan Stanley (Brynjolfsson and Hitt 1998). The paradox is that although the amount of computing power used in the service industry increased dramatically in the 1970's and 80's, the measured productivity remained flat. Other studies have partly contradicted these findings (Brynjolfsson and Hitt 1995; Dewan and Min 1997; Malone 1997). It is argued that ERP as well as other complex IT systems, fail to deliver benefits because the process of organizational change was not properly addressed (Hong and Kim 2002) and working practices were not changed (Brynjolfsson and Hitt 1998). Therefore, ERP implementation guidelines stress the need for a comprehensive re-engineering. Chircu and Kaufmann (2000) speak of two types of barriers that have to be overcome for a successful implementation: valuation barriers (how compatible is the technology to industry standards?) and conversion barriers (which supporting resources have to be considered to ensure the implementation succeeds?). The technical barriers are rather low, as much of the IT and identification technology has become ready available and comparatively inexpensive (Kuk 2004). The organizational barriers, on the contrary, are much more difficult to surmount. Soh, Kien et al. (2000, p. 47) speak of a "necessary disruptive organizational change." A change in the organization and its processes is also necessary in the case of a misfit between the ERP system and the existing company. A misfit occurs when the functionality offered by the ERP system and the requirements of the company differ. In case of a misfit, a spectrum of solutions is possible, requiring more or less change within the company (see Figure 9). An example for a greater required adaptation is when ASR systems presuppose an electronic inventory record. The working habits of store employees can be radically changed by this new process. Such electronic records need, for instance, an

44

3. Literature Research

accurate scanning of every product that is sold. The consequence is that a significant part of employees' time will be dedicated to inventory counts and data administration.
Greater organizational change

Adapt to the new Adapt to the new ERP functionality ERP functionality

Accept shortfall Accept shortfall in ERP in ERP functionality functionality

Workarounds to Workarounds to provide the provide the needed needed functionality functionality

Customisation to Customisation to achieve the achieve the required required functionality functionality

Greater customisation of ERP

Figure 9: Spectrum of misfit resolution strategies
39

3.3.3. ERP and Human Agency Kallinikos (2004) notes that many sources dealing with the influence of ERP on the organization have a strictly technical perspective. The behavioural assumptions underlying ERP implementation are seldom examined in detail. For example, not enough attention is paid to the construction of roles centred around new processes (cf. Markus, Axline et al. 2000; Sheer and Habermann 2000). A detailed investigation of the constitutive effects ERP packages have on work, human agency and organizational action has not yet been systematically researched. The influence of ERP on personnel that has been identified in theory is threefold: Change of working patterns Reduction of personal communication Retreat to own zone of duties, loss of interest in overall goals The direct influence of ERP implementation on human working patterns has already been stressed. Many ERP packages specify in detail every required work process and have therefore a significant influence on human agency. The system might constrain the way less structured human beings work by institutionalising patterns of action. For instance, some higher ASR systems suggest an exactly defined procedure for checking orders. For example, store employees may be required to check the order suggested by the IT system during a small, defined time window. In addition, the visibility of employees' actions increases. Increased visibility means that actions (e.g. the placing of orders) can be monitored from every PC

39

Source: adapted from Soh, Kien et al. (2000, p. 50).

3. Literature Research

45

connected to the ERP system. At the same time, because transactions are stored electronically, there is also the possibility to assess actions conducted in the past, a convenience often not available in manual systems. For example, in an OOS situation, it might be unclear in a manual system whether an order was placed in time or not, and every party can blame another for the fault. The second effect found in the literature describes the reduction of personal communication between employees (cf. Fleck 1994; Sawyer and Southwick 2002). In the case of advanced ASR systems, the contact between store employees and the central office takes place by digital means. Johansson (2002) has made several propositions on how the retailer-supplier relationship may change, when increasingly quantities of information are interchanged via EDI, Internet and market places. He examines the effects on three fields: relationship formation and development, distribution of power and influence and the responsiveness and flexibility in supplierretailer relationships. The author states that co-ordination of activities is easier through digital communication, at the cost of the richness of the communication and the building of knowledge. Johansson further argues that with the use of digital channels it is easier to handle continuous change in the supplier-retailer environment. An example for continuous change is the change of consumers' consumption quantity of a certain existing product. Yet for discontinuous changes, such as consumers demanding completely new products, personal communication channels would be the better choice. The findings of Johansson (2002) can only be partly transferred to the head office-store relationship, as this is a relationship within an organization. The distribution of power is another: there is no danger that the store will decide to change the supplier due to, for example, lack of trust. Finally, IT systems do not only reduce personal communication. In addition, typical human behaviour, such as improvisation, exploration and playing, is often limited. Kallinikos (2004) sees the danger of losing one of the ultimate goals of the company responding successfully to the demands of the market and customersin a maze of thousands of ERP transactions: "Concern with internal processes may obscure and finally replace external adaptation" (Kallinikos 2004, p. 18). The ordinary user, when overwhelmed by thousands of transactional possibilities, tends to retreat to his limited zone of duties (Zuboff 1988; March and Olsen 1989). The wider purpose of their work might be lost of sight. An aim of good ASR system design must therefore be to avoid the loss of employee interest and motivation. When the retailer MYFOOD introduced its ASR systems, one problem was that store employees lost any interest in the placing of orders, as it was, in their eyes, no longer their duty. There were fewer controls of the computer-generated ordering suggestions as wished by the central

46

3. Literature Research

office. The retailer had to make a significant effort to abolish the "if the goods are outof-stock, then it was the computer's fault" mentality. Therefore, by encouraging employees to make a minimum number of interventions, they were made to feel as if the orders placed were still "theirs". Employees should perceive the computer as a mere assisting technology.

3.3.4. Contributions and Deficits of a Business Information Systems Perspective The discussion of ERP-related theoretical sources offers several contributions for this thesis: ERP's major influence on business Pitfalls and success factors Change of human agency The literature has shown the major impact ERP systems have on business. Several authors stressed that to remain competitive, implementation of ERP systems has become mandatory. However, the implementation of such systems is a challenging task. ERP systems radically change the way employees work. Their effectiveness depends to a great extent on the design of the organization and processes. The right system can create an awareness of the effects of the actions taken by employees (Kallinikos 2004). As higher ASR systems are usually integrated into ERP system, the mentioned implication of the information business research can be transferred to the ASR context. This research stream therefore provides valuable insights on how to change the organization and organizational processes (sub-question Q5) and on which system to take, considering the available technology (Q4). The limits of the application of the ERP research to the subject of this thesis are that traditional ERP systems only target procedures of action and execution. Modules with real decision support logic are not common standard (Metaxiotis, Psarras et al. 2003). Advanced ASR systems go a step further than traditional ERP packages in that they implement knowledge aspects as well. Thus, additional considerations from the field of expert systems have to be taken into consideration. Decision support systems (DSS) are rule-based programs that often involve statistical or algorithmic analysis of data. Their decisions are made in real-time after weighing all the data and rules for a particular case (Davenport 2004). The implications of DSS systems for the business community cannot yet be fully anticipated. For Davenport, the impact of this new

3. Literature Research

47

technology on the employment market will be much greater than the effects of offshoring. From 2001 to 2004, 2.7 million US jobs were lost due to the increase of productivity, compared to only 3,000 lost through offshoring (Davenport 2004). Davenport recommends companies to incorporate DSS into their strategy to remain competitive in today's markets.

3.4. Contingency Theory Perspective

Fred E. Fiedler is considered to be the founder of contingency theory. He carried out research on the optimal leadership style. Fiedler (1967) examined the influence of two factorsthe leader's characteristics and the leading situationon the performance of employees. The result was that there is no optimal solution; the best leadership style depends on the situation. This was the beginning of contingency theory. In essence, contingency theory states that the effective level of an endogenous variable depends on the level of an environmental (exogenous) variable (Staehle 1991; Cadeaux 1994). In Fiedler's example, the contingency aspect is that the leadership type (endogenous variable) has to fit to certain characteristics of the situation (exogenous variable) in order to bring about the desired results. This theory can be applied at different levels of abstraction (e.g. organization, process level) as discussed in the following.

3.4.1. Contingency Theory at the Organizational Level One of the first contexts in which contingency theory was applied was the organization. Lawrence and Lorsch (1967) conducted empirical studies to formulate a contingency theory of the organization. They argued that differences between types of organizations are mainly due to the different environmental conditions they act in. In their study, the authors characterize the environment with the pairs secure-insecure and homogeneous-heterogeneous. The security dimension of an environment can be measured with the help of following three criteria: Certainty and reliability of information (environmental stability) Frequency of changes of the information (environmental change) Duration of the feedback cycle (between system and environment) Organizations reduce the complexity of their environments by creating sub-systems. For instance, the complex environmental factor "supplier" leads to the creation of the functional unit known as "purchasing department." Another example is provided by Staehle (1991). He examined the plastics industry and discovered that it has a heterogeneous environment: production is very secure, yet the basic research

48

3. Literature Research

department is classified as very insecure. Staehle then makes suggestions as to how to organize a company depending on his environmental classification. An application of contingency theory to logistics is made by Pfohl and Zllner (1987). In their theoretical work, they examine the influence of contingency factors on organization design. They put into question the effectiveness of the popular strategy of aggregating all logistical tasks into a major department. The authors look for alternative organization forms. Pfohl and Zllner argue that the choice of logistics organization should no be limited to just quantitative aspects (such as the size of the company). There are other important contingency factors that have to be considered as well: The complexity and uncertainty of the criteria Requirements of the decision-making levels Relationships with other functional units The behaviour of the employees The main conclusion of the authors is that the structures found in practice can only be partly explained by contingency theory. Decisive roles are played by the managers that set the overall strategy of the organization. Such managers choose structures that coincide with their cognitive and motivational orientation (Bobitt and Ford 1980). Different structures for logistics will emerge, depending on how the importance of logistics in an organization has been assessed (Pfohl and Zllner 1987). The authors cannot make a standard recommendation on how to implement best logistics in an organization, because this will depend on the requirements of the logistical task. For the type of companies in the focus of this thesis the consequence is clear. Logistics is for retailers a pivotal task (Krber 2003). Pfohl and Zllner (1987) recommend that in this case, companies should determine the organizational structure based strongly on logistical contingency factors. A last example for contingency theory at the organizational level stems from Kahn and Mentzer (1996). These authors suggest that the choice of the level of integration and interaction between departments should depend on the situational context, in this case the stability of the market.

3.4.2. Contingency Theory on Information Technology and Processes Contingency theory also plays a role when assessing the appropriateness of processes. Cadeaux (1994), for example, shows that the benefits from a branch-flexible planning of the store assortment strongly depends on the volatility of

3. Literature Research

49

the product category. Another example is the contingency between supply chain responsiveness and the need for forecasting abilities. It can be observed that suppliers have invested more in forecasting capabilities than retailers (Smros, Angerer et al. 2004b). An explanation given by these authors is that the need of retailers for forecasting is different, as their supply chain is more responsive (i.e. shorter lead times and higher service levels). Another branch of the contingency research stream examines the impact of information technology on business performance under the aspect of contingency theory. A good overview can be found in Bergeron, Raymond et al. (2001). In this paper, the authors list research executed on the study of specific impacts of IT, among them studies on the IT impact on learning (Leidner and Jarvenpaa 1995), the impact of IT problem-solving on task performance (Vessey 1991; Vessey and Galleta 1991), and the impact of IT on organizational performance (Chan and Huff 1993; Bergeron, Raymond et al. 1998). A common proposition among these papers is that a turbulent environment will induce firms to a more extensive use of information systems (Pfeffer and Leblebici 1971; Lederer and Mendelow 1990). Chagu (1996) argues that in risky environments, IT should be more flexible and managers more alert to adapt information systems to external changes. In addition, information acquisition should be more wide-ranging and continuous (Huber 1984). Generally speaking, there are many papers including the environmental fit as a contingency variable (Bergeron, Raymond et al. 2001) of IT performance. But very few show the link between the fit of such systems on environmental uncertainty and performance. An exception is, for example, the work of Sabherwal and Vijayasarathy (1994) which examines telecommunications technology fits and performance. Another application of contingency theory is the development of a contingency approach for the selection of logistics performance measures as conducted by Chow, Heaver et al. (1994). For example, the authors expect for a firm that has concentrated its strategy on cost dimensions, to focus its performance measures on cost dimensions as well. The transfer of these theories to the context of this research project means that retailers that have more complex replenishment systems will have at the same time a higher demand for elaborated performance measurements. If, for example, a complex forecasting algorithm is in use that requires more care on data accuracy, the company requires a more accurate image of the performance of this algorithm. Therefore key performance indicators (KPIs) such as MAPE40 would probably be introduced to test and control the algorithm.
40

Mean Absolute Percentage Error. For an explanation of MAPE see section 4.1.4.

50

3. Literature Research

3.4.3. Contributions and Deficits of a Contingency Perspective The basic idea behind contingency theory is of great relevance for the answering of research questions Q4 and Q5. The main contributions of the contingency theory for this thesis are: Choice of the right ASR system for each product category Fit of organization and processes to ASR system Consideration of the influence of store and product characteristics on ASR performance One of the main goals in this thesis is to help retailers to choose the right replenishment system. Yet it would be wrong to rely just on inventory management research for this, as with this approach the contextual factors of retail are not taken into consideration. A more differentiated approach is necessary. For example, a possible implication drawn from the inventory literature is the hypothesis that for turbulent products (i.e. products for which demand cannot be predicted) retailers should choose higher ASR levels. Yet in practice there are retailers that nevertheless prefer a manual approach for some high-demand-variance products such as fruit. The reason for this is that these retailers see their green grocery category as a way to differentiate themselves from the competition. As IT systems are not able to assess the quality of the products, these retailers prefer to have specialised planers making this decision manually. As this example shows, it would be wrong to give any recommendations without taking into consideration the context. This thesis therefore examines how the organization and the replenishment processes influence the performance of ASR systems. In section 6.5.3 technical and organizational requirements for each ASR system are presented, while in section 6.5.4 recommended store operations are proposed. A last implication of contingency theory is that ASR performance will also depend on product and store characteristics, as illustrated in the explanation model in section 4.3.

Nevertheless, one has to be aware of the limits of the contingency approach: Missing definition of term "fit" Explanation of "reality" limited The basic statement of contingency theory, namely that context and structure have to fit to each other, is at the same time the most critical part of this theory. The lack of

3. Literature Research

51

the definition of the term "fit" could alter the meaning and the relevance of the entire theory, as methodological rigor is missing (Schoonhoven 1981; Drazin and van de Ven 1985). As Bergeron, Raymond et al. (2001, p. 125) state: "Researchers were not cautious enough in defining the concept of fitwhich is central to any contingency modeland in selecting the most suitable data analysis approach to a given definition of fit." A second drawback of contingency theory is that it is useful for giving normative recommendations: in situation A, choose action B. Yet, it can be misleading to try to use this theory to explain existing structures and practices found in industry. An illustration of this statement is the paper by Ketokivi and Schroeder (2004) which shows that contingency arguments do not adequately explain why certain companies adopt innovative manufacturing practices such as TQM and lean management while others do not. Another example is Tidd (1993), which demonstrates that in many cases, organizational policies do not support technology strategies. This means that the contingency between a market's technological complexity and firm organization cannot be shown. In this example, there might be a tendency for companies to display certain structures depending on environmental influences. But there is no inevitability in this, other influences might be stronger, thus hiding the contingency effect. This obscuring effect can be the reason why in this thesis certain theoretical relationships do not correspond with observations made in reality.

3.5. Literature Research Overview

In the following table, the research streams analysed as well as exemplary sources are listed.

52

3. Literature Research

Research Stream Inventory management systems and OOS

Research Focus
Importance of inventory management for business Mathematical optimization of inventory Retail inventory control and OOS

Exemplary Sources
(Vollmann, Berry et al. 1992; Balachander and Farquhar 1994; Silver, Pyke et al. 1998; Dubelaar, Chow et al. 2001; Gudehus 2001)

(Galliher, Morse et al. 1959; Dalrymple 1964; Chang 1967; Groote 1994; Inderfurth and Minner 1998; Bassok 1999; Ketzenberg, Metters et al. 2000; Cachon 2001; Wagner 2002; Gudehus 2004) (Kotzan and Evanson 1969; Krueckenberg 1969; Cox 1970; Curhan 1972; Corstjens and Doyle 1981; Emmelhainz, Stock et al. 1991; Drze, Hoch et al. 1994; Andersen Consulting 1996; Taylor and Fawcett 2001; Zinn and Liu 2001; Gruen, Corsten et al. 2002; Sloot, Verhoef et al. 2002; Zinn, Mentzer et al. 2002; Roland Berger 2003b; Broekmeulen, van Donselaar et al. 2004a; Stlzle and Placzek 2004) (Zomerdijk and Vries 2002)

Organizational context of inventory control

Logistics and operations management

SCM: optimization along the supply chain

(Lee and Billington 1992; Fisher, Hammond et al. 1994; Womack and Jones 1996; Feitzinger and Lee 1997; Lee, Padmanabhan et al. 1997b; Christopher 1998; Anupindi 1999; Chen, Drezner et al. 2000; Hines, Holweg et al. 2000; Gudehus 2001; Duffy 2004; Pfohl 2004) (Lee and Billington 1993; Fisher, Hammond et al. 1994; Fiorito, May et al. 1995; Bell, Davies et al. 1997; Cottrill 1997; Christopher 1998; Closs, Roath et al. 1998; Lee 2001) (Davis 1993; Daugherty, Myers et al. 1999; Ellinger, Taylor et al. 1999; Stank, Daugherty et al. 1999; Achabal, McIntyre et al. 2000; Myers, Daugherty et al. 2000; Sabath, Autry et al. 2001; Kuk 2004) (Kurt Salmon Associates 1993; Cachon and Fisher 2000; Corsten 2002) (Raman, DeHoratius et al. 2001b; Wagner 2002; Fleisch and Tellkamp 2004; Tellkamp, Angerer et al. 2004; DeHoratius and Raman 2005) (Dreyfus and Dreyfus 1986; Akkermans, Bogerd et al. 1999; Chircu and Kauffman 2000; Markus, Axline et al. 2000; O'Leary 2000; Ptak and Schragenheim 2000; Kumar, Maheshwari et al. 2001; Hong and Kim 2002; Merkel 2003; Sarpola 2003) (Mumford 1934; Noble 1984; Fleck 1994; Bancroft, Seip et al. 1996; Ciborra 2000; Soh, Kien et al. 2000; Kallinikos 2004) (Zuboff 1988; March and Olsen 1989; Fleck 1994; Brynjolfsson and Hitt 1998; Markus, Axline et al. 2000; Sheer and Habermann 2000; Johansson 2002; Sawyer and Southwick 2002; Kallinikos 2004) (Metaxiotis, Psarras et al. 2003; Davenport 2004)

Demand-based supply chains (Pull-SC) Automatic replenishment programmes (e.g. VMI, CRP, QR) ECR and information sharing Retail operations management

Business information systems perspective

Guidelines for successful implementation of ERP systems Organizational implications Implications for employees and mode of work ERP and decision support systems

Contingency theory

Contingency organizational level Contingency at process level

(Lawrence and Lorsch 1967; Perrow 1967; Bobitt and Ford 1980; Pfohl and Zllner 1987; Staehle 1991; Cadeaux 1994; Kahn and Mentzer 1996) (Fiedler 1967; Pfeffer and Leblebici 1971; Huber 1984; Lederer and Mendelow 1990; Vessey 1991; Vessey and Galleta 1991; Chan and Huff 1993; Cadeaux 1994; Chow, Heaver et al. 1994; Sabherwal and Vijayasarathy 1994; Leidner and Jarvenpaa 1995; Chagu 1996; Bergeron, Raymond et al. 1998; Bergeron, Raymond et al. 2001)

Table 6: Summary of research streams perspectives

4. Development of Models

53

4. Development of Models
4.1. A Descriptive Model of Replenishment Systems
The first milestone in this work is the creation of a descriptive model of ASR systems. It provides an answer to the first research question Q1: how can automatic store replenishment systems be classified? This model does not claim to include all possible kinds of replenishment systems, as new, revolutionary, decentralized systems may be developed in the future.41 Yet it provides a helpful structure to examine the replenishment systems of today's retailers. As it is created on an abstract level, its core results can be applied to other industries dealing with piece goods replenishment processes as well. Regarded from a more abstract OR perspective, the store and the supplier are nodes in a network (cf. Carter and Price 2001). The store is herein viewed as a material sink, i.e. the place where the products are consumed.42 As stated before, material for the store can be either sourced from the retailer's DC or directly from a supplier. Obviously, supplier and retailer are only a small part of the entire distribution network. The initial source would be the raw material producer (e.g. a wheat supplier), the final sink would be the consumer (in contrast to the shopper43). In terms of information flows, stores and suppliers can be sinks and sources at the same time. In some systems, for example, the store sends an order to the supplier, the latter sends a dispatch advice back with the expected delivery time and quantity.44 The basic function of a replenishment system is to decide, when to order which amount (Wagner 2002). Characteristics of replenishment systems with make-to-stock policy (i.e. the product is already built when the order comes in) can therefore be captured with four elements (see Figure 10). To be able to decide which quantity to order, it is obviously necessary to know how many items are on stock. This is called inventory visibility, the ability to track the status of inventory in a supply chain tier or
41

For example, research has been conducted on transport networks where no central control is required, as the system controls itself by a non-centralized network of intelligent items (e.g. Gausemeier and Gehnen 1998). 42 As illustrated in Figure 5, the store can also be a material source, as products are sometimes sent back to the supplier. However, these quantities are negligible. 43 To illustrate the difference between shopper and consumer, here is an example: over 80% of all Gillette Mach-3 razor blades are purchased by women. However, over 90% of all Gillette Mach-3 razor blades are used by men. In this example, women are the shoppers and men are the consumers (Hofstetter 2004). In this thesis, though, no exact differentiation will be made between the two terms.
44

Cash flow between the parties, the third typical flow considered in logistics, does not play a role in the research framework of this project.

54

4. Development of Models

even across the entire supply chain (Woods 2002). The replenishment logic defines the business rules that are followed in order to decide when to order (Silver, Pyke et al. 1998). Some systems take into account order restrictions as well. For example, if the system can only order items in six-packs, an order of 10 units must not be placed. The last element is the forecasting part. Some systems try to anticipate future consumer demand by computing forecasts.

ForeForecasts casts

Order Order restrictions restrictions

Inventory visibility

Replenishment logic

Figure 10: Descriptive model of replenishment systems

Depending on the replenishment system, these four modules can be either IT-based or not. The sophistication of each of the four modules varies greatly between different systems. When automating replenishment processes, the question of which decision can be handled by the system autonomously and which not has to be examined. All replenishment systems have a replenishment logic module and require inventory visibility.45 The other two modules are facultative. For example, a Kanban-system functions without forecasts or explicit restrictions.46 In the following sections, the notations outlined in Table 7 will be used. The notation is based on Gudehus (2002), extended with elements of Silver and Pyke (1998) and the author's own contributions.

45

In theory, it is possible to have a replenishment system without a module for inventory visibility. Instead, the replenishment is triggered only by sales. Yet, the high shrinkage rates of grocery retailers would make this system very inefficient in practice. Kanbans are cards attached to units or containers that trigger replenishment. For a detailed description of Kanban systems see Christopher (1998, p. 185) and Gudehus (2004, p. 261).

46

4. Development of Models

55

C: Cost function Ci: Inventory carrying costs Cs : Ordering costs Coos: OOS costs D(i): Demand rate of item in period i in a store i: Time period iS: Time period with an order IA: Mean inventory accuracy In, system: Inventory record of SKU n in the system In, system: Real inventory stock quantity of SKU n in store L: Replenishment lead time m(i): Inventory position in store mO(i): Inventory on stock mS(i): Inventory ordered in store, not yet arrived n: SKU N: Number of SKUs Q: Fixed order quantity Q(i): Order quantity in period i Q*(i): Optimal order quantity for period i Qmax: Restriction, maximal order quantity Qmin: Restriction, minimal order quantity p: Period length s: Order point S: Order-up-to-level SS: Safety stock T: Order frequency z: Safety factor : D: Service level Standard deviation of demand

1, : Inverse of standard normal distribution

Table 7: Inventory notations

56

4. Development of Models

4.1.1. Inventory Visibility As stated before, inventory visibility is the ability to track the status of inventory across the supply chain (Woods 2002). This can be realized by defining "buckets" for every location and status of inventory. Simple systems do this with pen and paper, more sophisticated systems use identification technologies (e.g. barcode, RFID) and electronic systems (inventory management systems). For this research project, only the inventory in the store and in the pipeline towards the store is of interest. The physical stock level in a store is called the on hand stock mO(i). The inventory on the way to the store is called net stock mS(i). Both quantities have to be taken in to account before reordering. Therefore, the inventory position m(i) is defined as the sum of the on hand stock and net stock (see Figure 11). m(i)= mO(i)+ mS(i)

Inventory level

mO(i)

mS(i) m(i)

Q(i)

-D(i)

SS

iS L p

iS

Time

Figure 11: Exemplary time dependent course of the inventory stock level

47

Another important descriptive parameter that is closely related to inventory visibility is inventory accuracy IA. The accuracy of a store with N SKUs can be measured as the mean difference between recorded inventory levels and real inventory levels.

47

Adapted from Silver, Pyke et al. (1998, p. 250) and Gudehus (2004, p. 371).

4. Development of Models

57

N The inventory accuracy has not yet received the attention it deserves, though practitioners stress the importance of maintaining high data quality. There are very few papers concerning physical audits of retailers' inaccuracy levels; an exception is a study by DeHoratius and Raman (2005). The authors found, after having examined a US-retailer, that for 65% of the SKUs the inventory records in the systems do not equal the real physical inventory. An older empirical study shows that the distribution centres of a logistics operator had an inaccuracy rate of 36% (Millet 1994).

IA =

|I
1

n , system

I n ,real |

4.1.2. Replenishment Logic The store has to have an on hand stock mO(i) at the beginning of the period i that is bigger than the demand D(i), otherwise an OOS occurs and the demand cannot be completely filled. A replenishment system has to decide for every SKU whether it is time to place an order or not. Formally, it means that it has to decide whether i=iS. How does the system know that it is time to place the order? There are several sources dealing with the issue of describing the decision rules of inventory systems (e.g. Vollmann, Berry et al. 1992; Silver, Pyke et al. 1998; Arnold, Isermann et al. 2004; Stlzle, Heusler et al. 2004). The decision rules determines, when and in what quantity a replenishment order is made. The basic inventory decision rules are shown in Table 8.
Order Quantity Order Frequency Fixed Variable Fixed (T,Q) (s,Q)
48

Variable (T,S) (s,S)

Table 8: Basic inventory decision rules

The four basic decision rules are therefore: (T,Q) Order every T period the fixed quantity Q (T,S) Order every T period, fill up to level S (s,Q) Whenever inventory drops below s, order Q (s,S) Whenever inventory drops below s, fill up to level S

48

Adapted from Vollmann, Berry et al. (1992, p. 701).

58

4. Development of Models

For a discussion of the advantages of each decision rule see e.g. Silver, Pyke et al. (1998, section 7.3).

Order Frequency (s,Q) and (s,S) rules can have either a periodic review, making them (T,s,Q) and (T,s,S) or a continuous review (T=0). Most grocery retailers have a periodic review once per day (T=p=1 day). Systems that do not forecast use simple heuristics, such as the following: IF(m(i)<s; i=iS; 0)49 (An order will be placed in period i if the inventory position is below the reorder point). More sophisticated systems calculate whether there is a chance of an OOS in the case that the order is not placed until the next period. This means that the system compares whether the inventory position is larger than the demand at the time p+L (taking into account a certain desired service level). Consequently, the decision rule is IF(mO(i)<D(i+p+L); i=iS ; 0)

Order Quantity If the system decides to order, then the system orders the quantity Q(i): IF(i= iS; Q(i); 0) Which amount is to be ordered? Simple systems have a fixed quantity Q set or just fill up to a certain level S. More complex systems calculate the quantity Q(i) taking into account an optimal order quantity Q* and some other restrictions.50 This quantity is the result of optimizing a cost function C: C= CS+ CI+ COOS51

49

The function IF(A;B;C) stands for: If the condition A is true, than do B, otherwise do C. "0" stands for "do nothing." For an overview of existing restrictions see section 4.1.3. Cf. Gudehus (2004).

50 51

4. Development of Models

59

The cost function takes into account several parameters such as: CS: ordering costs - Cost of creating and placing orders This includes costs for creating the order, costs for sending it to the supplier and also costs for the supplier caused by the handling of the order. - Transport costs The cost of the physical transport of the goods, such as truck costs, road tolls and customs. CI: inventory carrying costs - Place-in storage costs These are costs that are created when the items are moved from the store gate to the back room. The time taken for reshelving is responsible for the major part of these costs. Additional costs may occur for bulky or heavy items that require special transport methods. - Unit variable cost Not the selling price of the unit, more its value. It is necessary together with the - Carrying charge to calculate the inventory carrying costs. The latter is the cost of having one money unit of the item tied up in inventory. - Backroom space cost The cost of having one unit in one backroom space. COOS: OOS costs These are probably the costs that are most hard to calculate. OOS costs can be split into lost sales, backorder and lost customers. Therefore, the computation needs not only to take into account the lost sales, but also the reduction in customer satisfaction and the risk of losing her for good. To reduce complexity, most retailers set a desired availability level and then calculate the optimal order quantity, without explicitly calculating the OOS costs.

Safety Stock It is worth noting that because the demand function has a certain distribution and variance, it is necessary to implement a safety stock to cope with this uncertainty. The lead time and the delivered quantity might have high volatility as well, therefore creating the need for additional safety stock. It is self-evident that the more volatile these factors are, the higher the safety stock will be to maintain a certain service

60

4. Development of Models

level. For instance, if the demand varies (in a normal standard distribution) with a standard deviation of D , then the safety stock will be SS= zD

p+L

where z is the safety factor. For normal distributions, z is calculated as z= 1, ( ) for a normal demand distribution N(,), where is the service level the company wants to maintain (Alicke 2003).

4.1.3. Order Restrictions Inventory replenishment systems are forced by many restrictions to deviate from an optimal calculated order quantity and replenishment time. For example, if minimum and maximum restrictions exist for the order quantity, the determination rule for the order quantity would be Q(i)= MAX(Qmin; MIN(Qmax;Q*))52 These order restrictions are necessary as retailers face certain business and logistics limitations. Many of these aspects were not taken into account in the optimization formula, and so the result has to be adapted. One of the most comprehensive order restrictions compilations was made by Bernard (1999) and it is seen in Table 9.
Restriction
Minimum quantity Maximum quantity

Type
Quantity constraint Quantity constraint

Explanation
Meet a specified minimum order quantity Meet a contractually agreed minimum quantity Meet a specified maximum order quantity Meet physical storage restrictions Meet physical shipping restrictions Meet an order minimum charge Limit orders to defined signature levels Security limit to reduce system errors Prevent multiple orders in a time period Reduce handling associated with frequent orders Constrain the order quantity to meet inventory turns limits Limit excessive purchases of low value parts Limit parts with a shelf life to defined days of supply Take advantage of volume discounts Order in case pack multiples Increase quantity to cover a minimum, average demand or period quantity Adjust quantity to account for spoilage, processing, shrinkage, inspections rejects, etc.
53

Minimum value Maximum value Minimum days supply Maximum days supply Price break quantity Rounding quantity Minimum demand quantity Order quantity multiplier

Cost constraint Cost constraint Time constraint Time constraint

Qualifier Qualifier Qualifier Qualifier

Table 9: Exemplary order restrictions

52

This formula assures that the calculated optimum quantity Q* is bigger that the minimum quantity Qmin and at the same time smaller than the maximum quantity Qmax.

4. Development of Models

61

Bernard (1999) differentiates between quantity restraints and qualifiers. The former provide upper or lower limits for the ordered quantity. There are numerous reasons to limit the order, among others transport restrictions, capacity restrictions, stock level at supplier restrictions, etc. Order quantity qualifiers (if/then rules) provide a micro-adjustment mechanism to fine-tune calculated quantities. This is undertaken to match the order quantities with specified case pack multiples. Both restrictions can be applied at SKU or category level. In manual systems, the order planners have to keep these restrictions in mind when ordering. They are also responsible for the internal consistency of the restrictions. The more sophisticated ASR systems are, the more restrictions automatically will be followed and less interventions will be necessary. Nevertheless, there will always be exceptions or temporary restrictions that have to be implemented manually (such as a strike at a supplier's plant).

4.1.4. Forecasts Types of Forecasting Techniques In this thesis, a forecast is regarded as a statement about the expected future demand. The importance of accurate forecasting has been stressed by several authors (e.g. Ritzman and King 1993; Moon, Mentzer et al. 1998)the success of entire companies depends on reliable forecasts. In theory, forecasts are not necessary if the probability distributions are already available. If this is the case, a formula can be directly applied to calculate the reorder point s, based on a service level the company wants to reach. Yet in practice, the demand function is unknown and remains a theoretical construct, making forecasts necessary (Kallinikos 2004). Hanke and Wichern et al. (2005) differentiate between two basic types of forecasting techniques: qualitative and quantitative (see Figure 12).

53

Source: adapted from Bernard (1999, p. 293).

62

4. Development of Models

Forecasting Techniques Forecasting Techniques

Qualitative Techniques Qualitative Techniques
Based on the subjective Based on the subjective expectations expectations and experiences of persons and experiences of persons

Quantitative Techniques Quantitative Techniques

Based on statistical analysis of historical data Based on statistical analysis of historical data One-dimensional One-dimensional Time series based Time series based forecasts forecasts Multidimensional Multidimensional Causal forecasts Causal forecasts

Figure 12: Qualitative and quantitative forecasting techniques

54

In practice, qualitative techniques, also known as judgemental techniques, are very popular among managers (Chase 1995). Such forecasts are based on the subjective expectations of a single person or group (Schneckenburger 2000). This means that demand is estimated based on personal experience and knowledge, and forecasts of this kind are often somewhat intuitive in character. The forecaster's judgement is a result of the mental extrapolation of past historical data (Hanke, Wichern et al. 2005). Popular techniques are (Mentzer and Kahn 1995): Estimates by the sales force ("sales force composite") Commissions of experts ("jury of executive opinion") Delphi method On the other hand, quantitative techniques are purely mechanical procedures that compute quantitative results and do not require any input of judgement. These statistical techniques estimate future demand based on historic sales data. There is a distinction made between one-dimensional forecasts (also called time series forecasts) and multidimensional forecasts (also called causal forecasts). The difference is that while time series techniques try to explain demand based on merely one parameter (i.e. time), causal techniques try to improve forecast accuracy by implementing more parameters that could have an influence on demand. Examples for such causal variables are, among others, price, weather and promotional activities. In the following, the focus is laid on quantitative techniques, as they are the ones that are used by higher ASR systems.

Accuracy of Forecasts A common statement in forecast literature is that every forecast is per se wrong (cf. Alicke 2003), as it is impossible to perfectly anticipate future sales. Therefore, a point
54

Source: based on Schneckenburger (2000).

4. Development of Models

63

forecast is never to be believed blindly. The credibility of a forecast is related to the strength of the theory underlying it (Wacker and Lummus 2002). The output is only an estimate of mean future demand (Wagner 2002); its accuracy will strongly depend on the variation of the demand. With formulas such as MAD (Mean Absolute Deviation) and MAPE (Mean Absolute Percent Error), the variability of forecasts can be measured: MAD=
1 N

| Di* Di
i =1
N

1 MAPE= N

i =1

| D* D* | i i

D* is the predicted demand, D the real demand. Managers face the challenge to choose the right forecasting methods. If two different forecasting methods are available, the one producing the lower MAD or MAPE is to be chosen. A practical solution to this problem was found by Fischhoff (1994). He has created a 4-level forecast classification enabling managers to decide how reliable a forecast is. Such classifications can be used to decide, whether an article can be ordered automatically or not (Toporowski and Herrmann 2003). All replenishment systems that make quantitative forecasts rely heavily on the quality of historical demand data. As in a typical store the demand cannot be observed directly (customers do not place an order, they just buy what is on the shelves), the actual sales data is regarded as an approximation to the true demand. It is not difficult to see that this approach is somewhat problematic. For example, if an OOS occurs and customers do not buy at all, then the sales data will underestimate the true demand. Wagner (2002) has coined the phrase "dirty data" for this and other similar problems. In Wagner's paper one can find a comprehensive list of data challenges companies face in reality that are often neglected in mathematical models. The data aggregation level has a major influence on forecast accuracy. An aggregation is possible regarding time dimensions (days-months-year), product dimensions (SKU-category) and geographical dimensions (city-region-country). The higher the aggregation level, the higher the forecast accuracy will be. However, at the same time the lower the information-richness of the forecast (Jain 2003). An example of aggregation levels found in retail is given by Achabal, McIntyre et al. (2000). The authors suggest for categories such as fashion goods, where due to different sizes, colours, etc. many SKUs exist, to pool four to ten articles within one category and to

64

4. Development of Models

use this to compute a common forecast. This clustering is made subjectively by selecting products that are expected to appeal to similar customers (e.g. a cluster of eight different styles of men's casual pants). For a stable forecast they recommend having at least 100 SKUs sold per time period. The allocation on store and SKU level is carried out based on historical fractions of sales. Achabal, McIntyre et al. claim that this is a better way to come to a forecast compared to store level forecasts. This is a contradiction for some software producers, who claim that only forecasts at SKU and store level will lead to the desired result (e.g. Beringe 2002).

Forecasts in Automatic Replenishment Systems Automatic replenishment systems that use forecasts have to decide at the time of order placement, whether the expected demand in the review period plus lead time is higher than the current inventory position. Some systems take into account in addition the uncertainty in demand and lead time. The order is triggered if the inventory position does not provide enough service protection; otherwise another period is waited (Wagner 2002) : IF(mk(i)> Dk(i+L+p) ; PLACE Order Qk(i); 0) An example of a multidimensional forecast formula used in an replenishment system is provided by Achabal, McIntyre et al. (2000). Their model is as follows: Sales = Baseline_sales * Seasonal_effects * Merchandising_effects This multidimensional model takes into account merchandising effects, effects from the elasticity of markdown prices, promotional frequency and length, advertisement size, among other events. However, the authors stress the importance of having as few parameters as possible ("parsimonious forecasts") as too many parameters can be counterproductive. In their experiment, with an increasing number of parameters the adjusted R-square increased in the estimation phase. This means that the models were quite good in explaining historical data. But these over-sophisticated forecasts were less accurate in prediction of future periods.

4.2. Classification of Automatic Replenishment Systems

A classification of the different ASR systems was developed based on the examination of dozens of retailers. Based on the degree of automation, the following classification system was developed, as seen in Figure 13.

4. Development of Models

65

ASR0 ASR0

ASR1 ASR1

ASR2 ASR2

ASR3 ASR3

ASR4 ASR4

Manual Manual ordering ordering

Electronic Electronic inventoryinventorybased based ordering ordering

Simple Simple ordering ordering heuristics heuristics

Time seriesTime seriesbased based forecasts forecasts

Causal Causal modelmodelbased based forecasts forecasts

Manual

IT-supported

Sales-based

Demand-based

Sophistication of the store replenishment system

Figure 13: Classification of automatic replenishment systems

In the following, the differences between the single levels of the presented classification will be discussed.55

ASR0: Manual Replenishment Systems This is the level where all replenishment systems were before retailers introduced IT systems. A manual replenishment system relies solely on the information and intelligence of the employees. This kind of system is still in use nowadays; in grocery retailers (e.g. VOLG), bakeries (Kamps), discounters (Denner) and opticians (McOptic). But only three out of 12 major European grocery retailers still rely solely on such manual ordering systems (Smros, Angerer et al. 2004a).

ASR1: Electronic Inventory-Based Replenishment Systems As soon as retailers have their inventory levels administered by an IT system, one can say that a state of semi-automation has been attained.56 In the descriptive model, this step means that the first module "Inventory Visibility" is represented electronically in the IT system. With this electronic information, it is for the first time possible to move replenishment decisions away from the store. A centralized system is now thinkable, where a central organizational unit monitors inventories at all the outlets and makes decisions concerning deliveries. In order to do so, a certain degree of inventory records accuracy has to be reached. It is worth noting that decisions about the time and quantity of the order are still made manually; therefore, the module "Replenishment Logic" is still not IT-based. Order restrictions can be displayed on an IT systems, but the planner still has to take them

55 56

See Table 10 for an overview of ASR differences. The term semi-automation is used because the inventory records are now automatically administered by the IT system, yet the ordering itself still takes place manually.

66

4. Development of Models

into account "manually" when placing the order. Fresh products, such as fruits and vegetables, are often ordered this way. Examples of companies using predominantly ASR1 systems are consumer electronics retailers (e.g. Media Markt CH) and book stores (Orell Fssli).

ASR2: Simple Replenishment Heuristics Basically, this is the first level where true automation starts. To reach this level, only a small step is necessary from ASR1 systems. At that level, the system already stores electronically inventory quantities. Consequently, just by implementing a simple heuristic it is possible to make the replenishment system work without any employee intervention. One of the most common heuristics used in retail is the order-point order-quantity (minimum-maximum) heuristic. The inventory management system is updated after each transaction or at least before a new order is to be placed. The ordered amount is either a fixed quantity (Q) or the difference between the actual inventory level and a desired inventory level (S). These replenishment parameters are set beforehand and are valid over a long period (months to years). This kind of replenishment system is called consumption-based, as the consumption of the items at the POS releases the order (pull-system). With such a replenishment logic no forecasts are necessary. This ASR level is the first at which basic order restrictions can be automatically accounted for (e.g. the system follows the quantity restrictions when placing an order). An example of companies with ASR2 systems are grocery retailers, such as Coop and Migros, along with specialised retailers dealing with cosmetics (Body Shop), mobile phones (Orange Zone), household electronics (Fust), among others. Later in this thesis an ASR2 system with a special difference is examined. The replenishment parameters are set at category level and not at SKU product level. The label ASR2* is introduced to make the difference clear.57

ASR3: Time Series-Based Forecasts At this level, for the first time the forecasting module is regularly in use. The system makes an estimate of the next period's demand and orders according to it. This system basically checks whether it is necessary to place an order in this period, keeping the actual stock level in mind, or whether it suffices to place the order in the next ordering cycle. As the demand is only a prediction, decisions are made taking into account a defined risk of OOS. The order quantity is either a cost-optimal amount

4. Development of Models

67

Q* or determined by a preset order-up-to-level point S. The main difference to the former levels is that the placing of the order becomes demand driven. The forecast is created by observing the behaviour of demand data in the past and making assumptions about future consumption. On this level, forecasts are computed with time series methods (one-dimensional techniques). More sophisticated programs take into account trends, cyclical variations, seasonal patterns and irregular random fluctuations (cf. Silver, Pyke et al. 1998). However, even such complex algorithms remain one-dimensional, as the only independent variable is time. As the IT systems used in ASR3 systems are more advanced, it is theoretically possible for them to take into account more order restrictions when ordering. Retailers such as Spar Switzerland, Orange SA and Spengler use these systems for some of their products.

ASR4: Causal Model-Based Forecasts The difference between ASR4 systems and the former level is the complexity of the forecasts used. Causal-based forecasts are obtained by taking into account not only past patterns, but also by anticipating the effects that future events will have. These systems claim to be able to predict the effect of promotions, price reductions, actions of the competition, the weather and more, on the demand. For this, the systems have to understand and compute the underlying effects influencing consumer demand. Demand is consequently treated as the dependent variable, and explained within a causal model with several, independent variables. This kind of system represents at the moment the most sophisticated software tools available on the market. These causal models have been successfully implemented, for example, at retailers such as dm-drogeriemarkt, Woolworth and Metro Group.

ASR5: Multi-Echelon Systems This level somehow breaks with the linear order of Figure 13. It is possible to enhance any of the ASR04 systems towards a multi-echelon replenishment system. This means that a system simultaneously controls the inventory stock and makes replenishment decisions on several levels of the supply chain. An example would be an ASR system that automatically and simultaneously calculates the optimal orders for retailers' stores and DCs. Such systems would need even more intelligence than normal systems, as the complexity of a multi-echelon system increases. The optimization becomes more difficult due to more restrictions. As stated on section

57

The difference between ASR2 and ASR2* is described in detail in section 6.2.2 in the case of MYFOOD.

68

4. Development of Models

2.1, multi-echelon systems are not included in the focus of this thesis, therefore they will not be further examined here. The following table summarizes the differences of ASR systems.
ASR0 Inventory visibility
Manual records Simple heuristics, e.g. (T,Q) (T,S) (s,Q), (s,S) Medium complex restrictions implemented into IT systems

ASR1

ASR2

ASR3

ASR4

Electronic inventory record system

Necessary inventory records accuracy increases

Replenishment logic

Manual decision: time and quantity

(s,Q*) or (s,S)

Order restrictions

Manual consideration of restrictions

Basic restrictions shown in IT systems

Complex restrictions implemented into IT systems

Forecasts
None or qualitative techniques None

Unidimensional (time seriesbased) forecasts

Multidimensional (causal) forecasts

Table 10: Characteristics of automatic replenishment levels

4.3. Explanatory Model

4.3.1. Purpose and Structure of the Explanatory Model The explanatory model in this paper is intended to be a representation of predicted relationships between factors that influence ASR performance. The explanatory model helps to quantify ASR1ASR4 system benefits (research question Q2) and to find the adequate level of automation for each retailer (Q4). The definition of the unit of analysis is critical for the further understanding of the model. In theory, retailers will face an optimal automation level for each SKU in every single store. Yet for practical reasons, a retailer can only have a certain number of ASR systems running in parallel. Therefore, retailers will have to define product clusters of items with similar characteristics that will be managed through a single system that fits the characteristics of that cluster best. For instance, a cluster could comprise slow-moving canned products with good demand predictability and low value. The explanatory model is also the basis for giving recommendations as to how to change the organization and replenishment processes (Q5).

4. Development of Models

69

A question that has occupied many practitioners is the question of reasons underlying OOS incidents. In the literature, there are many sources dealing with this question (eg.Gruen, Corsten et al. 2002; Roland Berger 2003b). Although these sources examine the underlying reasons, they remain descriptive. Furthermore, a question that has not been discussed in full detail in theory is the quantitative correlation between OOS and other variables such as product, store and replenishment system characteristics. In the model constructed in this thesis, it is postulated that it is possible to explain, at least partly, the OOS and inventory level performance of retailers by having the product, store and replenishment characteristics as independent variables. Therefore, the additive model that is explored in this thesis is:

Y= M + ASR + P + S + (ASR x P) + (ASR x S) + e

where Y: M: P: S: ASR: ASR x P: ASR x S: e:

OOS rate or inventory performance measure Overall mean Effect from the product characteristics Effect from the store and personnel characteristics Effect from the replenishment system in use Interaction effect replenishment system and product Interaction effect between replenishment system and store Error term, unidentified effects

The model consists of three main effects and two interaction terms. The first effect implemented is the replenishment system characteristics (ASR). It is the main hypothesis of this thesis that the inventory performance and the OOS rate58 can be influenced decisively by ASR system choice. The second main effect in the model consists of product characteristics. There are several statements found in literature and practice that product characteristics will influence the OOS rate and inventory performance (e.g. Andersen Consulting 1996; Grocery Manufacturers of America and Roland Berger 2002). For example, from the
58

In this thesis, the OOS definition formulated by Gruen, Corsten et al. (2002) is employed: the OOS rate is measured as a percentage of SKUs that are out-of-stock on the retail store shelf at a particular moment in time (i.e., the consumer expects to find the item but it is not available).

70

4. Development of Models

inventory management literature it is known that products with higher demand uncertainty need higher stock levels to achieve the same availability percentage (Alicke 2003).59 The third main effect implemented in the model consists of store characteristics. It is self-evident that this variable can play an important role in inventory performance. Several studies have stressed that OOS rates vary widely between stores (e.g. Gruen, Corsten et al. 2002; Roland Berger 2003b). In the case of MYFOODthe retailer that will be studied in detail in the quantitative analysisall stores have exactly the same distribution system; therefore logistics should not play a role. But the stores have different sizes, assortments, regions and managers. These differences most probably influence store performance. Furthermore, another difference in the results could stem from the personnel, as they still have a significant influence on the ordering and the quality of store operations (Grocery Manufacturers of America and Roland Berger 2002). Therefore, it is sensible to include this variable in the model, as a significant influence on the dependent variable can be expected. In addition to these three main effects, two interaction effects are included in the model. If the characteristics of a product have an influence on replenishment system performance, the ASR x P variable should be statistically significant. This would be the case, for instance, if store employees are influenced by the product's price when ordering while the ASR systems are not. One of the interviewees stressed that some store managers tend to have higher availability rates for products which they personally like. A machine doing the order is emotionless and is therefore less influenced by product characteristics. A similar result can be expected from the ASR x S interaction. It can be assumed that for retailers, such as MYFOOD, where store employees have greater influence on replenishment parameters, ASR performance will differ from store to store, making this interaction variable significant. In the following, hypotheses concerning possible interrelations between the dependent variables OOS and inventory performance and other independent variables are constructed. These hypotheses are constructed based on results from KLOG projects, interviews with practitioners and literature research. The hypotheses are tested in chapter 5 with the help of data from the grocery retailer MYFOOD. As
59

Sometimes the correlation is not as intuitive. For instance, Kotler and Bliemel (1999) emphasise that only 5% of unsatisfied customers complain to store employees. Products where consumers complain more often will probably tend to have a higher availability, as the employees are aware of the importance of the problem. As customers will tend to complain for certain products more than for others, there is clearly a strong relation between the product characteristics and the OOS rate.

4. Development of Models

71

this company has mostly ASR0 and ASR2 systems in use, the hypotheses are based on a comparison of these two systems. The quantitative analysis of the hypotheses will be carried out with the help of different types of analysis of variance.60

4.3.2. Hypothesis Development: Product Characteristics Sales Variance For a retailer it is more challenging to have the right amount of items on the shelves if there is a large sales variance in a store. There are two problems that arise when facing irregular selling behaviour. First, the replenishment system must be flexible enough to cope with the increased complexity. For example, I must be able to order on one day 50, on another only 5 items of a certain product. The case pack size (CU/TU ratio) will play a role here as it determines the minimum possible order quantity. Nevertheless, much more difficult to handle is the second effect. If, in addition to the high sales variance of a certain product, there is also low predictability, the replenishment becomes very complex.61 The amount of safety stock is strongly influenced by this parameter: higher sales variance leads to higher inventory levels (Krane 1994; Fafchamps, Gunning et al. 2000; Alicke 2003). The higher the demand uncertainty, the higher the OOS rates will be. For store employees, it is very difficult to keep track of inventory levels for products that have high sales variance. Besides, some store managers prefer regular ordering patterns, i.e. they order every second Monday a case pack (cf. Broekmeulen, van Donselaar et al. 2005). It is obvious that the bigger the sales variance, the worse such simple ordering patterns will work. For automatic systems, such as ASR2, inventory visibility is not affected by the sales rate or variance, the system always knows the quantity in the stores (inventory records accuracy aside). Besides, ASR2 systems decide every replenishment period (i.e. every day in the case of MYFOOD) whether it is necessary to place the order. For these reasons, one can expect that store personnel will have more difficulties replenishing correctly high-sales-variance products compared to ASR2 systems. Consequently, the following relationships are expected:

60

61

This procedure goes hand in hand with Backhaus, Erichson et al.'s (2003) recommendations, as the authors emphasise the necessity of having a priori hypotheses about the effects of the independent variables on the dependent variable before using ANOVAs. For the products analysed in this thesis it is not possible to say whether a product has a good predictability or not. To be able to compare the variance of slow and fast moving products the sales coefficient of variance is used. This coefficient is calculated by dividing the daily sales' standard deviation of each product by its mean sales.

72

4. Development of Models

H1a: ASR0 products62 with high sales variance have more OOSs than products with a low sales variance. H1b: For ASR2 products, the influence of sales variance on the OOS rate is smaller than for ASR0 systems. H1c: ASR0 products with a high sales variance have higher inventory levels than products with a low sales variance. H1d: For ASR2 products, the influence of sales variance on inventory level is smaller than for ASR0 products.

Speed of Turnover A product's purchasing frequency can have an influence on OOS rate. In the study by Andersen Consulting (1996) the OOS rates were higher for fast-moving products. Products that sell very often per day have to be replenished more often than those with lower sales. Therefore, the danger of an OOS is far greater. On the other side, very slow moving goods that are only seldom sold have the danger of being overlooked when ordered manually. In the study by Stlzle and Placzek (2004) slow moving products had the highest OOS rate. The authors explain this by the fact that slow moving products receive less attention. This statement is confirmed in the study by Andersen Consulting (1996), where the authors, in addition, stress the risk for these products of remaining OOS for a long time. Therefore, for manual systems (ASR0) a quadratic curve is expected, meaning that products with very low and very high speed of turnover will have the most OOS incidents. As automatic systems do not have the danger of overlooking orders, only a small influence of the speed of turnover on the OOS rate should occur. The inventory level in a store is determined by logistics aspects but also by marketing aspects. Retailers want to avoid having too few products on the shelf. The reason for this is explained by one interviewed store manager, who said, "we create demand by having full shelves." Therefore, one can expect to see for slow moving SKUs more on stock than is actually necessary from a logistical point of view. In addition, for some slow moving units a single case pack size lasts for weeks, therefore increasing average inventory levels. This means that the lower the turnover of a product, the higher will be the inventory range of coverage. The situation is similar for ASR2 systems. Their inventory level will also depend on case pack size and turning speed: the slow moving products will have higher inventory levels. Yet, there should be a difference for very slow moving products. While store managers tend to have a simple procedure for such items (e.g. order a case pack every third week) (cf.
62

The wording "ASRx product" is a short form for "a product that is ordered through an ASR level x system."

4. Development of Models

73

Broekmeulen, van Donselaar et al. 2005), MYFOOD's ASR2 systems check the inventory position daily, therefore optimizing the ordering and thus reducing inventory. Overall, it is hypothesised: H2a: ASR0 products have higher OOS rates for very fast moving and very slow moving articles than for the other products. H2b: For ASR2 products, the influence of speed of turnover on the OOS rate is smaller than for ASR0 products. H2c: ASR0 products with a low speed of turnover have higher inventory levels than fast moving products. H2d: For ASR2 products, the influence of speed of turnover on inventory levels is smaller than for ASR0 products.

Price The effect of price on OOS and inventory level is ambiguous. DeHoratius and Raman (2002) found that inexpensive goods were more likely to have inaccurate records than expensive ones, as expensive SKUs receive more attention. According to this fact, one would expect more expensive goods to be less often OOS, for both manual and automatic systems. Normally, store managers would try to reduce the amount of expensive goods on stock, as they are bounded capital. Yet, in the data set of this thesis the most expensive goods are only worth 17 euro, 90% of all goods are cheaper than six euro. This price range is very small compared to the range in DeHoratius and Raman's (2002) work, where the most expensive goods had a value of 2,800 euro. Yet, as Broekmeulen, van Donselaar et al. (2004b) show, the inventory costs are secondary for typical grocery retailers, as they only represent 10% of logistics costs. Much more important are the handling costs in the store, which are five times as high. Therefore, it is reasonable to argue that for the observed grocery retailer MYFOOD another relationship between price and inventory performance will appear. Assuming that the more expensive goods have higher margins, the store manager will tend to have a higher inventory level on the shelves to avoid any OOS and consequently customer dissatisfaction. Yet, it is not clear if this strategy is successful, as many sources from the just-in-time and lean management literature have stressed the negative impact of high stock levels on availability (Schonberger 1982; Krafcik 1988; Fisher 2004). The higher inventory levels may even worsen the availability rate of these products. For automatic systems, on the other hand, the product price should not play such an important role, as the automatic system reorders the product despite its price. Based on this research it is expected that: H3a: ASR0 products with high prices have more OOSs than lower priced goods.

74

4. Development of Models

H3b: For ASR2 products, the influence of price on OOS is smaller than for ASR0 products. H3c: ASR0 products with higher prices have higher inventory levels than low-priced goods. H3d: For ASR2 products, the influence of price on inventory level is smaller than for ASR0 products.

Case Pack Size (CU/TU) Trading units (TUs) are the cases used to transport the products to the stores and contain several small consumer units (CUs)the units that are finally bought by shoppers. For a typical grocery retailer, the size of the case packs is the same for all stores, as it is often determined by the supplier without taking into account the store's individual demand. The consequence is high inventory levels. Broekmeulen, van Donselaar et al. (2005) found in their research that 96% of the items in their sample had a larger order case pack size than the stores needed in a reordering period. Therefore, stores typically order a single case pack unit or wait a longer time to place the order. The higher the CU/TU ratio, the stronger this effect and the more inventory will be on stock. On the other hand, the smaller the CU/TU ratio is, the better a store's replenishment system can work. It is not only manual systems that are influenced by the CU/TU ratio. Falck (2005) has shown in theory that ASR3 systems will only manage to reduce inventory compared to ASR2 systems if the CU/TU ratio is small.63 Therefore, it is hypothesised that ASR2 systems will also be influenced by the CU/TU ratio. Big case packs force employees to place orders more seldom. Therefore, the danger of forgetting to replenish the item is greater. As automatic systems do not forget any order, one can assume that the OOS rate will be higher for manually ordered products compared to ASR2 ordered products. One effect that might cover the influence of the CU/TU ratio on OOS rate is shelf space. Depending on the shelf space allocation, a certain case pack may or may not be beneficial. Wegener (2002) hypothesises that the OOS rate will strongly be

63

It is worth noting that the resulting effect of the CU/TU ratio on performance depends on the store. Small stores are more affected by the CU/TU ratio than bigger stores. For example, for some non-food products of MYFOOD (e.g. shoe laces) the smallest case pack lasts in small stores for several months, while in a bigger store one case pack is sold within days. Therefore, instead of looking at the absolute case pack value, all analyses are made looking at the relative case pack size (case pack size divided by average daily sales), analogous to Falck's procedure.

4. Development of Models

75

influenced by whether it is possible to place the whole case pack directly on the shelf or whether part of its contents has to be stored in the backroom. The replenishment from the backroom not only creates high handling costs, it is also one of the major reasons for OOS incidents (Gruen, Corsten et al. 2002). Taking these studies into consideration, it is hypothesised: H4a: ASR0 products with a small CU/TU ratio have smaller OOS rates than products with a bigger CU/TU ratio. H4b: For ASR2 products, the influence of CU/TU on the OOS rate is smaller than for ASR0 products. H4c: Products with a small CU/TU ratio have lower inventory levels than products with a bigger CU/TU ratio. H4d: Products with a small CU/TU ratio have lower inventory levels than products with a bigger CU/TU ratio.

Product Size In the study by Andersen Consulting (1996) different findings on the effect of product size on OOS are encountered. One could argue that bulky products need more space on the shelf; therefore, retailers may opt to allocate them less shelf space than would be advisable from a logistical point of view. Consequently, small products will tend to have more facings to increase visibility to customers and therefore should be comparatively more on stock in the store. Wegener's (2002) data analysis of four product families from the beauty and household category showed that products with bigger package size were indeed more often OOS than products with small package sizes. The explanation for this effect given by an interviewed manufacturer was that especially for bulky products that are not visually pleasing (such as toilet paper), retailers tend to reduce the amount of facings on the shelves. Yet Wegener (2002) relativises this result: for other product categories the correlation effect of product size and OOS is likely to be smaller, as other product characteristics (such as speed of turnover) are probably the predominant influence. As regards inventory level, Broekmeulen, van Donselaar et al. (2004b) found in their analysis of two Dutch retailers that smaller products had more shelf space than required from a logistical point of view. Therefore, one can expect to see higher inventory levels for small products. For automatic systems, on the other hand, the product size should not play such an important role, as the automatic system reorders the product despite its size. This is why it is predicted that: H5a: ASR0 products with a small size have lower OOS rates than big-size products.

76

4. Development of Models

H5b: For ASR2 products, the influence of product size on the OOS rate is smaller than for ASR0 products. H5c: ASR0 products with a big size have lower inventory levels than small products. H5d: For ASR2 systems, the influence of product size on inventory level is smaller than for ASR0 products.

Shelf Life Retailers are constantly aiming to offer products with longer remaining shelf life, as consumers demand fresh products (Petrak 2005). Therefore, for products with a very short shelf life (e.g. fresh pasta), store managers seek to reduce the amount of store inventory. It is clear that the smaller the average inventory of any SKU (compared relatively to its mean sales) the fresher the product will be due to a higher turnover. Therefore, it is reasonable to expect low inventory levels for short-shelf-life products. This finding is supported by physical audits such as that carried out by Broekmeulen, van Donselaar et al. (2004b). Small (safety) stocks reduce the danger of spoilage, but at the same time increase the danger of OOS. On the other hand, fresh products such as green grocery are, in terms of marketing, the most important goods for many grocery retailers. Therefore, skilled store employees will pay extra attention to these products. In the study by Stlzle and Placzek (2004), cooled products with a short shelf life had smaller OOS rates. For products with a longer shelf life the retailer can afford to have bigger quantities stored, as the danger of spoilage is less. Consequently, inventory levels will increase as well as the time between the ordering thus increasing the danger of OOS incidents. Overall, this leads to a quadratic effect of shelf life on OOS: products with very short and very long shelf life will experience more OOS situations. Yet, the effect of shelf life on OOS could be covered by other factors, such as sales variance, that have probably a stronger influence on the inventory level. The reasons mentioned for the influences of shelf life on inventory performance do not hold for automatic systems. Systems such as ASR2 do not pay more or less attention to products with a certain shelf life, consequently there should be no influencing effect visible. Overall, it can be expected that:

H6a: ASR0 products with a very short shelf life have an increased OOS rate, as well as products with a very long shelf life.

4. Development of Models

77

H6b: For ASR2 products, the influence of shelf life on the OOS rate is smaller than for ASR0 products. H6c: ASR0 products with a long shelf life have higher inventory levels than products with a short shelf life. H6d: For ASR2 products, the influence of shelf life on inventory levels is smaller than for ASR0 products. The following table summarizes the product characteristics hypotheses H1H6:
Dependent Variables Independent Variables Sales variance Speed of turnover Price CU/TU Product size Shelf life Out-of-Stock Inventory Level

+ 0 +-+ 0 + 0 + 0 + 0 +-+ 0

(H1a, ASR0) (H1b, ASR2) (H2a, ASR0) (H2b, ASR2) (H3a, ASR0) (H3b, ASR2) (H4a, ASR0) (H4b, ASR2) (H5a, ASR0) (H5b, ASR2) (H6a, ASR0) (H6b, ASR2)

+ 0 0 + 0 + + 0 + 0

(H1c, ASR0) (H1d, ASR2) (H2c, ASR0) (H2d, ASR2) (H3c ASR0) (H3d, ASR2) (H4c, ASR0) (H4d, ASR2) (H5c ASR0) (H5d, ASR2) (H6c, ASR0) (H6d, ASR2)

"+" Positive effect "-" Negative effect "+ - +" quadratic relationship

"0" Smaller effect of ASR2 system compared to ASR0

Table 11: Overview of hypotheses concerning product characteristics

4.3.3. Hypothesis Development: Store Characteristics Several descriptive OOS studies have mentioned that there is a strong variance in the OOS rates between countries and retailers studied (e.g. Gruen, Corsten et al. 2002; Roland Berger 2003b). The same sources also indicate that the OOS rate differs from store to store, even though the stores' contexts (i.e. ASR system, distribution network, assortment, marketing activities, etc.) are fairly similar. This latter phenomenon will be analysed in the data of MYFOOD. In the following, hypotheses concerning the influence of store characteristics on the OOS rate are derived, as the understanding of these influences is important when assessing the effect of ASR systems.

78

4. Development of Models

Different Performance Between Stores As stated before, one of the common results found in previous availability studies is that the OOS rate differs significantly between stores (e.g. Gruen, Corsten et al. 2002; Roland Berger 2003b; Stlzle and Placzek 2004). A transfer thought would be that the same relationship holds true for inventory performance (i.e. inventory levels and inventory variance). As MYFOOD managers are at liberty to change the parameters of their replenishment system they have an important influence on the inventory level. The hypotheses are therefore: H7a: The mean OOS level of a store differs from store to store. H7b: The mean inventory performance of a store differs from store to store.

Shrinkage and OOS The pivotal question is how some stores manage to have lower OOS rates other stores facing the same logistical context and using the same replenishment systems. A possible answer could be that store managers decide willingly to reduce OOSs by having more shrinkage. Yet, the interviewed practitioners from MYFOOD stressed the importance of the capabilities of the store managers to the performance of the store. The literature has often stressed the importance of good management for performance (e.g. Fahrni 1998). In MYFOOD's interviewees' opinion, stores with experienced mangers are able to reduce shrinkage and OOSs at the same time. Consequently, the hypothesis is: H8: There are some excellently-managed stores that are able to reduce OOSs and shrinkage at the same time.

SKU Density The more complex a store is, the more difficult it will be to manage. Raman, DeHoratius et al. (2001a) report how difficult it is for employees ordering manually to keep track of a product's availability. Yet it would be misleading to compare the absolute number of products between stores. Bigger stores will tend to have higher sales and a broader product range, but at the same time there will be more store staff and more space on the shelves for the product. Therefore one cannot conclude that bigger stores will have higher OOS rates. A way to compare different-sized stores may be looking at the SKU density. The assumption is that the amount of products per square metre is a good way of quantifying the complexity of a particular store. This variable has also been used by DeHoratius and Raman (2002). The authors were able to prove a positive correlation between the inaccuracy of records in a store and the SKU density. The authors argue that in crowded stores it is more difficult to

4. Development of Models

79

track data inaccuracy, therefore increasing the danger of OOS incidents. Besides, the more products a store has per m2, the less space there will be on the shelves for each SKU, thus increasing the chances of an OOS. The hypothesis reflecting this considerations is: H9: Stores with high SKU density will have more OOSs than low-density stores.

Work Intensity The economic theory of work states that there is a quadratic relationship of work effort and efficiency (cf. Fairris 2004). Transferred to this context this means that the number of employees in the store should also have a quadratic influence on the quality of store operations. When there are very few employees in a store, every single one has to work more; exhaustion and stress can set in, thus reducing the efficiency of work. The replenishment of shelves and the accuracy of data will suffer. Gruen, Corsten et al. (2002) found overworked staff to be one of the reasons for OOSs. On the other hand, if there are too many employees, boredom can set in (Fairris 2004), motivation can sink (Connell 2001), the responsibility for tasks can become unclear, and store managers find it more difficult to coordinate tasks (Anon. 2004). Wegener (2002) states that the marginal rate of productivity sinks, therefore she expects stores with an average personnel intensity to have the lowest OOS rates. To be comparable, the number of staff should be seen relative to the size of the store. The bigger a store the more the staff necessary to manage the shelves accurately. Overall, the hypothesis can be formulated as: H10: Stores with too many or too few employees per m2 sales area will have more OOSs.

Store Managers' Experience For retailers that have a very decentralized decision system, the single store can be regarded as a little independent company. Therefore, the influence of the store manager as an entrepreneur on the performance of the store will be significant, as stated in classical entrepreneur theory (cf. Schumpeter 1935; Fahrni 1998). Stores with good leadership will tend to have lower OOS rates. The more experience the store manager has, the better the performance of the store will be as store operations have a radical impact on the OOS rate (Gruen, Corsten et al. 2002). In the case of MYFOOD, store operations are still within the responsibility of store managers; they have a strong influence on replenishment parameters. A rather crude way to asses the experience of a store manager is by looking at the number of years he or she has worked in that particular store. The longer the store manager has been working in a

80

4. Development of Models

particular store, the better the performance should be. Already Raman, DeHoratius et al. (2001a) confirm that store managers who have been working for a long time on a particular store had a smaller rate of misplaced SKUs (i.e. items that were in the backroom but not on the shelf). Another advantage of experienced managers is that they have gained knowledge about the demand behaviour of their customers. For instance, experienced managers know to quantify the influence of a soccer game in the city on beverage consumption. However, an effect of diminishing returns will most probably appear. This means that for managers that have been on site for a long time, additional years in the store only result in small improvements. Based on these statements it is hypothesised: H11: Store managers that have been working in the same store for a long time have lower OOS rates than other store managers.

Backroom Authors such as Krafcik (1988) and Schonberger (1982) from the just-in-time and lean management philosophy argue that inventory hides problems in manufacturing. Transferred to this thesis, this means that having too much stock can be counterproductive to the availability rate. First, in backrooms that are crammed with goods, it will be very difficult to achieve a high visibility. And second, high inventory impedes process improvement, as errors of suboptimal replenishment systems are obscured by high inventories. Raman, DeHoratius et al. (2001a) report on a case in which a certain shop suffered a substantial deterioration in performance after the capacity of the backroom was increased. Fisher (2004, p. 14) therefore demands from retailers "to get rid of the backroom." If this is really recommendable can be tested by the following hypothesis: H12: Stores with large backrooms have higher OOS rates than stores with small backrooms.

Customer Satisfaction Finally, a question asked by many retailers is, whether it is worth putting effort into improving the OOS rate at all. Although a high availability is crucial for the satisfaction of consumers (Sterns, Unger et al. 1981; Angerer 2004) one could argue that by investing too much time in increasing availability (by having, for example, store employees replenishing the shelves more frequently) the direct service to the customer could suffer, as e.g. less check-out lanes are open. In addition accurate scanning of goods takes longer, and is therefore not welcomed by customers.

4. Development of Models

81

Therefore, it is interesting to analyse whether there is indeed a positive correlation between customer satisfaction and OOS rate: H13: Stores with lower OOS rates have more satisfied customers than stores with high OOS rates. The following table summarizes the store characteristics hypotheses H7H13:
OOS rate differs from store to store Inventory performance differs from store to store Reduction of OOSs and shrinkage is at the same time possible The more SKUs per m the more OOSs Too many and too few staff increases OOSs More experienced store managers have fewer OOSs Bigger backrooms lead to more OOSs Stores with low OOS rates have higher customer satisfaction Table 12: Overview of hypotheses concerning store characteristics
2

(H7a) (H7b) (H8) (H9) (H10) (H11) (H12) (H13)

4.3.4. Hypothesis Development: ASR Characteristics The last group of hypotheses test one of the main questions of this thesis (Q2): can retailers achieve a better replenishment performance by using higher ASR systems? From the theoretical research of ASR-related topics64 it is known that ASR systems should be able to reduce OOSs (cf. Achabal, McIntyre et al. 2000), decrease inventory level (cf. Ellinger, Taylor et al. 1999), have lower inventory variance and show robust results compared to manual systems (cf. Closs, Roath et al. 1998). Therefore, these performance measures have been chosen to test the capabilities of higher ASR systems.

OOS Performance and ASR Level Fisher, Raman et al. (2000) point out that many retailers still use very basic methods for their ordering. According to the authors, rules of thumb and gut feelings are often used when determining the quantities to be ordered. They see a great opportunity in mixing "art and science" by using the vast possibilities of new technologies and data availability to improve retailers' replenishment systems. Such technologies are the ASR systems. It would be a major relief for retailers to have more reliable replenishment system, as 72% of all OOSs are caused by faulty store ordering and
64

See research on ARPs in section 3.2.

82

4. Development of Models

replenishment practices (Gruen, Corsten et al. 2002). The reports from practitioners (e.g. Beringe 2002) and the literature (e.g. Mau 2000) of the reduction of OOSs by ASR systems are very courageous, yet there are no tests available in academia that prove these statements. Therefore, this thesis aims to close this gap by showing that such systems do indeed have the potential of reducing OOSs caused by replenishment faults. As data for ASR2 systems is available, it is hypothesised: H14: If the ASR level is chosen and parameterised correctly for a certain product category, ASR2 products have less OOS incidents than ASR0 products.

Inventory Performance of ASR Systems While for the OOS performance the situation is clearthe fewer OOS incidents in a store the betterit is slightly more difficult to assess inventory performance. There are various definitions of what a good performing replenishment system should optimize regarding the stock level over time. Two possible perspectives on how to measure inventory performance used in this thesis are: Cost side: to reduce inventory costs, there should be as low stock in the stores as possible, therefore, the aim is to minimize the absolute inventory mean. Marketing side: there should be little variance between the stock levels each day, as the customer expects full shelves at any time. Consequently, the ordered quantity should be equal to the demand. The result of such a strategy is a constant inventory level. Effective replenishment systems should minimize inventory variance.65 Depending on the characteristics of the retailer, higher ASR systems should improve inventory performance, as long as the adequate level of ASR is not reached. The MYFOOD data includes data from three different automatic systems, which can be compared to ASR0 systems. While ASR2 and ASR3 products should benefit from the automation, an ASR2* automation is rather crude. ASR2* systems use the same reorder parameter for many products. As products might exhibit different sales patterns, the replenishment will probably be suboptimal. Overall, it is hypothesised: H15a: ASR3 products will have a better inventory performance than ASR0 products.66
65

Whether the shelves will be constantly filled or not, will also strongly depend on store operations, as employees have to refill the shelves from the backroom. However, a replenishment system that ensures that at any time the same quantity of goods is in the store is the basis for a good refilling of the shelves. The underlying assumption is that the ASR level was chosen and parameterised correctly for the product category that is examined in the sample.

66

4. Development of Models

83

H15b: ASR2 products will have a better inventory performance than ASR0 products. H15c: ASR2* systems will not perform better than manual systems. The Figures 1416 show an overview of the hypotheses.
Product characteristics

Sales variance Speed of turnover Price CU/TU Package size Shelf life

OOS OOS

H16 Dataset1

Inventory Inventory Performance Performance

Figure 14: Overview of hypotheses, product characteristics

67

Store characteristics Shrinkage SKU density Personnel / m2 Store manager experience Backroom size

H713 Dataset1

OOS OOS

Figure 15: Overview of hypotheses, store characteristics

67

It is worth noting that these three figures are not meant to be a model, even if they have some visual similarities to structural equation models. These figures are an illustration of the three different groups of hypotheses that will be tested in chapter 5. On the left side, there are the independent variables, on the right side the dependent ones. The tests of the hypotheses are carried out independently of each other. Furthermore, they are based partly on different data bases.

84

4. Development of Models

H14 Dataset1
ASR characteristics ASR system level

OOS OOS

H15 Dataset2

Inventory Inventory Performance Performance

Figure 16: Overview of hypotheses, ASR characteristics

5. Quantitative Analysis

85

5. Quantitative Analysis
To be able to test the hypotheses formulated in the previous chapter, two different data sets of the company MYFOOD, a major central European grocery retailer, are analysed.68 One part of the data stems from a physical OOS audit project that the KLOG conducted together with this retailer in summer and autumn 2004. In addition, historical sales and POS data was extracted from the retailer's ERP system to complement the findings. The quantitative analysis made on this dataset consists of three parts, analogous to the three groups of hypotheses. First, correlations between OOS rate and product characteristics are explored. Second, the influence of store characteristics on the OOS rate is examined. Finally, overall replenishment performance is tested. Table 13 gives an overview of the datasets that are used for the hypothesis testing and which will be presented in this chapter.
Hypothesis
H1H6

Data Set
Dataset1

Question Answered
Which product characteristics influence the OOS rate and inventory level? Is there a difference between manual and automatic systems? Which store characteristics influence the OOS rate? Is there a difference between manual and automatic systems? Do ASR2 and ASR3 systems perform better than ASR0 systems?

H7H13 H14H15

Dataset1 Dataset1 and 2

Table 13: Overview of the utilization of the two datasets for hypothesis testing

5.1. Sample and Methodology

5.1.1. Dataset1: Testing of Out-of-Stock Hypotheses Researchers who want to empirically test hypotheses concerning OOS incidents require first of all a reliable data sample. To obtain the availability rate of products on the shelves is, however, a difficult task, as today's IT systems only store the quantities in the store, without making a difference between the backroom and the sales area. To overcome this problem, an on-shelf availability project was launched in summer 2004 together with the European retailer MYFOOD. In the context of this project, 100 products were chosen across seven different categories and their on-shelf availability was audited manually. The choice of the products ensured that there was at least one article per each of the seven most important categories of the
68

For more information about the retailer MYFOOD, refer to chapter 6.

86

5. Quantitative Analysis

retailer.69 The top-selling products were selected in each category (e.g. bananas in the category green grocery), as their availability has a major impact on the retailer's financial results and customer satisfaction. In the sample are also products that are very important for customers as they are difficult to substitute (e.g. yeast in the category bakery products). In addition, care was taken to have all replenishment channels represented in the sample. This means that products were chosen that are distributed through regional and national DCs as well as products that are supplied directly from the manufacturers. Finally, products in promotion were taken to check for this influence as well. The final product sampling was determined by logistics and marketing directors of MYFOOD together with KLOG project members. The next step was to select the stores to be audited. Ten stores in a certain region of MYFOOD's home market were chosen taking care to represent three parameters that, according to the practitioners in the project, have most influence on current store performance: store size (large vs. small), location (rural vs. urban location) and historical performance (stores that had in the last years above and below average performance). The judgement of the stores' historical performance was carried out by marketing and logistics directors of MYFOOD, who took into account aspects such as financial performance, sales growth, happiness of customers, etc. Five students were sent during a period of two weeks in autumn 2004 to the stores to count the OOS incidents. Their task was to check twice a day the availability of the products, once in the morning and once in the evening. Therefore, this project resulted in an OOS data base of 100[products] * 12[days] * 2[measures per day] * 10[stores] = 24,000 data points. Furthermore, for 15 selected products, inventory levels were manually counted to check the accuracy of inventory records. At the end of each day, the students handed a list with all OOS incidents to the store managers. The managers' task was to check the underlying reasons for the missing products from a provided checklist.70 Store managers chose from this list the reasons for the OOS incidents for each SKU. The statements made by the store managers were sent to headquarters and were checked for plausibility by logistics and marketing representatives. The employees in the stores, as well as the store managers, were
69

70

The seven most important categories identified by MYFOOD managers are: packaged food (e.g. coffee), green grocery (e.g. tomatoes), non-food (e.g. toothpaste), dairy products (e.g. Parmesan cheese), meat and fish (e.g. minced meat), bakery products (e.g. white bread), frozen (e.g. French fries). The products in each of these categories are all distributed and ordered through the same logistics system. This checklist was analogous to the 13-items list ECR Europe has published (see Roland Berger 2003b).The ECR-list had to be enhanced by one OOS root cause: "Problems with the replacing of an SKU." This means that for some discharged SKUs the new replacing product was not yet available for ordering, therefore resulting in stock-outs.

5. Quantitative Analysis

87

informed that students would be in the stores during two weeks making a physical audit. But they were not informed what exactly was being measured. In personal face-to-face meetings, store managers were informed that the aim of the project was to measure the performance of the logistics system, not to check the performance of store staff. The managers were requested not to let the daily business be in any way influenced by the physical audit. This information strategy lead to the desired result, as the OOS rate for the first week, compared to the OOS rate in the second week, differed only by 0.1 percentage points.71 There were several cases where products were OOS for several days in a row, although the store manager could have intervened at any time. The results of other similar OOS audits undertaken before by other researchers differ radically from these results. Stlzle and Placzek (2004) had in their study a reduction of the OOS rate of 26% between the first, anonymous week and the second week, where the employees were informed. To calculate the weekly OOS rate of each product, the daily measures per product are converted into a percentage value, the same method as used by Gruen, Corsten et al. (2002). For instance, if an article was found to be absent from the shelves three times in the first week, the OOS percentage is 25%.72 For the final analysis, some products had to be discarded from the 100 original products available in Dataset1. These were products that were during any of the two weeks on promotion or out-listed. The reason for this is that promotional articles are not ordered regularly through the systems, instead the central office sets a fixed starting amount for each store (push-replenishment). As this procedure is executed manually, it would be not possible to compare the results of the ordering systems with each other, as the promotional articles change ASR system during the promotion. One article was OOS at all times in the store, as it was out-listed and the new product was not available for order. Overall, the final sample consists of 84 products in 10 stores, as can be seen in Figure 17. The unit of analysis is in most of the following each SKU in each store. Therefore, the sample population consists of n=840 units.

71

This statement is only valid for non-promotional articles. The OOS rates of products in promotion differ radically from one week to the other. 72 3 divided by 12, as during six days two daily measures were conducted.

88

5. Quantitative Analysis

Bread 3 Meat and fish 7 Dairy products 12

Frozen 3 Packaged food 27

Non food 16

Green grocery 16

Figure 17: Distribution of the 84 products in Dataset1

During the OOS project additional information on each SKU was acquired: product price, product size, shelf life and the CU/TU ratio. Another source of information was MYFOOD's ERP system, which provided for each product the delivered quantity and the daily sales. This data was used to calculate inventory performance (inventory level and its variance), the mean sales variance and the speed of turnover of each product.73 As the hypotheses H713 relate to the influence of store characteristics on the performance, some additional store related variables were collected as well. These are the yearly shrinkage rate (in percent of the turnover), the SKU density (number of SKUs per m2 sales area), the personnel density (number of employees per 100 m2 sales area), the years the store manager has been working in the particular store, and the size of the backroom in proportion to the sales area. The products in the sample are classified according to the replenishment process. The methods used are manually (ASR0: 31 products) and automatically (ASR2: 53 products). The latter group has to be split into two subgroups. For some ASR2 products, the store staff can only change the ordering parameters based on a discrete index scale from 1 to 10. This means that store managers can only decide if they want low replenishment parameters and stock levels ("1") or very high ones ("10") without specifying the exact quantity. Besides, the store manager can only do this on a group level. This means, for example, that store mangers have to change the replenishment parameters of all batteries, not only of one special battery type. As one will see, this procedure changes completely the performance of the products. Consequently, the term ASR2* level is applied to make a differentiation to the usual ASR2 systems discussed in this dissertation.

5. Quantitative Analysis

89

Table 14 summarizes the available store, product and ASR variables in use.
Variable Operationalization Variable Name

Product Variables
Sales variance Speed of turnover (=sales per day) Price CU/TU (case pack size) Product size Coefficient of variance of the daily sales Mean sales of an SKU per day: low, medium and high category Price in local currency (for all stores the same) Relative CU/TU ratio: number of consumer units per trading unit divided by mean sales Size of the product: small products (up to the volume of a closed fist) medium size products (up to the size of a one litre milk pack) and big products (for all stores the same) Shelf life Shelf life2 Product sales Product deliveries Shelf life in days Shelf life squared Daily sales per SKU and store Daily delivered quantity to the stores per SKU Shelf_life Shelf_life_squared Product_sales Product_deliveries Package_size Price CU_TU_relative Sales_coefvar Speed_of_ turnover

Store Variables
Store identification Store size Store location Shrinkage rate SKU density Personnel density Store manager experience Backroom size Number (110) to identify the stores Size category of the store (as defined by the retailer): small, medium, large Rural or urban location Yearly shrinkage in % of turnover Amount of SKUs in a store divided by sales area Number of employees in a store divided by sales area Number of years the store manager has been working in the store Ratio backroom size to sales room Store_location Shrinkage SKU_density Personnel_density Store_manager Backroom_size Store_ID Store_size

ASR Variable
Type of ASR system ASR0: manual ASR2, ASR3: automatic ASR2*: automatic, but crude parameterised ASR_level

73

In the appendix, the method for the calculation of inventory levels from the ERP-data is presented in detail.

90

5. Quantitative Analysis

Performance Variables
Inventory level Inventory range of coverage sales Inventory variance OOS Order-related OOS Inventory coefficient of variance All out-of-stocks Only OOS incidents are considered, where the underlying reason is an ordering fault.75 Weekly mean Inventory_coefvar OOS OOS_only_order
74

Inv_range_of_coverage

Mean of the stock level divided by average daily

Table 14: Overview of variables used in the analysis

76

The descriptive statistics of the product characteristics by ASR group can be seen in Table 15. Although the manual group has much higher selling quantities, the sales coefficient of variance is very similar for all three groups. The average CU/TU ratio is higher for the ASR2 group than for the ASR0 group, meaning that the latter products have more flexibility when ordering. A clear difference in shelf life exists between the groups; the manual products tend to have the shortest shelf life.

74

Instead of using absolute inventory levels, in the analysis the inventory range of coverage is computed. The inventory range of coverage is the absolute inventory level divided by the mean sales. This relative measure allows the comparison of products with different sales behaviour. As an example: if a product has a mean daily inventory level of 100, is this a lot or not? It depends; if the store sells 50 units per day, than the inventory is rather low as the amount is sufficient to meet demand for two days (100 units divided by 50 units per day equals 2 days of inventory). If the product is sold on average four times a day, than the store has a substantial amount of inventory. The inventory range of coverage is in this case 25 days (100/4). 75 The possible OOS causes that are directly related to replenishing behaviour are: order was created too late, the last order quantity was not sufficient, forecasts were not accurate, during the creation of the order an error occurred, fault with transmission to the DC, other order-related faults. In the case of MYFOOD, 60% of all OOS incidents are order-related. In the following, the wording "OOS (order-related)" is used to point out that only these kind of OOS incidents are meant. 76 Some of these variables, like price, are originally metric. To make it possible to use them in ANOVAs, they are banded before with the help of 33.33% quantiles. The result is three equal size groups that are called "low," "med" (medium) and "high." Whenever a confusion of the metric and the ordinal variable is possible, the suffix "(banded)" is used.

5. Quantitative Analysis

91

ASR Group ASR0 Sales per day [units] Sales coefficient of variance Price (local currency) Product size (category) CU/TU relative [days] Shelf life [days] ASR2 Sales per day [units] Sales coefficient of variance Price (local currency) Product size (category) CU/TU relative [days] Shelf life [days] ASR2* Sales per day [units] Sales coefficient of variance Price (local currency) Product size (category) CU/TU relative [days] Shelf life [days]

N 482 481 490 490 482 490 295 294 310 310 295 310 39 39 40 40 39 40

Minimum .25 .04 .0 1 .03 1 .50 .00 .5 1 .01 60 1.00 .28 2.1 1 .94 700

Maximum 299.75 2.83 27.5 3 40.00 700 161.92 2.50 23.0 3 20.00 700 26.67 1.12 4.9 1 17.50 700

Mean 41.16 .53 3.94 1.84 1.61 48.51 17.72 .57 3.41 1.77 2.83 460.00 6.82 .53 3.09 1.00 4.27 700.00

Std. Deviation 57.82 .26 4.71 .74 3.10 127.58 23.82 .27 4.18 .79 3.52 259.34 6.40 .18 1.09 .000 3.90 .000

Table 15: Product characteristics of Dataset1 by replenishment system

In the two week period there was in average a 3.0% rate of order-related OOS. A first evidence of the performance differences between ASR systems can be obtained by regarding the OOS rate separated by replenishment group, as depicted on Table 16. The products ordered manually had an OOS rate of 4.7% and a standard deviation of 11.7%. This contrasts to the much lower values of the ASR2 and ASR2* groups.

92

5. Quantitative Analysis

N
OOS (order-related) By ordering system ASR0 ASR2 ASR2* 310 490 40 840

Minimum
0.0% 0.0% 0.0% 0.0%

Maximum
91.7% 91.7% 29.2% 20.8%

Mean
3.0% 4.7% 0.6% 1.3%

Std. Deviation
9.4% 11.7% 3.1% 4.0%

Table 16: OOS rates (order-related)of the sample

This finding is consistent in almost all stores. As depicted on Figure 18 in all stores but in store 10, the ASR2 systems perform better than the ASR0 systems. Store 10 has overall very low OOS values, therefore, the absolute OOS rate difference between the ASR systems is minimal. Another finding depicted in Figure 18 is that the good performance of the ASR2 systems is not repeated with in the case of ASR2* products. Theses products have in all cases higher OOS rates than the ASR2 ordered products; for two stores, the ASR2* products perform even worse than the manual ordered products.

14% 12% OOS (only order) 10% 8% 6% 4% 2% 0% 1 2 3 4 5 6 7 8 9 10 Store ID ASR0 ASR2 ASR2*

Figure 18: Comparison of the replenishment systems by store

77

As with the OOS performance, a first indication on the inventory performance differences depending on the ASR system can be obtained by regarding the stock levels by replenishment group, as depicted on Table 17. The products ordered manually have larger inventory range of coverage and standard variation than the ASR2 and ASR2* groups. This difference is the more striking as the manual products
77

The stores 1-3 are the smallest stores in the sample. Their assortment do not include ASR2* ordered products, therefore there is no value for these products in the figure.

5. Quantitative Analysis

93

are mostly from fresh categories. These products should also have by far smaller inventory levels compared to the canned goods to keep the products fresh.
ASR Group
ASR0 ASR2 ASR2* Inventory range of coverage Inventory range of coverage Inventory range of coverage

N
442 287 36

Minimum
.00 .00 .33

Maximum
45.01 50.88 45.09

Mean
10.89 9.46 10.73

Std. Deviation
9.07 9.99 8.44

Table 17: Inventory range of coverage of Dataset1

5.1.2. Dataset2: Pretest/Posttest Analysis As valuable as Dataset1 is, there is a small deficit in its analysis, namely that a true pretest/posttest comparison is not possible. When assessing the performance of the ASR systems, one is comparing different product categories with each other. As seen before, the mean characteristics of the products (e.g. shelf life) are slightly different between the ASR groups. One could argue that these differences are responsible for the findings seen before, namely that higher ASR levels tend to perform better. To be able to make final conclusions, it is necessary to have a true pretest/posttest methodology as described by Sheeber, Sorensen et al. (1996) to underline the findings. This means comparing the performance of the same products in the same stores before and after the introduction of a new ordering system and having a control group where the replenishment system is not changed. This method has the highest control possibility for biasing influences. The goal of this comparison is to find out whether there is a significant benefit in implementing automatic systems or not. Therefore, a second dataset was extracted from the EPR system. As no OOS data is available from the ERP system, the dependent variables in these tests are inventory level and variance. For the pretest/posttest, data from the same 10 stores as in the OOS-audit were chosen. A period of 180 working days was defined, 15 weeks before and 15 weeks after the implementation of the new ASR system. This comparison was only possible for categories that experienced a change in the replenishment system in the last three years, namely dairy, beauty/household and non-food products. Only products that were included in the OOS study were chosen for this study, to profit from the comprehensive product information already collected in this first study. In order to increase the validity of the comparison it is important to reduce the effects of

94

5. Quantitative Analysis

non-controlled variables. One of these variables is seasonality. Therefore, when choosing the periods to compare, the same weeks in the year were taken to minimize the effect of holidays and seasonality. Otherwise, comparing the stock levels in summer with stock levels in winter could bias the result. For example, dairy products were automated in November 2004. Consequently, the weeks chosen for the test were the first 15 weeks of 2004 and 2005, respectively. Both data sets contain thus the weeks after Christmas and the Easter holidays. Another aspect taken into account is that there is normally a time period after the introduction of new ASR systems with strong irregularities in performance, as staff get used to the system. ASR systems need some time until performance stabilises. Therefore, the time period was chosen in such a way that there was at least a two months gap between ASR introduction and the performance measurement. Table 18 summarizes the periods chosen.
Number of Products
5 6 12 4 14

Category
Dairy products Control1 Beauty and household Non-food Control2

Data Period "Before"

Week 115, 2004 Week 115, 2004 Week 3044, 2003 Week 3044, 2003 Week 3044, 2003

Change of ASR System

November 2004 (ASR0 ASR3) None (ASR0) May 2004 (ASR0 ASR2) May 2004 (ASR0 ASR2*) None (ASR0)

Data Period "After"

Week 115, 2005 Week 115, 2005 Week 3044, 2004 Week 3044, 2004 Week 3044, 2004

Table 18: Dataset for the pretest/posttest

For each product the data from the inventory systems was extracted in analogy to Dataset1 supplying in this way per day and store the delivered quantities and the sales. This procedure results in 41[products] * 10[stores] * 180[days] * 2[measures] = 147,600 data points. In addition, information about whether and when a product was in promotion was provided. There was no logistical change for any of the groups during the period under examination; an unbiased comparison of the replenishment systems was thus possible.

Comparison 1: Dairy Products (ASR0 ASR3) In November 2004, a new ASR3 system was introduced for dairy products. Based on the sales of the preceding four weeks, the new system creates a forecast for the following days and dynamically calculates and places the new order. From the 100 products in Dataset1, the dairy products where chosen for further analysis as they

5. Quantitative Analysis

95

experienced a change from ASR0 to ASR3. Only products without promotional activity were included in the analysis, resulting in a sample of five products. The choice of non-promotional SKUs is of most importance, as during promotions the inventory levels sometimes increased by a factor of 20. Having these products in the analysis would distort the results enormously. Besides, promotions are still handled manually. The products studied were yoghurts, cheese and eggs. As a control group (Control1), six other randomly selected products were taken: packaged (e.g. convenience food, frozen goods) and non-packaged food (vegetables, fruit). The control group products were ordered in the same way during the entire period, namely manually. As the data set consists of measures during 180 days in ten stores, this results in 9,000 data points for the ASR3 and 10,800 data points for the ASR0 Control1 group. These figures can be considered large enough to make significant comparisons. Descriptive statistics from these two categories can be seen in Table 19.
N78
ASR0 Control1 ASR3 Dairy

Minimum
ASR0 Control1 ASR3 Dairy

Maximum
ASR0 Control1 ASR3 Dairy

Mean
ASR0 Control 1 ASR3 Dairy

Std. Deviation
ASR0 Control 1 ASR3 Dairy

Sales coefficient of variance Price Package size [category] Shelf life [days] Speed of turnover [units sold per day]

120 120 120 120 120

100 100 100 100 100

.27 1.00 1.00 2.00 1.14

.24 .60 1.00 10.00 1.80

2.35 5.60 3.00 350.00 317.44

1.02 3.87 2.00 12.00 349.60

.82 3.28 2.33 63.67 41.55

.54 2.21 1.20 11.60 53.80

.38 1.84 .75 128.62 66.38

.15 1.37 .40 .80 80.78

Table 19: Descriptive statistics of the dairy products (ASR3) and Control1 group (ASR0)

When the dairy products data is split by stores the result is that on average, before the introduction of the ASR3 system, in four out of ten stores the dairy products had a bigger inventory range of coverage than the control group. After the change, this was still the case in only one store.

Comparison 2: Beauty and Household Products (ASR0

ASR2)

Typical products in this category are beauty and hygiene articles such as handkerchiefs, toothpaste, nappies and toilet paper, as well as household articles such as washing powder. A change in the replenishment system for these 12
78

The unit of analysis is again the SKUs per store and time period. The 90 days periods for each product are condensed into one mean value. Therefore, the number of n=120 for the ASR0 group results from the computation: 6 products * 10 stores * 2 time periods (before/after) =120.

96

5. Quantitative Analysis

products was made in May 2004, changing from manual ordering (ASR0) to a minimun-maximun system (ASR2), in which a fixed order-up-to quantity exists for every product/store combination. These replenishment parameters are normally stable during the course of the year (no seasonality is taken into account) but can be changed by store managers at any time without involving the retail central office. The control group (Control2) consists of 14 randomly chosen ASR0 products including fruit and vegetables, convenience food, meat and fruit juice. Analogous to the analysis of the ASR3 group in the last section, promotional activities are a major problem when analysing inventory performance. In contrast to dairy products, promotions are in this category much more frequent. Virtually all products were on promotion for at least a week during the period. Therefore, it was not possible, as before, to reduce the analysis sample to products without any promotional activity at all. Instead, the weeks with promotional activities were ignored when calculating stock levels and variances.79 It was observed that the effect of a promotion lasts sometimes a couple of days longer than the promotion itself. Stores that had ordered manually too many SKUs still had above-average inventory some days after the end of the promotion. Therefore, the week following the promotion was taken out as well. The same procedure was applied to the control group, thus eliminating the promotional effect for these products as well. With this procedure, approximately 15% of the daily data is lost, but the gain in accuracy far outweighs this loss. The same method of dealing with promotions is also applied to the ASR2* group in the next section. As depicted in Table 20, the product groups have exactly the same sales variance. ASR2 products tend to be larger and have a lower rate of turnover than ASR0 products.

79

This is the same method as used by the ordering system of MYFOOD to compute forecasts for ASR3 articles: dates with promotional activities are ignored.

5. Quantitative Analysis

97

N
ASR0 Control2 ASR2 B&H

Minimum
ASR0 Control2 ASR2 B&H

Maximum
ASR0 Control2 ASR2 B&H

Mean
ASR0 Control2 ASR2 B&H

Std. Deviation
ASR0 Control 2 ASR2 B&H

Sales coefficient of variance Package size [category] Price Shelf life [category] Speed of turnover [units sold per day]

253 266 266 266 266

243 243 243 243 243

.16 1 1 1 0.82

.00 2 1 2 0.76

1.69 3.00 2.00 2.00 339.63

2.21 3.00 2.00 2.00 139.40

.63 1.93 1.22 1.50 30.61

.63 2.46 1.23 2.00 13.29

.26 .46 .41 .50 56.97

.32 .50 .42 .00 17.99

B&H: Beauty and household products

Table 20: Descriptive statistics of the beauty and household group (ASR2) and Control2 (ASR0)

Comparison 3: Non-Food Products (ASR0

ASR2*)

With just four articles, this is the smallest category studied. The products examined were batteries, baking foil, glue and stickers. The replenishment system for this category was automated at the same time (May 2004) as that for the beauty and household articles category. After May 2004, the non-food products have been ordered through an ASR2* system. As before, the main difference between the nonfood products and the control group (Control2) is the very low rate of turnover of the ASR2* products (cf. Table 21).
N
ASR0 Control2 ASR2* Nonfood

Minimum
ASR0 Control2 ASR2* Nonfood

Maximum
ASR0 Control2 ASR2* Nonfood

Mean
ASR0 Control 2 ASR2* Nonfood

Std. Deviation
ASR0 Control 2 ASR2* Nonfood

Sales coefficient of variance Package size [Category] Price Shelf life [category] Speed of turnover [units sold per day]

253 266 266 266 266

74 74 74 74 74

.16 1 1 1 0.82

.36 1 1 2 1.05

1.69 3 2 2 339.63

.95 2 1 2 24.82

.63 1.93 1.22 1.50 30.61

.59 1.51 1.00 2.00 6.76

.26 .46 .41 .50 56.97

.13 .50 .00 .00 6.02

Table 21: Descriptive statistics of the non-food group (ASR2*) and Control2 (ASR0)

Methodology of Statistical Analysis The datasets presented were utilized to test the three groups of hypotheses: products, store and ASR characteristics influence.80 In the following, the methodologies of the statistical analysis are presented.

80

See Figures 1416.

98

5. Quantitative Analysis

First, the relationships between product characteristics and OOS rate/inventory performance are examined. This analysis is conducted with the help of Dataset1, as there the true OOS rates for each product are available. The OOS rate of a product is the average rate of the two weeks physical audit. As this analysis aims to test the role of the replenishment systems, examining all kinds of OOS incidents would lead to distortions. A missing article on the shelf caused by a strike at the manufacturer cannot be influenced by the retailer's ASR system. Yet, with the analysed dataset, there is the unique opportunity to consider only OOS incidents caused by ordering behaviour (OOS order-related). Therefore, other OOS root causes do not distort the analysis of the hypothesis dealing with ASR performance. To establish statistical relationships between the variables correlations, linear regressions and above all ANOVAs are used.81 To prove that the effect of product variables on automatic replenishment systems is not as strong as for ASR0 systems, the interaction variable ASR x Product is considered. If this variable is significant, then the performance of the ASR system depends on the characteristics of the product. Furthermore, post hoc tests show whether there is a difference between the individual levels of product characteristics. For example, post hoc analysis tests whether the OOS rate of lowpriced products that are ordered through ASR2 systems is different from the OOS rate of medium- and high-priced products, respectively. Second, with Dataset1 and the same statistical tools it is also possible to test the influence of store characteristics on the OOS rate. In order to do this, a mean weekly OOS was calculated for each of the 10 stores, resulting thus in a sample size of n=20.82 The evidence from this rather small sample are intended to illustrate the possible existing correlations rather than being a final statistical proof. Third, the results of both these analyses are a necessary prerequisite for answering research question Q2: the benefits of ASR systems. To prove that ASR2 products perform better than ASR0, Dataset1 is examined with the help of ANOVAs. Yet, the final proof of the performance of the ASR systems is given with the help of the second dataset. As no OOS rates are available in Dataset2, the tests concentrate on
81

82

The significance level used in this thesis is 10%. As the ANOVA method is the primary method being used, the most important requirements for this methodology are discussed in the appendix in section 8.1.2. For this second analysis, the mean weekly OOS is calculated for the whole range of 100 products, therefore also including the products in promotion. This procedure was chosen because the promotions are the same for all stores and have the same duration. The performance of a store is therefore also measured by the availability to handle promotional goods. For the same reason all the OOS causes are taken into consideration, in contrast to the former hypothesis block, where only order-related OOS causes were considered. Therefore, in this second analysis reasons like faulty replenishment of shelves from the backroom are also taken into consideration.

5. Quantitative Analysis

99

the inventory levels and variance. With a controlled pretest/posttest methodology as described by Sheeber, Sorensen et al. (1996) it is possible to prove that regardless of the products chosen, higher ASR systems have significantly better results than manually ordered products. The statistical method used are so-called linear mixed models.83

5.2. Hypothesis Testing: Dataset1

5.2.1. Influence of Product Characteristics In this section, two questions are covered. First, the influence of product characteristics on OOS rates and inventory levels will be examined. If there are any substantial correlations, retailers would have an instrument to know for which products the danger of OOS is eminently high and which products tend to have high inventory levels. Second, the interaction between product characteristics and ASR system are tested. These tests make it possible to confirm whether the influence of product characteristics on ASR performance is stronger for ASR0 than for ASR2 systems.84

Test of Hypothesis H1: Sales Coefficient of Variance The variable sales coefficient of variance is banded into three categories (low, medium and high)85 and an ANOVA conducted. The results are significant for all variables but the interaction effect.

83

The basic model underlying the analysis of variance and multiple regression is the general linear model. One of the extensions of this model is the linear mixed model, the main tool for testing the hypothesis in the pretest/posttest methodology. The difference to the general linear model is that as the same subject is measured several times the assumption of independence of the error terms is relaxed. The linear mixed model allows the modelling of the values of a dependent scale variables (e.g. OOS) measured at multiple time periods based on its relationships to category (e.g. package size) and scale (e.g. price) predictors and the time at which they were measured (SPSS Inc. 2003). 84 For all tests in this section, the ASR2* products are omitted, as their sample size is very small (n=40) compared to the other two groups (nASR0=490, nASR2=310). ANOVAs results would be strongly biased by this large group size difference.
85

As stated before, the banding of the variable to create three groups is made by using three quantiles (33.33% percentiles). The result of this is three groups with nearly the same number of units in each subgroup, thus making the ANOVA analysis much more reliable.

100

5. Quantitative Analysis

Dependent variable: OOS (order-related) Source Corrected model Intercept ASR_level Sales_coefvar ASR_level x Sales_coefvar Error Total Corrected total Type III Sum of Squares 3195.433 4034.514 2526.136 463.234 185.062 47616.619 56753.472 50812.052 df 5 1 1 2 2 769 775 774 Mean Square 639.087 4034.514 2526.136 231.617 92.531 61.920 F 10.321 65.157 40.797 3.741 1.494 Sig. .000 .000 .000 .024 .225 Partial Eta Squared .063 .078 .050 .010 .004

Table 22: ASR level and sales coefficient of variance ANOVA on OOS (order-related)

On the first look, Figure 19 shows a quadratic relationship between sales variance and OOS. Yet, a Tamhane's T2 post hoc test shows that for the ASR0 products only the difference between the medium and the high sales variance group is significant (p=.071). This means that the ANOVA cannot confirm a significant difference between the low and medium ASR0 group. For this, hypothesis H1a can only be partly confirmed, i.e. higher sales variance will lead to higher OOS rates for ASR0 products.

6.00%

ASR Group
ASR0 ASR2

5.00%

Estimated Marginal Means

4.00%

3.00%

2.00%

1.00%

0.00% low med high

Sales Coefficient of Variance (Banded) Sales coefficient of variance (Banded)

Figure 19: Estimated OOS (order-related) rate by sales coefficient of variance

The interaction variable ASR_level x Sales_coefvar is not significant (p>.225). This means, that the forms of the two ASR lines, as depicted in Figure 19, do not significantly differ from each other. Put in another words, the missing significance of

5. Quantitative Analysis

101

the interaction variable means that one cannot reject the hypothesis that these two lines are parallel. Yet, a post hoc analysis of the ASR2 group shows that the individual variance levels are not statistically different from each other (all p>.16). Therefore, H1b is supported, namely that sales variance has a lesser impact on ASR2 products than on the manually ordered ones Overall, given that, in the opinion of experts, this is the most important parameter influencing replenishment systems, it is surprising that the results above are not stronger. The model estimates that the ASR0 low-variance group has a 4% and 6% OOS rate for the low and high-variance group, respectively. This difference of only two percentage points is rather low. A possible explanation is that replenishment systems cope quite well with a certain demand variance. Yet there will be a certain point at which this variance becomes so strong that performance will start to decline. This would explain that there is no significant difference between the low and medium variance group for the ASR0 system. It is worth noting that the mean sales coefficient of variance in this dataset is slightly smaller for the ASR0 group than for the ASR2 group (53% compared to 57%). This means that although the automatic systems have to deal with slightly higher variance, they perform better than their manual counterparts. For the average inventory level as the dependent variable, the results are clearer. The products of the ASR0 group that have a higher sales variance tend to have higher stocks (all post hoc tests are p<.002), supporting H1c.
Dependent variable: inventory range of coverage Source Corrected model Intercept ASR_level Sales_coefvar ASR_level x Sales_coefvar Error Total Corrected total Type III Sum of Squares 40506.130 159995.245 8865.787 22270.375 4483.293 383233.775 603923.142 423739.904 df 5 1 1 2 2 766 772 771 Mean Square 8101.226 159995.245 8865.787 11135.188 2241.647 500.305 F 16.193 319.795 17.721 22.257 4.481 Sig. .000 .000 .000 .000 .012 Partial Eta Squared .096 .295 .023 .055 .012

Table 23: ASR level and sales coefficient of variance ANOVA on inventory range of coverage

The interaction variable is significant (p=.012) and, as can be seen in Figure 20, the ASR2 replenished products do not react as strongly on the increase of sales variance (H1d). The only post hoc test that is significant for the ASR2 group is the one

102

5. Quantitative Analysis

between the low and high variance group (p=.010), while for the ASR0 group all post hoc tests show a significant difference between the variance levels (p<.022).
30.00

ASR Group
ASR0 ASR2

Estimated Marginal Means

25.00

20.00

15.00

10.00

5.00 low med high

Sales Coefficient Variance Sales coefficient of variance (Banded)

Figure 20: Estimated inventory range of coverage by sales coefficient of variance

Test of Hypothesis H2: Speed of Turnover As expected in hypothesis H2a, the best curve that fits the scatter plot of speed of turnover and OOS is a quadratic one. Products with low speed of turnover and with very high speed of turnover have the highest OOS rates. Yet, the explained variance (R2) is, with 1%, rather low. When the average sales volume per product and day is regarded, the ASR0 group has a higher mean volume than the ASR2 products (41.2 units compared to 17.7 units). This could bias an ANOVA. Therefore, to neutralise the effect of the unequal distribution of speed of turnover, this metric variable is made into a category variable. An ANOVA shows that all the effects are significant (p<.001).

5. Quantitative Analysis

103

Dependent variable: OOS (order-related) Source Corrected model Intercept Speed_of_turnover ASR_level Speed_of_turnover x ASR_level Error Total Corrected total Type III Sum of Squares 5280.535 4541.636 1245.637 2868.753 900.493 52704.280 64548.611 57984.815 df 5 1 2 1 2 771 777 776 Mean Square 1056.107 4541.636 622.818 2868.753 450.247 68.358 F 15.450 66.439 9.111 41.966 6.587 Sig. .000 .000 .000 .000 .001 Partial Eta Squared .091 .079 .023 .052 .017

Table 24: ASR level and speed of turnover ANOVA on OOS (order-related)

Post hoc tests of the ASR0 group show that the products with the highest speed of turnover (=high sales per day) have significantly lower OOS rates than the slow turning products (8.0% compared to 2.4%; p<.001). The ASR2 products only show a difference between the medium and high group (post hoc tests: p=.08), but not between the medium and low group (p>.99). As depicted in Figure 21, the absolute influence of the speed of turnover on the ASR2 products is very small, therefore confirming hypothesis H2b.

ASR Group
ASR0 8.00% ASR2

Estimated Marginal Means

6.00%

4.00%

2.00%

0.00% low med high

Speed Speed of Turnover (Banded) turnover (Banded)

Figure 21: Estimated OOS (order-related) rate by speed of turnover

The testing of hypothesis H2c confirms that the higher the speed of turnover, the lower the stock levels are. The differences are significant for the ASR0 group (all post hoc tests: p<.003). For the ASR2 products, there is no significant difference between

104

5. Quantitative Analysis

the medium and high group (p=.23). As can be seen in Figure 22, the effect of speed of turnover on the inventory level is stronger for the ASR2 group, as the line for this group is steeper. Because of this, hypothesis H2d can be regarded as confirmed.
Dependent variable: inventory range of coverage Source Corrected model Intercept Speed_of_turnover ASR_level Speed_of_turnover x ASR_level Error Total Corrected total Type III Sum of Squares 50320.073 149988.612 31316.425 11667.890 2633.599 373885.421 603923.142 424205.494 df 5 1 2 1 2 768 774 773 Mean Square 10064.015 149988.612 15658.213 11667.890 1316.799 486.830 F 20.673 308.092 32.164 23.967 2.705 Sig. .000 .000 .000 .000 .068 Partial Eta Squared .119 .286 .077 .030 .007

Table 25: ASR level and speed of turnover ANOVA on inventory range of coverage

30.00

ASR Group
ASR0 ASR2

Estimated Marginal Means

25.00

20.00

15.00

10.00

5.00 low med high

Speed of Turnover (Banded) Speed of turnover (Banded)

Figure 22: Estimated inventory range of coverage by speed of turnover

Test of Hypothesis H3: Price As hypothesised (H3a), there is a slight trend towards higher price products having higher stock outs (linear regression: p<.001), yet the explanation portion is very small (R2=.039). In an ANOVA with the ASR level as category factor, the interaction between price and ASR is significant (p=.029), as well as the entire model (p<.001).

5. Quantitative Analysis

105

A post hoc analysis shows that only for the ASR0 group is there a significant difference between the high price subgroup and the other subgroups (p<.02).
Dependent variable: OOS (order-related) Source Corrected model Intercept ASR_level Price_category ASR_level x Price_category Error Total Corrected total Type III Sum of Squares 4931.502 5169.567 3024.956 649.854 607.124 67879.609 80416.667 72811.111 Df 5 1 1 2 2 794 800 799 Mean Square 986.300 5169.567 3024.956 324.927 303.562 85.491 F 11.537 60.469 35.383 3.801 3.551 Sig. .000 .000 .000 .023 .029 Partial Eta Squared .068 .071 .043 .009 .009

Table 26: ASR level and price ANOVA on OOS (order-related)

Looking at the estimated means by this model one sees that the manual system reacts much more strongly to price level than the ASR2 system. The latter has no significant differences between any of the price groups (post hoc: all p>.46). The automatic system is not influenced by the price when ordering, in contrast to human beings (H3b).

8.00%

ASR Group
ASR0 ASR2

Estimated Marginal Means

6.00%

4.00%

2.00%

0.00% low med high

Price (Banded) Price (Banded)

Figure 23: Estimated OOS (order-related) rate by price

An ANOVA with inventory range of coverage as the dependent variable shows in a post hoc test clearly that the price levels differ from each other (p<.001). Again,

106

5. Quantitative Analysis

hypothesis H3c is confirmed, the more expensive the goods the higher the stock levels are.
Dependent variable: inventory range of coverage Source Corrected model Intercept ASR_level Price ASR_level x Price Error Total Corrected total Type III Sum of Squares 87031.665 169851.557 4334.233 57741.651 5582.179 337173.830 603923.142 424205.494 df 5 1 1 2 2 768 774 773 Mean Square 17406.333 169851.557 4334.233 28870.825 2791.089 439.028 F 39.647 386.881 9.872 65.761 6.357 Sig. .000 .000 .002 .000 .002 Partial Eta Squared .205 .335 .013 .146 .016

Table 27: ASR level and price ANOVA on inventory range of coverage

For the ASR0 products, all post hoc tests are significant (p<.001). The post hoc test of the low price and medium price of the ASR2 products is the only non-significant (p=.110). As the significant interaction variable shows, ASR2 and ASR0 products react differently to price influence. The two lines on Figure 24 cross between the low price and medium price group, confirming that manually ordered groups react more strongly on the price influence (H3d).

ASR Group
ASR0 ASR2 30.00

Estimated Marginal Means

20.00

10.00

0.00 low med high

Price (Banded)

Figure 24: Estimated inventory range of coverage by price

5. Quantitative Analysis

107

Test of Hypothesis H4: Case pack size (CU/TU) As stated before, the analyses of the case pack are conducted with the quotient case pack size divided by mean sales (CU_TU_relative). The result is that all the main effects as well as the interaction effect are significant (all p<.001)
Dependent variable: OOS (order-related) Source Corrected model Intercept ASR_level CU_TU_relative ASR_level x CU_TU_relative Error Total Corrected total Type III Sum of Squares 5412.554 4629.639 2873.172 1197.892 1136.784 52572.261 64548.611 57984.815 df 5 1 1 2 2 771 777 776 Mean Square 1082.511 4629.639 2873.172 598.946 568.392 68.187 F 15.876 67.896 42.137 8.784 8.336 Sig. .000 .000 .000 .000 .000 Partial Eta Squared .093 .081 .052 .022 .021

Table 28: ASR level and CU/TU ANOVA on OOS (order-related)

As Figure 25 depicts, the OOS rate of manual order products is strongly influenced by the relative case pack size. The bigger the ASR0 case packs, the greater the OOS rate, as hypothesised in H4a. Post hoc tests show a significant difference between the low and high group (p<.001) and the medium and high group (p=.01). In contrast, the automatically ordered products do not show a difference between any of the groups (all p>.59), confirming H4b.

10.00%

ASR Group
ASR0 ASR2

Estimated Marginal Means

8.00%

6.00%

4.00%

2.00%

0.00% low med high

CU/TU Relative (Banded) CU/TU relative

Figure 25: Estimated OOS (order-related) rate by CU/TU group

108

5. Quantitative Analysis

As expected in H4c and in H4d, bigger case packs lead indeed to higher inventories (p<.001). For both manually and automatically ordered products, the SKUs with the highest CU/TU value have at the same time the highest inventory levels. The interaction effect is not significant (p=.13); all the post hoc tests for the ASR0 and ASR2 products show significant differences between the case pack size (all p<.02). This means that both replenishment systems are equally influenced by the case pack variable.
Dependent variable: inventory range of coverage Source Corrected model Intercept ASR_level CU_TU_relative ASR_level x CU_TU_relative Error Total Corrected total Type III Sum of Squares 37077.842 137012.570 12754.988 22483.296 2061.403 387127.652 603923.142 424205.494 df 5 1 1 2 2 768 774 773 Mean Square 7415.568 137012.570 12754.988 11241.648 1030.702 504.072 F 14.711 271.811 25.304 22.302 2.045 Sig. .000 .000 .000 .000 .130 Partial Eta Squared .087 .261 .032 .055 .005

Table 29: ASR level and CU/TU on inventory range of coverage

30.00

ASR Group
ASR0 ASR2

Estimated Marginal Means

25.00

20.00

15.00

10.00

5.00 low med high

CU/TU Relative (Banded) CU/TU relative (Banded)

Figure 26: Estimated Inventory range of coverage by case pack

5. Quantitative Analysis

109

Test of Hypothesis H5: Product Size An ANOVA analysis shows that product size has a significant (p=.001) influence on the OOS rate. All other effects are significant as well.
Dependent variable: OOS (order-related) Source Corrected model Intercept ASR_level Product_size ASR_level x Package_size Error Total Corrected total Type III Sum of Squares 6496.079 4666.274 2812.039 1098.617 1453.912 66315.033 80416.667 72811.111 df 5 1 1 2 2 794 800 799 Mean Square 1299.216 4666.274 2812.039 549.308 726.956 83.520 F 15.556 55.870 33.669 6.577 8.704 Sig. .000 .000 .000 .001 .000 Partial Eta Squared .089 .066 .041 .016 .021

Table 30: ASR level and product size ANOVA on OOS (order-related)

For the ASR0 products, the differences between the smallmedium (p<.001) and smalllarge products (p=.002) are significant. Larger products have indeed more OOS incidents as hypothesised (H5a). For the ASR2 systems, on the other hand, there is no significant effect of product size on OOS (all p>.37), as predicted in H5b.

8.00%

ASR Group
ASR0 ASR2

Estimated Marginal Means

6.00%

4.00%

2.00%

0.00% small medium large

Product Size (Banded) Product Size

Figure 27: Estimated OOS (order-related) rate by product size

The ANOVA on the inventory range of coverage shows that all factors are significant (p<.04). Yet, as can be seen in Figure 28, the direction of the effect is not clear.

110

5. Quantitative Analysis

Dependent variable: inventory range of coverage Source Corrected model Intercept ASR_level Product_size ASR_level * Product_size Error Total Corrected total Type III Sum of Squares 25963.724 147353.369 2263.971 10526.892 6644.695 398241.770 603923.142 424205.494 df 5 1 1 2 2 768 774 773 Mean Square 5192.745 147353.369 2263.971 5263.446 3322.348 518.544 F 10.014 284.168 4.366 10.150 6.407 Sig. .000 .000 .037 .000 .002 Partial Eta Squared .061 .270 .006 .026 .016

Table 31: ASR level and product size ANOVA on inventory range of coverage

Post hoc analyses demonstrate for the ASR0 group significant differences between the smallmedium size group (p<.001) and mediumlarge group (p=.005). The result of this ANOVA is that medium-size products have the highest inventory levels, contrary to hypothesis (H5c). For ASR2 products, the largest products have the highest inventory on stock (smalllarge post hoc test: p=.009). The interaction effect is significant, however, as both groups are strongly influenced by the product's size, hypothesis H5d cannot be supported.

25.00

ASR Group
ASR0 ASR2

Estimated Marginal Means

20.00

15.00

10.00

small

med

big

Product Size (Banded) Product size (banded)

Figure 28: Estimated inventory range of coverage by product size

Test of Hypothesis H6: Shelf Life To test hypothesis H6a, a simple quadratic regression was made separately for every store. The result is that for all stores but one a quadratic function fits the data better

5. Quantitative Analysis

111

than a linear one. As expected, products with a low or very long shelf life have a tendency to have higher OOS than products with a medium time shelf life (R2 in mean 7.9%). The ASR2 group contains only products with a shelf life longer than 60 days, while 85% of the ASR0 products have a shelf life shorter than 12 days. Therefore, the results of the ANOVA have to be taken with caution. Figure 29 shows the estimated results of an ANOVA analysis. For ASR0 products, there is a clear difference between the low and medium group (post hoc test: p<.001), while the ASR2 group is not influenced by shelf life (p>.79).

7.00%

ASR Group
ASR0 ASR2

6.00%

Estimated Marginal Means

5.00%

4.00%

3.00%

2.00%

1.00%

0.00% low med high

Shelf life (Banded) Shelf Life (Banded)

Figure 29: Estimated OOS (order-related) rate by shelf life

Next, two independent regressions were made, splitting the group by ASR classes and making a regression analysis on Shelf_life and Shelf_life_square. The result is that for the ASR0 group the shorter the shelf life, the more OOS it had (p=.046). The quadratic variable is not significant (R2=11%). For the ASR2 group there was no significant correlation at all. Also the insertion of the variable as a covariate to an ANOVA does not give significant results. Overall, one can confirm that shelf life has an effect on OOS for the manual systems, but not for the ASR2 systems. As the distribution of shelf life is not even, hypothesis H6b can only be partially supported.

112

5. Quantitative Analysis

Dependent variable: OOS (order-related) ASR Unstandardized Group Coefficients B Std. Error ASR0 (Constant) Shelf life Shelf_life_ square (Constant) Shelf life Shelf_life_ square 5.312 -.024 2.819E-05 .541 -8.355E-05 2.697E-07 .586 .012 .000 .735 .005 .000

Standardized Coefficients Beta -.263 .179

Sig.

9.059 -1.966 1.340 .737 .007 .019 -.017 .046

.000 .050 .181 .462 .987 .963

ASR2

Table 32: Regression of shelf life and shelf life squared on OOS

Figure 30 shows that, basically, the same results as before apply to the test of the inventory levels. The ASR0 products show a significant difference between the low and medium shelf life group (p=.033). This means that hypothesis H6c cannot be supported, as products with short shelf life have higher inventory levels than the other products. For the ASR2 products, on the other hand, the variable shelf life is significant neither in an ANOVA nor in a regression (p>.79 and p>.94, respectively). This could be evidence of the smaller effect of shelf life on ASR2 systems, yet due to the known restrictions of these tests, H6d is only partially supported.

25.00

ASR Group
ASR0 ASR2

Estimated Marginal Means

22.50

20.00

17.50

15.00

12.50

10.00 low med high

Shelf life (Banded) Shelf Life (Banded)

Figure 30: Estimated inventory range of coverage by shelf life

5. Quantitative Analysis

113

5.2.2. Influence of Store Characteristics In the last section, the influence of product characteristics on OOS rate and inventory performance was shown. Yet, this alone cannot explain the results from other studies showing that there is a high variance between the stores. Therefore, the correlation between store characteristics and the OOS quote is examined in the following by conducting analyses on Dataset1. A correlation matrix of the store characteristics can be seen in Table 33.
Yearly Shrinkage in %
.654(**) .002 .654(**) .002 .167 .483 .018 .940 -.322 .166 .706(**) .001 -.581(**) .007 .351 .129 -.334 .150 -.597(**) .005 .318 .172 -.437 .054 .010 .966 -.353 .127 .132 .579 -.292 .212 .212 .369 .006 .980 -.547(*) .012 -.326 .161 .598(**) .005 -.478(*) .033 1

OOS per Week OOS per week Yearly shrinkage in % SKU density
Pearson correlation Sig. (2-tailed) Pearson correlation Sig. (2-tailed) Pearson correlation Sig. (2-tailed) Pearson correlation Sig. (2-tailed) Pearson correlation Sig. (2-tailed) Pearson correlation Sig. (2-tailed) Pearson correlation Sig. (2-tailed) 1

SKU Density
.167 .483 .351 .129 1

# Personnel per 100m2

.018 .940 -.334 .150 .010 .966 1

Store ManagerYears in Store

-.322 .166 -.597(**) .005 -.353 .127 .212 .369 1

Ratio Salesroom to Backroom

.706(**) .001 .318 .172 .132 .579 .006 .980 -.326 .161 1

Customer Satisfaction Index

-.581(**) .007 -.437 .054 -.292 .212 -.547(*) .012 .598(**) .005 -.478(*) .033 1

# Personnel per 100 m2 Store manager years in store Ratio salesroom to backroom Customer satisfaction index

** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed).

Table 33: Correlation between OOS per week and store characteristics (Dataset1)

Test of Hypotheses 7a,b: OOS Rate/Inventory Performance per Store As hypothesised before, there is indeed a strong correlation of the store variable on the OOS rate (H7a), meaning that the OOS (order-related) rate differs significantly from store to store. An ANOVA with the store as a factor is highly significant (p<.001). In all stores but one, the mean OOS rate of ASR2 products is smaller than those of the ASR0 products. The interaction effect can be interpreted on how well the ASR2 systems perform compared to ASR0 in each store. As it differs significantly (p=.005), one can assume that the store indeed has an influence on the ASR system. The biggest absolute difference exists for store number 1. This store profits most from the ASR2 implementation.

114

5. Quantitative Analysis

Dependent variable: OOS (order-related)86 Source Corrected model Intercept Store_ID ASR_level StoreID x ASR_level Error Total Corrected total Type III Sum of Squares 9249.192 5226.104 2816.143 3176.885 1928.730 63561.919 80416.667 72811.111 df 19 1 9 1 9 780 800 799 Mean Square 486.800 5226.104 312.905 3176.885 214.303 81.490 F 5.974 64.132 3.840 38.985 2.630 Sig. .000 .000 .000 .000 .005 Partial Eta Squared .127 .076 .042 .048 .029

Table 34: ASR level and Store ANOVA on OOS (order-related)

14.00%

ASR Group
ASR0 ASR2

12.00%

Estimated Marginal Means

10.00%

8.00%

6.00%

4.00%

2.00%

0.00% 1 2 3 4 5 6 7 8 9 10

Store ID

Figure 31: Estimated OOS (order-related) rate by store

The results for the inventory ranges of coverage of each store are not that clear. An ANOVA is neither significant for the factor Store_ID nor the interaction Store_ID x ASR_level. Only the ASR_level variable is significant. In all but two stores the ASR2 products have lower stocks than the products ordered manually. The range of the inventory is estimated from 8.2 to 12.2 days.

86

One obtains similar results when making the ANOVA on all OOS reasons. Yet in this case, the interaction variable ASRxStore_ID is no longer significant.

5. Quantitative Analysis

115

Dependent variable: inventory range of coverage Source Corrected model Intercept Store_ID ASR_level Store_ID x ASR_level Error Total Corrected total Type III Sum of Squares 1428.936 72004.635 744.789 359.901 435.329 63764.766 142943.244 65193.702 df 19 1 9 1 9 709 729 728 Mean Square 75.207 72004.635 82.754 359.901 48.370 89.936 F .836 800.619 .920 4.002 .538 Sig. .664 .000 .507 .046 .847 Partial Eta Squared .022 .530 .012 .006 .007

Table 35: ASR level and Store ANOVA on inventory range of coverage

13.00

ASR Group
ASR0 ASR2

12.00

Estimated Marginal Means

11.00

10.00

9.00

8.00

7.00

6.00 1 2 3 4 5 6 7 8 9 10

Store ID

Figure 32: Estimated inventory range of coverage by store

The variance of the stocks is significantly lower (p=.000) for the ASR2 group, as an ANOVA shows. The result is consistent for each store, in all the stores the ASR0 products show a smaller coefficient of variance of the stocks compared to the ASR2 group. In this case, there is no interaction effect ASR x Store. Overall, hypothesis H7b can only partially be supported, as there is a difference between the inventory variance between the stores, yet no difference between the inventory levels.

116

5. Quantitative Analysis

Dependent variable: inventory coefficient of variance Source Corrected model Intercept ASR_level Store_ID ASR_level x Store_ID Error Total Corrected total Type III Sum of Squares 1.949 375.430 .788 1.079 .243 31.154 422.729 33.102 df 19 1 1 9 9 752 772 771 Mean Square .103 375.430 .788 .120 .027 .041 F 2.476 9062.248 19.014 2.894 .651 Sig. .000 .000 .000 .002 .754 Partial Eta Squared .059 .923 .025 .033 .008

Table 36: ASR level and store ANOVA on inventory coefficient of variance

0.90

ASR Group
ASR0 ASR2

Estimated Marginal Means

0.85

0.80

0.75

0.70

0.65 1 2 3 4 5 6 7 8 9 10

Store ID

Figure 33: Estimated inventory coefficient of variance by store

Test of Hypothesis 8: OOS and Shrinkage To test this hypothesis and the following, a simple linear regression is made. As on Figure 34 depicted, the line fits rather well the data (p=.002) and explains 43% of the variance. The higher the OOS rate in a store, the higher the shrinkage. There are indeed some stores that manage to increase the availability without increasing shrinkage, H8 is thus confirmed. A repeated measure ANOVA for all points shows that the high shrinkage group has higher OOS rates (p=.08).

5. Quantitative Analysis

117

0.120

0.100

0.080

OOS per Week

0.060

0.040

0.020 R Sq Linear = 0.428

0.000

1.00

2.00

3.00

4.00

5.00

6.00

7.00

Yearly Shrinkage in %

Figure 34: Relationship between shrinkage and OOS

Test of Hypothesis 9: OOS and SKU Density There is a correlation between the store size and the weekly OOS rate, resulting in larger stores having smaller OOS rates. This correlation is at the first glance counter intuitive, as larger stores should be more difficult to handle. The explained variance is with 19.5% rather small. The examining of the variable SKU density reveals another picture. In Figure 35 the positive trend between SKU density and OOS rate is much more visible. Yet, a repeated measure ANOVA for all points shows no significant results. A possible reason for this non-significance is that a single store has more than twice the SKU density than the mean of the other stores, and nevertheless the lowest OOS rates. When this exceptional store is not regarded, the explained variance of a simple regression sky rockets to 49%. A repeated measure ANOVA without this outlier shows that the high SKU density group has significantly more OOS incidents than the other group (p<.011). The OOS difference is estimated to be 3.9 percentage points. Overall, the hypothesis H9 can be regarded as partially supported.

118

5. Quantitative Analysis

0.120

0.100

0.080

OOS per Week

0.060

0.040

0.020

0.000

5.00

7.50

10.00

12.50

15.00

17.50

SKU Density

Figure 35: Relationship of OOS and SKU density

Test of Hypothesis 10: OOS and Personnel Density To be able to compare the number of personnel between stores, the ratio employees per 100m2 is calculated. A linear regression is not significant, as stores with very few and many employees tend to have higher OOS rates. Consequently, a quadratic line fits the data rather well (R2=0.39) and underlines the hypothesis H10. Yet, a repeated measure ANOVA with the variable personnel (split in 3 categories) does not give a significant difference between the three groups.

0.120

0.100

0.080

OOS per Week

0.060

0.040

0.020 R Sq Quadratic =0.391

0.000

2.00

3.00

4.00

5.00

6.00

7.00

# Personnel per 100m2

2

Figure 36: Relationship of OOS and number of personnel per m

5. Quantitative Analysis

119

Test of Hypothesis 11: OOS and Store Manager The relation between these variables is not as high as with the other store characteristics variables. A linear regression explains only 10.4% of the variance. Higher R2 values can be obtained by a logarithmical transforming of the years scale, taking so into consideration the effect of diminishing returns. A repeated ANOVA with three groups according to the years of experience is significant (p=.053). Again, the Tamhane's T2 post hoc analysis show only a significant difference between the group with the least years in the store to the medium group, confirming the theory of diminishing returns. Overall, the hypothesis H12 can be supported, yet the effect is not as clear as with the other store variables.

0.120

0.100

0.080

OOS per Week

0.060

0.040

0.020 R Sq Linear = 0.104

0.000

0.0

2.5

5.0

7.5

10.0

12.5

Store Manager- Years in Store

Figure 37: Relationship of OOS and number years of the store manager working in the store

Test of Hypothesis 12: OOS and Backroom Size The hypothesis H12 can be fully supported, the larger the backroom in comparison to the sales room, the higher the OOS rate. A simple regression model explains 49.8% of the OOS variance, a repeated measure ANOVA analysis is significant (p=.028).

120

5. Quantitative Analysis

0.120

0.100

0.080

OOS per Week

0.060

0.040

0.020 R Sq Linear = 0.498

0.000

0.60

0.80

1.00

1.20

1.40

1.60

1.80

Ratio Salesroom to Backroom

Figure 38: Relationship of OOS and the size of the backroom

Test of Hypothesis 13: OOS and Customer Satisfaction A strong correlation between customer satisfaction87 and OOS can bee seen in Figure 39, as predicted in hypothesis H13. Stores with lower OOS rates tend to have more satisfied customers, underlying the importance of high on-shelf availability. The significance of the variable in a repeated measure ANOVA is p=.023. The explained variance is 33.7%.

87

The satisfaction of the consumers was measured through a survey of MYFOOD. The highest satisfaction score is 6.

5. Quantitative Analysis

121

0.120

0.100

0.080

OOS per Week

0.060

0.040

0.020 R Sq Linear = 0.337

0.000

4.80

5.00

5.20

5.40

5.60

5.80

6.00

Customer Satisfaction Index

Figure 39: Relationship of OOS and customer satisfaction

Test of Hypothesis H14: Influence of ASR on OOS In the descriptive statistics of Dataset1 one could see that the products ordered manually have on average almost an 8-times higher OOS rate than those ordered through the ASR2 system. The standard variation is also far smaller. In the section 5.2.2 the influence of product characteristics on OOS and inventory performance was examined. Although this difference is very impressive, there could be a small chance that this difference is the result of a random process. To test Hypothesis H14, that the ASR2 system will perform better compared to manual ordering, a one way ANOVA is made.
Dependent variable: OOS (order-related) Source Corrected model Intercept ASR_level Error Total Corrected total Type III Sum of Squares 3304.928 1394.349 3304.928 70266.170 81111.111 73571.098 df 2 1 2 837 840 839 Mean Square 1652.464 1394.349 1652.464 83.950 F 19.684 16.609 19.684 Sig. .000 .000 .000 Partial Eta Squared .045 .019 .045

Table 37: ASR level ANOVA on OOS (order-related)

As one can see, the differences between the different replenishment methods is statistically significant (p<.001). One of the assumptions of ANOVA is that the group

122

5. Quantitative Analysis

variances should be equal. Yet the Levene's test of equality of error variances rejects this assumption (p<.001). The ASR0 group has a higher variation (11.6%) than the ASR2 group (3.13%). Nevertheless, as the ANOVA method is robust to this violation Lindman (1992) does not recommend ignoring the F-test carried out above. Instead, a Tanhame's T2 post hoc test is conducted, with the assumption that the variances are not equal. The result is that the difference of 4.09% between ASR2 and ASR0 is still significant (p<.001). The difference in the OOS rates between ASR2* and ASR0 is not significant (p=.678). Both a Welch and a Brown-Forsythe test, show a significant difference between the ASR0 and ASR2 group (p<.001). An ANOVA with ASR_level as the independent variable on inventory level shows that there is indeed a significant difference between the ASR groups (p<.045). The ASR2 group has an estimated mean inventory range coverage of 9.5 days which is significantly lower than the 10.9 days of the ASR0 group. The coefficient of variance of the inventory, however, is smaller for the manual group: 69% compared with 75% of the ASR2 group (ANOVA: p=.024).
Dependent variable: inventory range of coverage Source Corrected model Intercept ASR_level Error Total Corrected total Type III Sum of Squares 359.102 72055.366 359.102 64834.601 142943.244 65193.702 df 1 1 1 727 729 728 Mean Square 359.102 72055.366 359.102 89.181 F 4.027 807.968 4.027 Sig. .045 .000 .045 Partial Eta Squared .006 .526 .006

Table 38: ASR level ANOVA on inventory range of coverage

Therefore, the hypothesis H14 can be supported: the ASR2 ordered products have on average lower OOS rates than ASR0 products. In addition, there exists strong evidence that the same holds true for inventory performance. The final proof of this is checked in the next section.

5.3. Dataset2: Pretest/Posttest

In this section, a comparison of ASR inventory performance is undertaken by analysing Dataset2. For this, linear mixed models are employed in a pretest/posttest methodology. Data from a period after the introduction of higher ASR systems is compared to data from one year before, when the ordering was still done manually.

5. Quantitative Analysis

123

5.3.1.

Dairy products: ASR3 performance (H15a)

Inventory Performance: Mean Inventory Levels In order to compare the performance of the inventory between the categories, the mean inventory range (i.e. the mean of the daily inventory divided by the sales mean) is used as before. A first glance at the mean inventory range of coverage reveals that comparing 2003 to 2004, the mean stock level of the ASR0 control group (Control1) increased by 11.8% (from 2.8 days to 3.2 days) while for the ASR3 products the inventory was reduced by 6.4% (from 2.6 to 2.5 days, see Figure 40).

3.5 Inventory range of coverage 3.0 2.5 2.0 1.5 1.0 0.5 0.0 2004 Year 2.85 2.65

3.18 2.48

ASR0 ASR3

2005

Figure 40: Mean inventory range of coverage in days of dairy products and Control1

When this reduction is spilt by store, one can see that six out of 10 stores had a higher reduction for the ASR3 group (in percentage points) than for the control group. For two stores that development of both groups was similar, and only for two stores the ASR3 group developed considerably worse than the control group.

124

5. Quantitative Analysis

Store ID

ASR Group

Inventory Mean (days) 2004 3.88 4.31 2.41 2.71 3.95 1.96 2.96 2.46 2.97 3.38 2.57 2.06 3.06 2.80 2.50 1.52 2.17 3.50 2.00 1.76

Inventory Mean (days) 2005 4.81 3.39 2.43 2.12 4.23 1.77 2.48 2.51 4.40 3.57 3.03 2.47 2.84 2.67 2.42 2.08 2.32 2.09 2.88 1.80

Reduction in %

Performance of ASR3 Group Compared to Control Group (percentage points) 45.3% 22.3% 16.8% -18.3% 43.1% -2.3% -2.8% -40.3% 47.0% 42.2%

1 ASR0 Control1 ASR3 Dairy 2 ASR0 Control1 ASR3 Dairy 3 ASR0 Control1 ASR3 Dairy 4 ASR0 Control1 ASR3 Dairy 5 ASR0 Control1 ASR3 Dairy 6 ASR0 Control1 ASR3 Dairy 7 ASR0 Control1 ASR3 Dairy 8 ASR0 Control1 ASR3 Dairy 9 ASR0 Control1 ASR3 Dairy 10 ASR0 Control1 ASR3 Dairy

-24.01% 21.34% -0.51% 21.80% -7.13% 9.66% 16.26% -2.02% -48.52% -5.46% -17.63% -19.94% 7.39% 4.55% 3.28% -36.98% -6.70% 40.29% -44.41% -2.22%

Table 39: Performance of ASR3 group compared to the Control1 (ASR0)

To be able to statistically asses the effect of ASR3 introduction on inventory performance, a linear mixed model analysis is made. The repeated factor is each product in each store, as it is measured twice: before and after the introduction of the new system. The result of such an analysis is that the introduction of the ASR3 systems did indeed have a significantly different impact for the dairy products compared to the control group (p=.038). While both groups start 2004 at a similar level (between 2.6 and 2.8 days inventory range of coverage), the ASR3 group has the next year only in mean 2.45 days of inventory, while the control group has increased its stock to 3.18 days. This confirms the first part of hypothesis H15a.
Dependent variable: inventory range of coverage Source Intercept Date ASR_level Date x ASR_level Numerator df 1 1 1 1 Denominator df 108 108 108 108 F 345.739 .287 2.450 4.413 Sig. .000 .593 .120 .038

Table 40: Repeated ANOVA on inventory range of coverage, ASR3 group

5. Quantitative Analysis

125

Adding the factors Store and Store x ASR_level to the model results in a slightly better model fit. The Schwarz's Bayesian Criterion (BIC) is now 757.9 compared to a BIC of 802.7 in the former model. The store factor is significant on a 10% level (p=.073), the interaction between the store and the ASR group is not. The basic result of this model is the same, the ASR3 group improves its mean stock level while the control group worsens (p=.038). Higher degrees of interaction (Store x ASR_Group x Date) are not significant.

Inventory Performance: Coefficient of Variance A linear mixed model with the coefficient of variance of the stock as the dependent variable supports the theory of the good performance of the ASR3 systems. In analogy to the section before, the Date x ASR group interaction is significant (p=.025). Again the ASR3 ordered products reduce their inventory coefficient of variance (from 132% to 122%), while the control group worsens (from 160% to 191%), this confirming the second part of hypothesis H15a. This model results in a BIC value of 573.0.
Dependent variable: Inventory coefficient of variance Source Intercept Date ASR_Group Date x ASR_Group Numerator df 1 1 1 1 Denominator df 108 108 108 108 F 330.969 1.252 8.522 5.200 Sig. .000 .266 .004 .025

Table 41: Repeated ANOVA on the coefficient of variance of the stock level, ASR3 group

Inventory coefficient of variance

2.5 2.0 1.60 1.5 1.0 0.5 0.0 2004

Year

1.91 1.32 ASR0 1.22 ASR3

2005

Figure 41: Means of the repeated ANOVA on the coefficient of variance of the stock level, ASR3 group and Control1

126

5. Quantitative Analysis

5.3.2.

Non-food Products: ASR2 and ASR2* Performance (H15b, c)

Inventory Performance: Mean Inventory Levels The mean inventory range of coverage by year and ASR system can be seen in Figure 42. Just by looking at this graph one can see that the ASR2 group showed a very marked improvement, reducing their mean inventory range of coverage by 4.1 days. The ASR0 group (Control2) reduced their stock as well, but only by 0.9 days. Contrary to this trend, the ASR2* inventory levels increased on average by 1.1 days after the introduction of automatic ordering.

Inventory range of coverage

12 10 8 6 4 2 0 2003
Year

9.83 7.01 5.52 6.09 5.74 ASR0 ASR2 ASR2*

6.63

2004

Figure 42: Mean inventory range of coverage in days of non-food products and Control2

Two separated linear mixed models analyses were made to compare the ASR2 and the ASR2* products with the control group. The interaction between the variable Date and the ASR2 group is significant (p=.005), confirming H15b.
Dependent variable: inv_range_of_coverage Source Intercept Date ASR_level Date x ASR_level Numerator df 1 1 1 1 Denominator df 289.421 226.741 289.421 226.741 F 166.127 22.540 .846 7.981 Sig. .000 .000 .358 .005

Table 42: Repeated ANOVA on mean inventory range of coverage, ASR2 group

The estimated inventory means show that Control2 products reduced its stock from 6.8 days in 2003 to 6.0 in 2004 (-12%). But the reduction of the control group from 9.0 days to 5.9 days (-34%) is much more dramatic.

5. Quantitative Analysis

127

10 Inventory range of coverage 9 8 7 6 5 4 3 2 1 0 2003 6.84

9.01

6.05

5.86

ASR0 ASR2

2004 Year

Figure 43: Estimated means of the repeated ANOVA on the inventory range of coverage, ASR2 group

For the ASR2*, another picture results from the analysis. Here again, there is a significant interaction between the Date and the replenishment type (p=.011). Yet the difference is that while the control group managed to reduce its stocks, the ASR2* showed an increase. The automatic replenishment system did not work as expected on these products, confirming H15c.

8 Inventory range of coverage 7 6 5 4 3 2 1 0 2003 Year 2004 ASR0 ASR2* 6.85 6.03 5.41 6.63

Figure 44: Estimated means of the repeated ANOVA on the inventory range of coverage, ASR2* group

Inventory Performance: Coefficient of Variance The second performance indicator of stock is the coefficient of variance. As depicted in Figure 45, the ASR2 group reduced its variance from 65% to 57%, while the control group experienced and increase from 47% to 59%. In this case, the ASR2* group also managed to reduce the coefficient of variance by 5.6 percentage points.

128

5. Quantitative Analysis

Inventory range of coverage

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0.47

0.65 0.56

0.59 0.57 0.51 ASR0 ASR2 ASR2*

2003
Year

2004

Figure 45: Mean inventory coefficient of variance in days of ASR2, ASR2* group and Control2

Although the better performance of the ASR2 group seems clear, a linear mixed model fails to prove a significance for the ASR x Date variable (p=.335). Because of this, hypothesis H15b can only be partially supported. The same holds true for the difference between ASR2* and the control group: the differences depicted between the means could have happened by pure chance.

5.4. Quantitative ResearchConclusions

The quantitative analysis carried out provided answers to two research questions. On the one hand, the influences of store and product characteristics on replenishment performance could be identified (research question Q3). On the other hand, the statistical analysis clearly showed that MYFOOD's replenishment performance has benefited from the implementation of ASR2 and ASR3 systems (research question Q2). In the following, a summary of the results is given. The hypothesised direction of influences of product characteristics on the performance of manual replenishment systems could be shown for most of the product characteristics. The analysis demonstrates that products with higher prices, larger case packs and with larger volume size will tend to be more often OOS when ordered manually. The influence of sales variance, shelf life and speed of turnover could be only partially proven. Furthermore, it was confirmed that products that have higher sales variances, that are slow movers, that are expensive or that have large case packs will tend to have higher inventory levels in the store. A major result of these tests was to show the different reactions of automatic and manual systems on

5. Quantitative Analysis

129

different product characteristics. The data reveals for the ASR2 products a much weaker influence of the product characteristics on the OOS and inventory performance. None of the product variables had an influence on the OOS rate. The inventory levels of ASR2 ordered products are only influenced by the case pack and product size. This is a strong argument in favour of the use of automatic systems, as their performance is much more robust.
Dependent Variables Independent Variables Sales variance Speed of turnover Price CU/TU Product size Out-of-Stock Inventory Level
part. sup. sup. part. sup. sup. sup. sup. sup. sup. sup. sup. part. sup. part. sup.

+ 0 +-+ 0 + 0 + 0 + 0 +-+ 0

(H1a, ASR0) (H1b, ASR2) (H2a, ASR0) (H2b, ASR2) (H3a, ASR0) (H3b, ASR2) (H4a, ASR0) (H4b, ASR2) (H5a, ASR0) (H5b, ASR2) (H6a, ASR0) (H6b, ASR2)

+ 0 0 + 0 + + 0 + 0

(H1c, ASR0) (H1d, ASR2) (H2c, ASR0) (H2d, ASR2) (H3c ASR0) (H3d, ASR2) (H4c, ASR0) (H4d, ASR2) (H5c ASR0) (H5d, ASR2) (H6c, ASR0) (H6d, ASR2)

sup. sup. sup. sup. sup. sup. sup. sup. not sup. not sup. not sup. part. sup.

Shelf life

"+" Positive effect; "-" Negative effect; "0" Smaller effect of ASR2 system compared to manual system; "+ - +" quadratic relationship; sup.: supported; part. sup.: partially supported

Table 43: Results overview: product characteristics hypotheses

Five out of eight hypotheses concerning store characteristics could be fully supported, three only partially. There is a marked difference in OOS performance between the individual MYFOOD stores, despite their equal logistical context. Some stores manage to have low OOS rates and at the same time low shrinkage rates. Small backrooms might be one of the success factors for this performance, as stressed by supporters of the lean management theory. Another success factor is surely the store manager. To achieve good results experienced managers are necessary that have been working in the store for a certain time. There exists strong evidence for the fact that the right number of staff and the right number of SKUs per m2 are also important for replenishment performance, although their influence in the present data could not be fully proven.

130

5. Quantitative Analysis

(H7a) (H7b) (H8) (H9) (H10) (H11) (H12) (H13)

OOS rate differs from store to store Inventory performance differs from store to store Reduction of OOS and shrinkage is at the same time possible The more SKU per m the more OOS Too many and too few staff increases OOS rate More experienced store managers have fewer OOSs Larger backrooms lead to more OOSs Stores with low OOS rates have higher customer satisfaction
2

Supported Partially supported Supported Partially supported Partially supported Supported Supported Supported

Table 44: Results: store characteristics hypotheses

Finally, the statistical analysis showed the expected results for the performance of automatic systems. Products that are ordered through ASR3 systems show significantly better inventory performance than ASR0 products; the automation reduced the stocks of the ASR2 products by 6.4% while the control group increased by 11.8%. The same holds true for ASR2 systems, after automation these products had a 3.1 days lower inventory range of coverage, i.e. inventory could be reduced by 42% while the control group only reduced its inventory by 13%. The average ASR0 product has, with a 4.7% OOS rate, an almost 8-times higher OOS rate than ASR2 products. Yet, in this second case the improvement in inventory variance is not significant. Last but not least, the quantitative analysis showed that a crude automation will not lead to the desired results. The badly parameterised ASR2* systems have a worse replenishment performance than manual systems.
(H14) ASR2 systems have lower OOS rates than ASR0 systems Supported Supported Partially supported Supported

(H15a) Products changed from ASR0 to ASR3 systems improve their inventory performance (H15b) Products changed from ASR0 to ASR2 reduce inventory stocks, however not inventory variance (H15c) Products changed from ASR0 to ASR2* systems do not improve their inventory performance Table 45: Results: ASR hypotheses

A summary of the results of the hypotheses can be seen in Table 46.

5. Quantitative Analysis

131

Findings
Product Characteristics
H1a H1b H1c H1d H2a H2b H2c H2d H3a H3b H3c H3d H4a H4b H4c H4d H5a H5b H5c H5d H6a H6b H6c H6d ASR0: Products with high sales variance have more OOS incidents than products with medium sales variance ASR2: No influence of sales variance on OOS rate ASR0: Products with higher sales variance have higher inventory levels ASR2: Smaller influence of sales variance on inventory level ASR0: Slow-moving products have more OOS incidents ASR2: Smaller influence of speed of turnover on OOS rate ASR0: Slow turning units tend to have higher stock levels than fast movers ASR2: Smaller influence of speed of turnover variance on inventory level ASR0: Expensive products have more OOS incidents ASR2: No influence of price on OOS rate ASR0: Expensive products have higher inventory levels ASR2: Smaller influence of price on inventory level ASR0: Products with large CU/TU ratio have higher OOS rates ASR2: No influence of CU/TU ratio on OOS rate ASR0: Products with high CU/TU ratio have higher inventory levels ASR2: Same influence of CU/TU on inventory rate as ASR0 products ASR0: Large products have more OOS incidents ASR2: Product size has no influence on OOS ASR0: Medium-sized products have the lowest inventory level ASR2: Large products have higher inventory levels ASR0: Very short and very long shelf life leads to more OOS incidents ASR2: Smaller influence of shelf life on OOS, but data missing ASR0: Short shelf life products have higher stocks than medium shelf life products ASR2: No difference between medium and high shelf life products, but data missing

Method/ Significance
ANOVA: p<.001 ANOVA post hoc ANOVA: p<.002 ANOVA post hoc ANOVA: p<.001 ANOVA post hoc ANOVA: p<.003 ANOVA post hoc ANOVA: p<.02 ANOVA post hoc ANOVA: p<.001 ANOVA post hoc ANOVA: p<.001 ANOVA post hoc ANOVA: p<.001 ANOVA post hoc ANOVA: p<.002 ANOVA post hoc ANOVA: p<.005 ANOVA post hoc Regression ANOVA post hoc ANOVA: p=.033 ANOVA post hoc

Hypothesis Supported?
Part. supported Supported Supported Supported Part. supported Supported Supported Supported Supported Supported Supported Supported Supported Supported Supported Supported Supported Supported Not supported Not supported Part. supported Part. supported Not supported Part. supported

Store Characteristics
H7a H7b H8 H9 H10 H11 H12 H13 The OOS level and inventory performance varies strongly from store to store The inventory levels vary strongly from store to store, the inventory variance not Stores with high OOS rate have high shrinkage as well Stores with high SKU density have higher OOS rates Stores with too little or too many employees per m have more OOS incidents Stores with new store managers tend to have more OOS incidents Stores with large backrooms tend to have more OOS incidents Stores with low OOS rates tend to have more content customers
2

ANOVA: p<.001 ANOVA Rep ANOVA: p=.08 Regression Regression Rep ANOVA: p=.053 Rep ANOVA: p=.028 Rep ANOVA: p=.023

Supported Part. supported Supported Part. supported Part supported Supported Supported Supported

ASR Characteristics
H14 H15a H15b H15c ASR2 systems have lower OOS rates than ASR0 systems Products changed from ASR0 to ASR3 systems improve their inventory performance Products changed from ASR0 to ASR2 reduce inventory stocks, however not inventory variance Products changed from ASR0 to ASR2* systems worsen their inventory performance Rep ANOVA: ANOVA with repeated measures; Part. supported: Hypothesis is partially supported Rep ANOVA p=.005 Rep ANOVA p=.011 Part. supported Supported ANOVA: p<.001 Rep ANOVA p<.05 Supported Supported

Table 46: Overview of the results of the hypotheses tested

132

6. Field Research and Managerial Implications

6. Field Research and Managerial Implications

As seen in the quantitative analysis, higher ASR systems can significantly improve retailers' inventory performance, reduce OOS incidents, and show more constant results. Yet it was also found that badly or crudely parameterised systems will not deliver the desired results. As stated by the contingency theory, the context of the company is decisive in determining the success of an automatic replenishment system. Therefore, in order to be able to give recommendations to practitioners it is necessary to examine in detail how retailers successfully introduced ASR systems and in which fields they can still improve their operations. In this chapter, the case studies of four retailers are depicted. For each company, a description of their replenishment processes, the organizational changes and personnel issues linked to the ASR introduction are highlighted as well as the outcome of ASR system's introduction. Next, recommendations for management are derived from these findings. First, advice is given to retailers on which ASR systems they should implement, depending on product characteristics (research question Q4). Second, a cost-benefit analysis assesses the effect of automation. Third, practitioners learn the necessary organizational and technical prerequisites for each ASR level; the answer to research question Q5.

6.1. Research Sample

6.1.1. Company Selection As Eisenhardt (1989) states, the selection of cases is a most important aspect when carrying out field research. Given the limited number of cases which can usually be studied, she does not promote a random selection (statistical sampling). Instead, Eisenhardt recommends the use of cases with polar types, filling each of the categories that are relevant for the research.88 The cases are selected to highlight theoretical issues and to refute or challenge the theory being tested (Bansal and Roth 2000). The main companies chosen for this research project are presented in Table 47.

88

This process is called theoretical sampling by Denzin (1989).

6. Field Research and Managerial Implications

133

Retailer name and type A: MYFOOD Grocery retailer B: GROCO Grocery retailer C: NORTHY Grocery retailer D: DRUCKS Over-the-counter chemist retailer

ASR Systems in use

ASR0: fruit and vegetables ASR2: packaged food and non-food ASR3: dairy products ASR0: most of items ASR4: goal 80% of goods ASR2: non-food ASR3: most packaged goods ASR3: fruit and vegetables ASR4: most products

Experience with system89

Several years 2 years 3/4 of a year Several years Pilots Several years 3 years 3/4 of a year Several years

Table 47: Selected companies for the field research

The choice of these four retailers is illustrated in the following. The first retailer chosen was MYFOOD. The quantitative analysis in the former chapter is based on data from this grocery retailer. The richest insights into the replenishment systems comes from this case study. The next retailer chosen was GROCO. This retailer is piloting ASR4 systems. These systems are rarely found in European grocery retailing; for this reason the studying of GROCO provided a most interesting case study. The third grocery retailer, NORTHY, is known as one of the most innovative retailers in northern Europe. No other grocery retailer known by the author has acquired more experience with ASR3 systems in the last few years. The company has already adapted its logistics, operational processes and organization to best support the automatic systems. The last case study examines the over-the-counter chemist retailer DRUCKS. Although it is not a grocery retailer, it is the company with the most experience with ASR4 systems known to the author. DRUCKS has a very broad product range, selling not only pharmaceuticals, beauty and health articles but also packaged food and near-food products (2003: 11,000 SKUs). Furthermore, the distribution channels and delivery frequencies to DRUCKS's stores are very similar to the other three cases. For these reasons the company is increasingly comparable to a regular grocery retailer. The only differences one can find between DRUCKS and other grocery retailers is the absence of very large stores (>1500 m2) and the absence of fresh goods such as green grocery. The similarities support the trend

89

As of August 2005.

134

6. Field Research and Managerial Implications

found by Eichen and Eichen (2004) that the assortments of chemists and grocery retailers are increasingly overlapping. To sum it up, these companies were chosen because they represent the most advanced ASR levels one can find at this point in time in retail, namely ASR3 and ASR4. At the same time, the chosen retailers are still also ordering some products through lower-sophistication levels, making thus valuable comparisons possible.

6.1.2. Market Characteristics The four retailers come from two European regions: the Nordic countries and central continental Europe. In terms of market share, they are among the top three in their respective core markets. The four companies are under the influence of the same market developments that can be observed in many western European countries. First, the markets the sample companies are in are oligopolies; a consequence of several consolidations of smaller retailers in the last few years. Second, as stated before, a trend observed is the increasing overlap of the assortments between the classic grocery retailers and chemists. The result is fiercer competition between both retail types. The third trend observed is the power situation between the suppliers and retailers. According to Eichen and Eichen (2004), manufacturers tend to dictate prices and quantities, their power resulting from their branded products and category share. However, retailers are becoming more powerful through their increasing market share.90 The big players in the market in particular are profiting from the consolidations that are taking place (Aalto-Setl 2001).

6.1.3. Supply Chain Structure The retailers MYFOOD and GROCO have both basically the same supply chain structure. They distribute slow movers and special goods (such as frozen products, wine) through central warehouses and regional distribution centres. Direct store delivery (DSD) quantities are negligible. This is different at NORTHY. DSD is used by NORTHY for many categories such as brewery products (e.g. beer and soft drinks), fresh bread, dairy products and ice creams. Measured in turnover, DSD accounts for about 30% of NORTHY's distribution volume. The remaining products are transported through a main

90

This power shift opinion is shared by many researchers and practitioners, even if its quantitative verification is somewhat difficult. See Ailawadi, Borin, et al. (1995).

6. Field Research and Managerial Implications

135

distribution centre as well as smaller regional distribution centres and cross-docking terminals. DRUCKS changed its distribution system in the 1980's from a DSD philosophy towards a system with 3 distribution centres: one central and two regional. With this new system came a shift of power and responsibility away from the stores. Nowadays there is centralized control of the logistics, which has led to major improvements in the lead time and full truck loads due to bundled transport possibilities. The national DC handles 80% of the SKUs. The national DC takes care of all small products where the case packs have to be opened before being transported to the stores. Smros, Angerer et al. (Smros, Angerer et al. 2004a) found in their analysis of the major European retailers two trends that can also be observed in the present sample: the increasing employment of cross-docking and a move towards centralized storage of slow moving goods (Smros, Angerer et al. 2004a). Retailers employ crossdocking for a part of their products, typically for fast moving goods such as fruit and vegetables or meat products. One-third of the companies examined by Smros and Angerer (2004) have centralized storage of slow movers, just like MYFOOD and GROCO in this sample. Overall, as depicted in Figure 46, all retailers have a two-level distribution structure. All four retailers have national DCs and heavily use cross-docking and regional distribution centres. This eases the comparison between the ASR replenishment systems, as all retailers have a comparable logistics context.

Supplier Supplier

Supplier Supplier

Supplier Supplier

National DC National DC

National DC National DC

National DC National DC

Regional DCs/ CD Regional DCs/ CD

Regional DCs/ CD Regional DCs/ CD

Regional DCs/ CD Regional DCs/ CD

Retail Stores Retail Stores

Retail Stores Retail Stores

Retail Stores Retail Stores

MYFOOD/ GROCO

NORTHY

DRUCKS

Figure 46: Supply chain structure of sample

136

6. Field Research and Managerial Implications

6.1.4. Chains and Store Formats Along with supermarkets, GROCO's business also includes a variety of non-food stores such as chemists and non-food specialities. The non-food sector accounts for a major part of its turnover. In this thesis, only the classic grocery stores are studied that have a size between 200 and 8,000 m2 and that carry on average 10,000 SKUs. The retailer MYFOOD has a similar format channel concept ranging from small stores with 200 m2 up to hypermarkets with up to 9,000 m2. The MYFOOD group also owns some non-food chains. The total assortment has a size of over 300,000 SKUs. NORTHY has three different retail chains. The first type of stores consists of small neighbourhood shops. The stores are between 150 and 300 m2 in size and have assortments of about 2,000 products. The next group consists of supermarkets that are typically relatively small, although they come in a wide range of sizes, from 400 to 1500 m2. The stores carry between 3,000 and 5,000 SKUs. The third type of stores consists of hypermarkets located in cities with slightly trendier products than the other supermarkets. Their size is between 4,000 and 10,000 m2 and have assortments of about 18,00025,000 products. DRUCKS has several hundred store outlets in its home market in central Europe. A typical store has about 500 m2 and carries 10,000 SKUs. The biggest stores have around 1000 m2. In the last few years, an expansion away from its home market towards other European countries can be observed.

6.1.5. Delivery Frequency and Order-to-Deliver Lead Times MYFOOD and GROCO have a very frequent delivery to their stores. Both stores have daily delivery for most of their products, i.e. six times per week. Smaller stores on rural areas are have slightly less frequent deliveries, that is, between three and five times a week. For some very fresh products such as bread, MYFOOD even has several deliveries per day. The order-to-deliver time, depending on the product category, is typically around 12 days. NORTHY delivers non-foods, packaged goods, fresh goods and green grocery on average 5 times a week. Frozen goods are brought to the stores 2 or 3 times a week on average; larger stores receive replenishment 5 times per week. DRUCKS has a delivery frequency of between 6 times per week to only once a week. The replenishment frequency depends mainly on the store turnover and in second

6. Field Research and Managerial Implications

137

place on the store size. If two stores have the same turnover but one is much smaller than the other, then the smaller store will have a more frequent replenishment because the storage space is smaller. The entire assortment can be ordered at any of the defined delivery days.
MYFOOD MYFOOD Delivery Delivery Frequency Frequency
6 times a week 6 times a week Rural stores:3- 5 Rural stores:3- 5 times times Bread: twice a Bread: twice a day day

GROCO GROCO
6 times a week 6 times a week Rural areas and Rural areas and small stores: 3- 5 small stores: 3- 5 times times

NORTHY NORTHY
Large stores: 5 Large stores: 5 times a week times a week Frozen goods: Frozen goods: 2- 5 times a week 2- 5 times a week

DRUCKS DRUCKS
Large stores: 6 Large stores: 6 times a week times a week Very small stores: Very small stores: Once a week Once a week

Figure 47: Delivery frequency of sample

Overall, one can see that all retailers have a very frequent delivery frequency. In addition, all the stores experience a high delivery service quality from their distribution centres. For the bulk of products there are typically daily deliveries, combined with a very fast response time. The order-to-delivery lead times between distribution centres and stores are usually between 24 and 36 hours. This result is analogous to the findings in the study by Smros, Angerer et al. (2004a). In this European study the average delivery lead time of fresh goods to the stores is 24 hours, with a range of 14 to 32 hours. For other goods, the median value is about 34 hours, with a range of about 15 to 48 hours. The companies studied have an orderto-delivery lead time from manufacturers to retailers' distribution centres of typically less than 48 hours. Many fresh and green grocery products can be even replenished within 24 hours. Therefore, the authors conclude that the supply chain of typical European groceries has improved significantly over the last few years and has now reached a very high level, when compared, for example, to American retailers (c.f. Roland Berger 2003a). The four companies in this field research are no exception to these European trends.

6.2. Replenishment Processes

This section takes a closer look at the four modules a replenishment system consists of: inventory visibility, forecasting, replenishment logic and ordering restrictions.91

91

For more details on these modules see section 4.1 with the descriptive model.

138

6. Field Research and Managerial Implications

6.2.1. Inventory Visibility The importance of having extremely accurate inventory records increases with the automation level in use. As the companies in this thesis' sample order products with the help high ASR levels, the need for high accuracy grows accordingly. In the following, the different processes that have an influence on inventory visibility are described. To compute the stores' inventory levels, the input and output product streams have to be considered. The basic idea behind an inventory holding system is relatively simple. The inventory in the store equals the initial stocks plus the amount delivered by suppliers minus the cash-out, returns (e.g. wrongly delivered goods) and write offs (e.g. spoilage of fresh goods). It is clear that it is not possible with such a system to know how many products are on the shelf and how many in the backroom, as shelf replenishment is not monitored separately. However, it makes a great difference for shoppers whether the products are on the shelf or in the backroom. Figure 48 depicts the elements that have an influence on inventory.
Shrinkage Store Inventory Replenishment from the DC/supplier Backroom Backroom inventory inventory Returns Shelf replenishment Shelf Shelf inventory inventory Cash out/ write off

Figure 48: Inventory storage places and product flow processes

Inventory record books can be either administered on paper or electronically. The four companies in the sample have for most of their products ASR24 systems, therefore the use of electronic inventory management systems is mandatory. The method for transferring information into the system is the barcode at trading unit and SKU level. For instance, a single toothpaste SKU arrives at MYFOOD stores in case packs of 24; each toothpaste tube and each case pack has a barcode on it. As stated by a NORTHY logistics manager, such barcodes are "the backbone of any automatic replenishment system." Non-food products are an enormous challenge regarding one-to-one SKU identification, as they tend have a huge variety. An example are apparel products; the

6. Field Research and Managerial Implications

139

same T-shirt cut can have several colours and sizes. But also in the food sector there are some products that are difficult to track. The GROCO ASR project leaders reported that they had considered having products from the served cheese counter also in the inventory recording system. This idea was abandoned rather quickly as the only way of having an exact inventory record was to weigh each product in the evening. As the larger cheese counters might carry more than a hundred different products, the weighing of every SKU would have been very time consuming and thus very expensive. Consequently, GROCO still orders these products manually.92 For an automatic system to work properly, one-to-one identification is necessary. This means that every SKU needs its own barcode number. The technology is not the problem, there are in theory enough digits available for every SKU. Yet several of the small suppliers often do not comply with EAN-standards93, thus making individual identification very difficult. In all four cases, the electronic inventory record system is an integrated part of the ERP system of the company. Therefore, in the following no difference will be made between the inventory record system and the ERPthey are regarded as a single entity. To transfer the inventory level-information to the system, the goods are scanned at different points in the supply chain; every transaction in the supply chain is registered. Ideally, this scanning should happen after leaving and before entering any of the storage places along the SC. However, there are in practice several deviations from this procedure. For example, there is no retailer known to the author that tracks the movement from the backroom to the store shelves.94 In the following, the two main flow processes in the store are considered, the replenishment from the supplier and the check-out.

Store Replenishment The input stream of a store consists of the goods that are delivered from a supplier. Here lies the first possible source of inaccuracy, as a GROCO distribution centre director points out: "In our system the sales automatically initiate the replenishment process. If any picking errors are made at the distribution centre, then this has a dramatic influence on inventory records, distorting the entire replenishment system."

92

Most Spanish retailers have found another solution for this problem: they abandoned any served counters, fresh products such as meat, cheese, etc. are nowadays only offered in pre-packed portions. EAN: European Article Numbering. The Metro shop of the future is piloting RFID labels and smart shelves that promise to track the inventory on the shelf.

93 94

140

6. Field Research and Managerial Implications

Companies such as MYFOOD and GROCO have stopped internal checks on deliveries from the DCs to the stores at the store gates. These retailers trust in the quality of their own pickers and scanners and feel that the increase in accuracy is not worth the cost of the physical audits. Consequently, an error that happens during picking at the DC can no longer be detected at store level. Deliveries that arrive directly from the supplier are still normally checked, as any delivery difference is reclaimed and refunded. Some consultants (e.g. Arthur Andersen LLP 1997) highlight the possibility of skipping checks for suppliers with good track records, as the operations costs are higher than the refunds. A logistics director from MYFOOD stressed the importance of having synchronized the DC's deliveries to the processes at store level. In his opinion, it is not sufficient to deliver the required quantities at the right day at the store. The exact hour and timing of the deliveries is just as important. If all the deliveries from the different suppliers arrive at the same time at the store, chaos would ensue. Store staff are not able to cope with such high delivery quantities at one time. Therefore, MYFOOD, just like many other retailers, has defined exact delivery windows for all the transports to the store.

Check-out MYFOOD has identified the importance of high POS data accuracy for ASR performance. To improve the personnel's awareness of accurate scanning, the store employees are trained intensively. A former problem was that there existed a generic scanning number that was used at the check-out every time a product could not be scanned. This source of inaccuracy was abolished, as it was used too frequently by staff. To be able to introduce new ASR systems, it was necessary for MYFOOD to install new tills. The problem was not missing functionality in the old machines but rather that the stores did not have the same standardized systems, making a centralized system connecting all tills impossible. The average scanning accuracy is currently 99.4%. GROCO's plan to improve check-out accuracy relies on a new function of their tills. By examining individual scanning processes the retailer found out that in some stores the repeated scanning button was unusually frequently used. The problem was that some cashiers thought they could save the time of the shoppers by scanning fast, but wrong. Instead of scanning five different yoghurts flavoured separately, only one yoghurt was scanned and then the "repeat" key was pressed five times. The result was that all these products had a wrong inventory record. To inhibit this behaviour in

6. Field Research and Managerial Implications

141

the future, GROCO's new check-out system allows configuring the tills individually for each store and cashier so that this key can be blocked. NORTHY, on the other hand, already had modern equipment at the time of the ASR introduction. The challenge was in this case not the usage of a new technology but moreover to educate the people on the importance of using the tills right. As NORTHY receives many direct deliveries from smaller companies it frequently faces the problem of suppliers using the wrong barcodes. Consequently, suppliers had to be instructed on the role of exact and reliable barcodes for the functioning of the entire system. At DRUCKS, modern check-out systems and PCs at store level were already in use. In theory, no process had to be changed. Nevertheless, as the accuracy of the inventory records had to increase after the introduction of the ASR4 system, a special emphasis was placed on check-out accuracy.

Shelf Inventory Visibility As stated before, the most important storage space for goods, the shelf, is at the same time the most difficult place to control the inventory levels, at least for stores with a backroom. The quantitative results in the former chapter showed the counter-productive effect that excess backroom space can have (see Figure 38). Some retailers have abolished these backrooms, for example Manor in Switzerland or Globus in Germany. For these retailers, the inventory on the shelf equals the total inventory in the store. The four retailers in the sample still use backrooms, just like most European grocery retailers. A precondition to measure the availability rate in a store is to know the quantity of goods on the shelf. From the research by Smros, Angerer et al. (2004a) it is known that most retailers do not have systems or processes to measure whether there is at least one SKU left on the shelf. Of the 12 major European retailers in the sample, only six companies have a process for regular shelf availability checks; four companies check availability on a daily level, two regularly but less frequently. In addition, three companies have commissioned one-time studies to examine their availability in selected product categories. Three out of 12 companies do not measure on-shelf availability at all. However, on-shelf availability and approaches for measuring it are considered extremely important development areas by the majority of the interviewees. The four retailers in this thesis sample have had physical audits of OOS in pilots, yet no regular control has been implemented. Nevertheless, all four companies expressed the need for a tool that is able to measure OOS situations automatically from the ERP systems.

142

6. Field Research and Managerial Implications

Processes to Increase Inventory Records Accuracy There exist many factors that can make inventory records inaccurate. Among others there are shrinkage, returns to the suppliers that are not recorded, and wrong scanning at the check-out. The four companies in the sample check overall inventory accuracy only during annual stock takes, as demanded by commercial law. Consequently, it is not surprising that the interviewees had only a rough guess of their inventory accuracy; in their opinion it lies between 9799%. This figure is much better than that reported in the study by Raman, DeHoratius et al. (2001b). And indeed, a physical audit of inventory records accuracy of MYFOOD revealed a high inaccuracy in the results. For many products, the quantities shown in the system were different from the real physical inventory. An example of a product (baking powder) that was manually counted during 14 days is depicted in Figure 49. As one can see, the real inventory deviates constantly by 13 SKUs.
Units in 40 store 35
30 25 20 15 10 5 0 1 2 3 4 5 6 7 8 9 10 11 12
0 5 10

Inventory range of coverage in days

Inventory according to the system

Real inventory in store

Figure 49: Comparison of inventory records and real inventory in one store

Even by counting manually the products in the store, an exact measure of inventory accuracy is rather difficult for two reasons. First, problems occur for inventory levels that are not updated in real time. In the case of MYFOOD, the inventories are only calculated once a day, when the regional sales system (i.e. how many items were sold at the store?) communicates with the nationwide logistics system (i.e. how many items were delivered to the store?). Only during this small time window, in the early morning before the stores are replenished, the physical inventory in the store can be compared with the records in the IT system. The second major problem is simply the amount of goods in a store. For instance, MYFOOD has in an average sized store around 200 packages of a nut cookies SKU. When there are that many products of a single SKU on the shelf, the manual counting of the products is not only time-consuming, but also error-prone. Already Millet (1994) reports on how in some cases manual counting can actually worsen the accuracy of the inventory records.

6. Field Research and Managerial Implications

143

Therefore, several retailers such as DRUCKS have introduced a counting process that is exception-based: whenever an inventory record is zero or even negative, employees receive an alert and make a counting check. A drawback of this procedure is that some inventory inaccuracies will remain undetected. For example, if the records show positive stock levels even though there are no goods in the store, no alert will be given. As the ASR believes that there is still inventory left it will place no new order. As no items can be sold, the quantity in the inventory system will never decrease until by chance someone detects this error. A possible solution for this dilemma is an intelligent alert system that also takes into consideration sales data (cf. Corsten and Gruen 2004). The system compares the sales of historical articles with previous data. If an article is not sold for a certain period at all, the employees receive an alert. GROCO's plan to increase inventory accuracy is called "evening-walks." At the end of each day, employees are supposed to walk along the aisles and check for "0123 products." This means that the store staff look for products that have three or fewer items left on the shelf. With this method out-of-stocks can be detected as well as products which are in danger of becoming OOS in near future. For the products with such a low inventory level the employees compare the quantity on the shelves with the books and check, whether any orders have been placed. The idea behind this procedure is that it is unproblematic to check the inventory records for products that are almost sold out, as there are very few items to be counted. Another advantage of this system is that in the long run, all SKUs are checked. Another advantage of this procedure is that SKUs with a high risk of becoming OOS will also be checked as well more frequently, as they often have near zero stock levels. Not only are products with very low inventory levels checked, but the employees are also asked to pay attention to products with unusually large stocks. NORTHY has introduced a similar method, but not as structured as GROCO. The store employees are required to check "unusual" amounts of SKUs, without specifying what this means. This checking is not undertaken on a regular basis; employees are asked to always keep an eye open for such anomalies. DRUCKS's store managers are free to decide whether an inventory record accuracy process is necessary. Central office produces a handbook which contains recommendations on how to increase inventory accuracy. The processes described in this operations-handbook are just suggestions and not at all compulsory. For example, it is recommended to manually count the products that have an unusually low sales ratio compared to the inventory records. The employees are responsible for

144

6. Field Research and Managerial Implications

finding such anomalies independently, as there are no alerts provided automatically by the IT systems. As described above, the only exceptions are products with a zero or negative inventory; these products are specially marked by the IT systems. Inventory accuracy plays an important role for DRUCKS, as without any exception all SKUs have electronic inventory records. DRUCKS demands from every one of its suppliers an electronic dispatch advice; this is something many retailers are still fighting for, among them NORTHY. All the retailers interviewed stated that with the introduction of higher ASR systems training sessions were organized to increase the awareness of check-out personnel. The problem is to find the right balance between accurate checking and customer service. Many customers do not understand why it is necessary to scan every article even though they have the same price. Therefore, cashiers are also instructed on how to behave when facing complaints from customers. The high turnover of cashiers and the large number of part-time employees makes it difficult and costly to train every one appropriately. To check and correct inventory records, all the four retailers have mobile PDAs in use. This allows employees to check directly on the shelf or in the backroom the accuracy of the inventory records, at least roughly if there is no real time updating of the inventory records. Store employees are also allowed to change the records. On the one hand, this is in the opinion of most interviewees the best method for increasing the quality of the inventory records. On the other hand, some interviewees see the danger that this ability might be misused to influence the automatic replenishment system. One MYFOOD logistician reported a case where the correction of the inventory records was misused to influence the automatic ordering. By making the system believe that there were too few articles in the store, the employee steered indirectly the delivery quantities. By now, MYFOOD has restricted the ability of staff to change records. For its new ASR4 ordered products GROCO has plans to introduce a central control mechanism to prevent misuse of this feature. Employees will still be allowed to change the inventories. But this information is reported to headquarters, where central planners watch for anomalies. Incentives Related to Inventory Accuracy The wages of MYFOOD's store managers consist of a basic salary and some additional bonuses. One of the bonuses depends on inventory accuracy. The problem with this incentive is that inventory accuracy is only checked manually once or twice a year due to the high level of effort required. OOS rates and inventory levels are not part of the incentive system.

6. Field Research and Managerial Implications

145

GROCO also offers a salary incentive for the accuracy of inventory records at store level. The maximal difference must be no higher than 0.3%. Yet, this incentive system takes into account exceptions for special products with high shrinkage rates such as razor blades or condoms. In a three-month test, GROCO found that only 32% of the razor blades that were delivered to the stores were sold regularly through the check-out. The rest (68%) were stolen. Razor blades are therefore nowadays only sold at the checkout. The biggest change concerning incentives came with the introduction of the new ERP system. This new technology increased inventory visibility and makes it now possible to check whether a store shows unusually high inventory levels. In such a case, the central office would warn the store manager. At DC level, there are monetary incentives for the accuracy of picking. According to a GROCO logistics director, the new incentive scheme has shown excellent results. NORTHY has adopted another approach. For this retailer it is of the utmost interest to have a standardized assortment on its stores across the country. Consequently, it has introduced negative incentives (i.e. punishments) for stores that do not stick to the standardized planograms. Finally, DRUCKS has decided not to implement any incentive system for their store managers. The logistics director of DRUCKS stated that employees already have a great interest in high inventory accuracy, as the stores' replenishment performance is dependent on accurate values. This manager stated: "Accurate records are like the foundations of a building. If the foundations are skewed, the entire building will be skewed as well." In his opinion, the low inaccuracy level of well below 3% (measured in the annual stocktake) shows that no incentive system is required.

Inventory Records Accuracy and Performance In spite of all the difficulties mentioned in this section about record accuracy, all companies in the sample report very good results from their ASR implementation for most of their goods. There are two possible explanations for this. On the one hand, the ASR systems in use could be robust against inventory accuracy. The safety stocks could have been set very generously, making the systems robust. This is an open question that has to be solved in further research. Another explanation could be that if at any time inaccuracy reaches a high level, the processes implemented to detect this such as the "0123 walk" seem to work pretty well. An exception is NORTHY, as this retailer is rather unhappy with the performance of its ASR3 system

146

6. Field Research and Managerial Implications

with perishable goods (products with a shelf life shorter than 10 days). Further research is required to determine if poor inventory accuracy is the underlying problem for this.

6.2.2. Forecasts and Replenishment Logic At MYFOOD, an ASR3 system with integrated forecasts has been implemented for dairy products. The forecasts on store level are rather simple, on every weekday a forecast for the next day is calculated taking into consideration the sales of the last four same weekdays (e.g. on a Monday the last four Tuesdays are taken into account). Any days with special events (such as Christmas) or with promotions are skipped. If the current inventory plus the goods on the way are lower than the estimated required quantity for the next day, an order is placed. To be on the safe side, the quantity necessary for the second day after the current day is also ordered (e.g. on a Monday an order is placed to have enough for the Tuesday and the Wednesday). For packaged products and non-food SKUs, a simple ASR2 system is in use. This means concretely that there are minimum and maximum levels for each product's inventory that are not to be under or over scored. These parameters are constant throughout the year. The shop is responsible for the setting and updating of the ASR2 system parameters. An exception is non-food articles. As there are so many SKUs in this category, it would be too time-consuming for each store to set the parameters on item level. Therefore, the store has a restriction on changing these parameters. Employees can only change the quantity based on a discrete scale ranging from 110. Furthermore, they can only do this on a group level, meaning, e.g., that they have to change the replenishment parameters for all coffee machines, not only of one special SKU.95 An logistics manager interviewed expressed the opinion that it would make no sense to have a forecasting system for packaged and non-food goods, as there is a daily delivery service to the stores. The parameters for non-food and packaged food are set in such a way that the stores have an inventory range of coverage of approximately three days. This rule is not applicable to slow-moving articles as they tend to have much longer inventory ranges of coverage. The exact inventory level depends on the size of the trading unit and the shelf space. MYFOOD tries to set the parameters in such a way that an entire case pack can be put directly on the shelf to avoid inventory in the backroom. Still, there are situations where this procedure does not work due to the formula that is used for the replenishment. The parameters used
95

This system was called ASR2* in chapter 5, where the performance of such systems was examined.

6. Field Research and Managerial Implications

147

for the replenishment are based on average sales. To make this clear, imagine an SKU that sells on average 10 units per day. If the shelf holds 35 units and the replenishment takes only one day the parameter will be set in such a way that as soon as the inventory drops below 25 units, a case pack with 20 SKUs will be ordered. The system calculates that within the lead time of one day, 10 articles will be sold making hence a complete refilling possible. Obviously, there can be problems for articles with high sales variance. An OOS incident is not very probable, as the safety stock is quite large. The main problem occurs if sales are unusually low. In this case, there are additional handling costs of having a half-empty case pack stored in the backroom. The author has seen in many stores "creative" ways of stacking the last units left of an open case pack on the shelf to avoid any additional reshelving effort. Another unsolved problem is small stores, where the number of SKUs in a case pack will be always too large for the shelves. The order suggestions made by the systems are controlled via PDA (personal digital assistant). Employees look at the automatically-generated orders and compare them to the existing inventory. In the beginning, store employees were able to alter any generated proposition. As too many suggestions were altered, the correction possibilities were later restricted by the central office. The logistics director of MYFOOD defends this procedure by saying: "Either you do automatic replenishment the right way, or you better do nothing at all." One of the last influences MYFOOD's store managers have on ordering is that they are able to change the reorder point levels. Yet there are thoughts in the central office of abolishing this right as well. Goods in promotion and new product introductions are not handled through this automatic system; the orders are placed manually. For fresh products there used to be a very simple automatic ordering system for many years. Employees received sheets with tables where they wrote with a pen the new orders. The tables were not empty, the last order was already pre-printed on them. This way, if an employee forgot to place the new order, the last quantity was ordered automatically. The retailer GROCO still has at the moment an ASR0 system in use for most of its products. This means in practice that employees walk along the aisles and decide subjectively the replenishing quantity by looking at the inventory levels. The only support is a PDA to check the electronic inventory records. Yet, in the past months, an ASR4 system has been in pilot in two stores for two categories. In the near future it is planned to introduce this system throughout the country for many categories. The new ASR4 system will use the forecasting engine of a popular ERP package to

148

6. Field Research and Managerial Implications

compute the forecasts. For each article in each individual store, a separate forecast is calculated based on the sales data of the last 2 years. Any promotions in that time are ignored in order to avoid biasing the result. For the employees at the stores, as well as for the employees in the central office, the forecast is a "black box" as they do not know the exact algorithms behind it. The retailer's central office feeds the system with so-called events to make the forecast more precise. The variables that are used at the moment to improve the causal forecasts are calendar events, opening times, holidays, regional festivals and actions of the competition. The interviewed ASR project member said that he is thinking of implementing weather data and substitution-products in the future as well. Every day, the forecast is calculated for each day of the following four weeks. For new products without sales history, comparable products are taken as the basis of the calculation. The results of the forecasting are two groups of order suggestions. In the one group there are the SKUs where the software has a high confidence on the results. In the second group, the software has some doubts about its own calculations and requests the manager to check the results for plausibility. Approximately 5% of all computed orders are classified as insecure. The system classifies products into the insecure cluster if the inventory levels are negative, the order size calculated has an unusual size, or if it is an article that is required to be counted in the formerly described "evening walks." The software has up to 650 predefined reasons that can raise an alert. This is a theoretical number, only a small proportion of them are practically implemented. The store manager sees only about 510 exceptions, the central headquarters slightly more. For example, if an article does not sell at all in a day, this is reported directly to the replenishment group in the central office. The planners there have to decide if any action is required and thus whether the store manager has to be informed. Independently from this alert system, the store manager still has complete rights to change any order suggested by the ASR system. NORTHY's ASR3 system has a very simple forecasting algorithm. It uses exponential smoothing to create the forecast for the next weekday. The exponential parameter is the same for all products and with a value of 0.2 rather low. Such a low value means that more weight is put on the sales history than on recent sales. Therefore, unusual sales influence the forecast only slightly. NORTHY has experimented with several other values and found this one to be the most appropriate. The retailer's actual forecasting method does not take seasonality effects, promotions or holidays into account. It is hence one-dimensional in the classification terms of section 4.1.4. The parameters are fixed for the whole year. Nevertheless, sometimes the central planners change the parameters for entire product groups when they expect high demand changes. The overall accuracy of the forecasting is rather low, but as stated

6. Field Research and Managerial Implications

149

by a logistics manager, a better forecast is not necessary for their company. The interviewee shares the opinion of MYFOOD managers: when the replenishment supply chain is fast and responsive then high forecast accuracy is not necessary. A view that is shared by many retailers and researchers (cf. Alicke 2003; Smros, Angerer et al. 2004b). Some of the articles are ordered through ASR2 systems and thus no forecasts at all are required. Six parameters are necessary to make the ASR3 system work: sales data, lead time, safety parameters, shelf space, minimum shelf space and CU/TU ratio.96 The sales data is collected for each store, product and day. Based on these numbers, the inventory position is calculated and together with the lead time the schedule for the next order is determined. The system requires manual information about how big the safety stocks should be. For each SKU there exists one safety stock parameter (in days) that is the same across all stores. Yet, sometimes the safety parameters are determined at category level. The next required parameter for the logic is shelf space. Shelf space is not determined for each store, moreover there exist six predefined assortment types depending on store size. For each one of these assortment types the shelf space is strictly defined as well as a minimum shelf filling parameter. The last parameter is a percentage value of the total shelf space and defines a minimum shelf stock amount that should never be reached; otherwise, the shelf would look too empty and thus would deter sales. Finally, the CU/TU ratio plays an important role; this parameter is mostly set by the supplier. It is worth noticing that if any of these parameters changes by more than 30% (e.g. when the case pack becomes bigger or an SKU sells less than 70% of the average quantity) an alert appears and the order has to be checked manually. For special occasions with increased demand (e.g. Christmas or sports events), employees are requested to place additional manual orders. DRUCKS's ASR4 systems use a forecasting engine that was programmed by the same company as the one from GROCO. As described before, the forecast is made for each product in each store on the basis of the sales of the last two years. DRUCKS has an every-day-low-price policy, therefore promotions do not have to be omitted from this database. Overall, there are about 50 parameters and event types that can be implemented in theory to improve the forecast. Replenishment is calculated taking into account a minimum level at each store (for aesthetic reasons), a desired availability level, seasonal effects, public holidays, new stores opened by the competition and price changes. There are six different classes of forecasting algorithms, while the slow moving products receive a very basic forecast, the fast moving products receive the most attention and have forecasts where the entire
96

See also section 4.1 for a description of the elements of an ASR system.

150

6. Field Research and Managerial Implications

bandwidth of events is taken into account. The classification of an SKU to a certain forecast type is fixed for all the stores, even if in a store a product is sold more often than in another one. The forecast for a single product requires only milliseconds and is created by a PC in the store. This way, the forecast for all the 10,000 SKU's in a typical outlet is finished within a few minutes. Overall, the forecasting quality of the system is described by the DRUCKS logistics director as fantastic, yet there are no KPIs in place assessing its quality. The interviewees could not name any regular product where a manual forecast would be better than the ones calculated by PC. Only for articles that are listed for the very first time, a manual ordering and forecasting is made until the system has enough data to make a reliable forecast. As with GROCO, the system automatically detects products where the forecast is insecure and asks the personnel for a manual counter-check. The system displays the reason for the alert, such as a zero stock level. About 510% of all articles are counter-checked every day. The final replenishment decision is still in the responsibility of the store manager. He or she can change any order at any time. It is thus possible to fill an article with zeros to delist a product without the approval of the central office. Smros, Angerer et al. (2004a) found in their research that the majority of the observed European grocery retailers have sophisticated forecasting capabilities only at central warehouse level. At this level, forecasts are made from the aggregated stores demand in order to place the orders to the supplier. Yet, the trend found in the four retailers examined in this thesis is that more and more complex forecasts are also made at store level. Retailers decide from category to category whether it is worth investing in better forecasts or not. It is not always clear whether the decision that lead to the choice of a certain ASR level was based purely on objective reasoning. For some categories the choice is obvious. For example, for very fresh products (e.g. bakery products) no forecasts are computed as there are not even electronic inventory records available. For other categories the "importance" of the product for the success of the retailer is the driver behind the choice of a higher ASR category. MYFOOD, for example, has introduced ASR3 systems for dairy products and in the future will do so also for green grocery, as this category is in their opinion crucial to customer satisfaction. The GROCO ASR project leader estimates that it makes economic sense for a typical retailer to have around 80% of all products replenished with complex ASR4 systems, while 20% of the goods are better served with manual replenishment. In section 6.5.1 a systematic look at the appropriateness of a product category for a certain ASR level is made.

6. Field Research and Managerial Implications

151

6.2.3. Order Restrictions The most common restriction taken into account is the case pack size restriction (CU/TU ratio). In all the observed companies the system orders only multiples of the TU size. A second restriction has been implemented by MYFOOD and takes into consideration the shelf size. As described before, the ASR system is parameterised to place the order in such way that the case pack units fit completely into the shelf. The aim of this restriction is to reduce the number of items stored in the backroom. No other restrictions were observed. NORTHY, for example, does not have a minimum order quantity, ensuring that minimal orders are not placed with its suppliers. In the past, this has led to goods being delivered in near-empty trucks, making deliveries very costly. But this is an exceptional case; there are no plans to implement additional restrictions as it would be too costly compared to the potential benefits. In addition, NORTHY has set aside the idea of having maximum order caps, as all stores have a backroom in which to store excess deliveries. Neither GROCO nor DRUCKS see the necessity of introducing any additional restrictions at store level. An optimization of the logistics, e.g. by introducing order caps to ensure full truck loads, is not an option at the moment.

6.3. Organizational Changes and Personnel Issues

The implementation of ASR systems is often a small part of major ERP system introduction. The change of ERP systems has been in turn seen as an opportunity to centralize and homogenize IT systems. In some cases, the entire logistics system had to be changed. The first part of this section describes how the retailers in the sample changed their organization. As one will see in the following, the organization in the four cases has become more centralized. In addition, there are insights as to how the ramp up phase of the new systems was realized. In the second part changes in the processes that influence the way employees work are illustrated. The retailers approach differently the question of how great an influence employees should have on automatically generated orders. 6.3.1. Structural Changes and Setup MYFOOD and GROCO were some years ago very decentralized companies. Their regional organization units had extensive decision freedoms. The main problem with this organization structure was that several IT systems existed without compatibility among them. It was therefore not possible to interchange information and cooperate with other sub-organizations. The result was the loss of economies of scale and scope. MYFOOD hence decided to start a centralization programme. As the basic

152

6. Field Research and Managerial Implications

independence of the subdivisions should remain, the programme was limited in scope. Nevertheless, a fundamental change took place in both logistics and marketing. The first step was to introduce centralized logistics for slow moving goods. In order to do so, a new central DC was built and a central logistics services group was created that works as a separate profit centre. Delivery frequency for some of the small stores was reduced, as automatic systems need fewer deliveries to achieve the same service level compared to manual ordering. Next, the logistics as well as the IT infrastructure were adapted. In order to maintain good service for its customersthe regional unitsthe central unit had to harmonise the regional IT infrastructures. Another goal of the new central unit was to maintain the master data of all SKUs up to date. Within this major project that changed IT and distribution strategy, the implementation of an advanced replenishment system at DC level was also realized. As a second step, MYFOOD began to introduce advanced ASR systems at store level as well. This top-down procedure is in contrast to the strategy of GROCO, which is implementing its ASR systems at store level before changing and centralizing its national distribution systems. For the manager in charge interviewed at MYFOOD, this is the logical way to introduce an ASR system, but because of the strong influence of the individual regional sub-divisions, it was not possible for MYFOOD work in this way. GROCO will have an even more radical change to its structure. The interviewed ASR project members stressed that the introduction of the new IT systems together with centralization will have an effect on almost everyone in the company. Eighty per cent of employees will experience a substantial change in the way they work and, for this reason, they will require training. In regard to the ASR system itself, a new organization has been created. There will be ASR-dedicated planners in charge at national level as well as regional planners. The exact distribution of tasks between the two organization types still has to be decided. The ASR-planners will be responsible for the setting and updating the parameters in the stores. They also will check the computer-generated orders made by store employees and will decide whether they are appropriate. Further, they will decide which products can be ordered automatically and which not. They will be responsible for most of the exceptions alerts created by the systems. For NORTHY's ASR2 and ASR3 systems, the organizational change was not so significant. In the stores no change in the number of employees was required. And at head office, there was only a shift of responsibilities between existing positions, no new function had to be created or personnel lost. The initial setup of the millions of parameters was a rather simple project, as most of the values could be taken from

6. Field Research and Managerial Implications

153

the existing systems, minimizing manual input. NORTHY had previously been organized on strict centralized lines; most of the information was available at the central office. Unlike MYFOOD and GROCO, the planograms were strictly followed by the stores. Before the nationwide implementation of ASR3, different parameters were tested in a number of stores. Also, comparable to the case of MYFOOD, the product categories were automated one by one. The DRUCKS case was similar, in that the implementation of ASR systems required only minor changes because a central organizational structure was already in place. The logistics did not experience any change, and the distribution channels and replenishment frequency remained the same. Neither was it necessary to make any changes to the existing technology. The same handhelds that were used before at the shelf to place orders are now used to check inventory levels. The same holds true for the tills and the utilized barcodes. The only process that was adapted was the one regarding inventory records accuracy. A new KPI was introduced to check the number of automatically generated orders that are accepted by the store managers: the value is around 95%. The very first time the ASR system was tested, a pilot store was chosen and a single category was automated. Then, each category was implemented one after the other until all the goods (around 10,000 SKUs) were ordered via the new system. Now, a new store gets a standard parameter set. This parameter set is copied from a store with a similar size and sales volume. The effort for this setup is very low, after 23 hours the parameters for the new store are implemented. For every new store a set of performance goals is elaborated. In the following months, store turnover and other KPIs are compared to the planned figures; if there is a strong misfit the replenishment parameters are adapted.

6.3.2. Personnel and Change Management Maybe the most radical change resulting from the implementation of higher ASR systems is the change in organization and the role of employees. The change in human agency is most visible in the standard manual ordering process. The main intellectual task of an employee whose store orders manually is to take into account many information sources at the same time. When determining the new replenishment quantity at the shelves, employees have to take into consideration the inventory levels in the backroom, how many goods are on the way to the shop, and they have to estimate the demand for the next days. Furthermore, before creating the final order the store staff has to take into account the lead time and other restrictions such as shelf space or minimum order quantities. This task requires a lot of experience so that only employees working for a long time for a certain retailer

154

6. Field Research and Managerial Implications

achieve good replenishment results.97 With the new ASR systems, all this information is stored in a computer, and the main task is now to keep the records accurate. Therefore, day-to-day operations change radically, and new intellectual challenges arise. For an IT director at MYFOOD, the change from an ASR0 system to an ASR3 system for fresh food products meant a radical change in the philosophy of the store staff: "We have changed our philosophy from an 'order culture' to an 'inventory management culture'. In seminars we tell our store staff how important it now is to keep the inventory records accurate. We tell them 20 times a day until they cannot hear it any more." The new ASR system relieves employees of the ordering procedure. During training, people are told to invest this freed-up time and energy in updating the inventory records. For ASR2 systems, there is still the intellectual task of determining the right replenishment parameters. The right setting of the order point decides whether it is possible to place the complete TU on the shelf and has thus a major influence on the effort required for reshelving. Another task that still requires intellectual effort is the case of exceptional demand and upcoming holidays. MYFOOD's employees are able to alter the suggestions for these cases, meaning that skilled labour is still required. The same holds true for promotions and new product introductions. The first main change for the employees connected with the installation of ASR systems was the introduction of computers and PDAs. A couple of years ago, all the orders were made on pieces of paper. Nowadays the use of computers is compulsory for every employee, as the inventory records have to be updated electronically, the automatic orders checked, and exceptional orders have to be placed. Consequently, MYFOOD's employees first receive basic training on the use of computers. They learn how to use a mouse and the basic functions of the operating system (Windows). The next step is to get acquainted with the ERP programme as this software is used for all the orders. This first steps are often the hardest, as many employees have been working for many years without using any IT at all. According to one IT manager, one of the goals of personnel training is to promote personal responsibility. Consequently, voluntary e-learning programmes were offered to employees. As an additional incentive, the employees receive a voucher worth 500 euro for an adult education centre where additional computer courses can be taken.
97

See the analysis of the influence of the time a store manager has been in a store on the OOS rate on Figure 37.

6. Field Research and Managerial Implications

155

The company does not control what courses the vouchers are spent on as this would not be part of its company philosophy, and employees could spend the vouchers on language courses instead. The interviewed MYFOOD ASR expert estimates that about five years are required for the introduction of such systems into all stores. The employees need around 23 years until their mind-set has completely adapted to the new philosophy. The ASR project leader at GROCO stresses the importance of the human factor for the success of an ASR project: "80% of our [ASR4] project concerns the human factor." As for the previous retailer, the human factor is considered to be one of the main reasons for the length of such projects. According to the responsible ASR manager, it takes up to 5 years until such a system can be used nationwide. The store staff needs a significant part of this time to get used to new system. It was stressed by the interviewees that the role of store employees as entrepreneurs should remain in the new system. Therefore, store managers will be allowed to override orders made by the ASR4 system. Yet, new KPIs will be introduced to check the amount of altered suggestions. Another education activity will be targeted at the employees in the new central replenishment organization. They require special training, as these jobs are created from the scratch. Finally, the interviewed managers also stressed the importance of superior management support, as such a fundamental change creates much uproar in the organization. The managers interviewed at NORTHY's confirmed the dramatic changes in the role of store staff described before. Almost every process they were used to changed due to the implementation of the ASR systems. The loss of the right to order and the focus on inventory records accuracy brought the biggest changes. With this retailer, store employees face the strongest limitation of the influence on the automatic generated order in the sample. It is not possible for NORTHY's store employees to cancel an order. The suggestion cannot be reduced, but it is possible to order additional quantities. These strong limitations were partially the consequence of misuse of the systems. There were cases where the staff tried to influence the system indirectly. For example, inventory level records on the computer were changed to show incredibly high levels to make the system stop deliveries of undesired products. However, this problem was an exception; one year after the introduction of ASR systems the new processes and rules became standard. This can be partly attributed to the comprehensive educational courses the store managers received. They were afterwards responsible for teaching their employees the new processes. At head office, less than six logisticians are responsible for the

156

6. Field Research and Managerial Implications

ASR systems in all stores. Such a small number of planners is only possible because the product managers are in charge of updating the parameters of the ASR systems in each store. This way, the central planners can concentrate on exceptions. The DRUCKS interviewee confirmed the radical change for the staff, as suddenly inventory management became the major task. But the introduction was not as problematic as with the other retailers because DRUCKS' personnel already had experience with computers in the stores. Neither the check-out process nor goods reception had to be adapted. The ordering procedure is basically the same, as the same PDAs still are used to confirm the order suggestions made by the ASR system. The regional managers were convinced of the automatic replenishment by showing them the successful results of the store pilots. The main argument in favour of ASR systems was that the time savings realized by the store personnel could be used to increase service to the customer. One of the major strategies of DRUCKS was the communication strategy. The words "automatic ordering" were never used, instead the management spoke of "PC-supported ordering." The word "automatic" was avoided to never give the employees the feeling that their work could be replaced by a computer. Overall, all the interviewees agree on the importance of dispelling the main concerns of employees. The first concern is that the ASR system will result in logistical chaos. This fear can be overcome by having successful pilots. Second, many employees expressed the fear that jobs could be cut due to the automation. The four companies stress that nobody lost his job due to the automation. This is not always the case, as the example of the retailer Globus in Germany shows. After an ASR introduction at DC level this retailer made several employees occupied full time with ordering tasks redundant (Shalla 2005). And third, employees face with automation the loss of a fundamental part of their current work. Being relieved of the ordering task is not always seen as something positive, as one logistics director points up: "It is more fun to buy then to sell."98 The satisfaction of employees can dramatically suffer with the introduction of ASR systems. One controversial topic is what intervention rights the personnel should have as regards the IT-calculated replenishment orders. DRUCKS and GROCO stress the importance of still having entrepreneurs in the stores making their own decisions, and so store managers are able to ignore the computer-generated

98

Source: Dr. Michael Krings, director Logistics Douglas GmbH, 24.6.2005.

6. Field Research and Managerial Implications

157

suggestions completely. This is the opposite of MYFOOD's and NORTHY's policy in that both retailers have limited the influence of their store employees.

6.4. ASR Performance

One of the most interesting questions for practitioners is the question about the performance of the systems. As seen in the quantitative analysis, it is at the same time one of the most difficult questions to answer. The introduction of ASR systems often goes hand-in-hand with a change of the distribution channels or the introduction of a new ERP system. Consequently, very few retailers are able to quantify the overall effect. In this section, first a hypothesis derived from the contingency theory concerning performance measurement is verified. Later, the benefits of ASR implementations on stock levels, OOS incidents and other effects for the four retailers will be depicted.

6.4.1. Performance Measurement One of the assumptions made in the contingency theory section 3.4.2 was that the more sophisticated the system, the more performance measure variables would be in use by the company. This statement can only be partly supported. In favour of this theory is the fact that with new ASR systems new KPIs were introduced. An example is the rate of computer-generated orders that were accepted by store employees. Another example is GROCO: this retailer now checks the inventory levels of its stores through a new KPI, something that was not possible before ASR introduction. On the other hand, the two retailers with the most advanced systems (DRUCKS and GROCO) do not measure any special logistics KPIs such as the MAPE or the inventory accuracy that are not in use by the other two companies. Therefore, it is difficult to support or decline this hypothesis.

6.4.2. Inventory Level Performance When asked about inventory levels at the distribution centres, the interviewees in the sample were quite content with their stock levels. The retailers are still trying to reduce their inventory levels, but they do not expect any radical improvement in the near future. The picture changes radically when the interviewees are asked about inventory quantities in the stores. The interviewed retailers stated that a high proportion of the goods in the supply chain they control used to be located at the stores. These findings concur with the situation described in the study by Smros,

158

6. Field Research and Managerial Implications

Angerer et al. (2004a). Their average inventory levels of 12 major European grocery retailers are depicted in Table 48. Distribution centre Fresh 2 days 0 days* 10 days Other 13 days 7 days 20 days Store Overall (incl. fresh and other goods) 12 days 4 days 16 days
*These goods are cross-docked
99

Inventory Median Minimum Maximum

Table 48: Inventory range of coverage of European grocery retailers in days

The retailer NORTHY is very satisfied with the development of its inventory levels after the ASR introduction. As a logistics manager said, the inventory level could be "considerably" reduced. The same statement comes from the retailers MYFOOD and GROCO. Concrete numbers could not be delivered by the interviewees. Yet, in GROCO's pilot the stock reductions were in part so strong that the central planners had to intervene and artificially raise the parameters. Too low stock levels are, from an aesthetic point of view, not desirable. One of the categories profiting most from the ASR systems are very fresh products, as the mean remaining shelf life of these products could be increased. Only DRUCKS is able to report a concrete figure. With an ASR4 introduction, this retailer has reduced its store stocks by about 1020%. The quantitative analysis of this thesis shows, that a reduction of this magnitude is also possible for grocery retailers.

6.4.3. OOS Reduction and Overall Performance MYFOOD is very pleased with the results from the ASR2 and ASR3 system implementations. The logistics managers report a reduction in OOS incidents, without been able to quantify it. Another benefit is that the time necessary for the daily orders could be reduced dramatically. A positive side-effect of the pull system is that the orders from the stores to the DCs are now more stable and thus more predictable. One of the highest financial impacts might have a last effect: the delivery frequency of some small stores could be reduced to three times a week. All these positive results have encouraged MYFOOD to introduce higher ASR systems for almost all categories across the country.

99

Source: Smros, Angerer et al. (2004a).

6. Field Research and Managerial Implications

159

The results of GROCO's ASR4 pilot are rated very positive by the ASR project members. The first benefit is again the time savings from the lost ordering procedure. The savings depend very strongly on the size of the store, for larger stores up to four working hours are saved, while the smallest stores only experience a reduction of half an hour. As in the previous case, the overall distribution system profited, as the DCs experience nowadays less variance in the stores' orders. Overall, GROCO is very pleased with the new ERP system, as the IT creates a transparency that is a requisite for any further improvement of the supply chain. For example, it is now for the first time possible to calculate the impact of a reduction of stores replenishment frequencies, as the ERP systems have created a new visibility level. This would in turn lead to greater cost savings. Another positive effect is described by one logistics manager: "After the introduction of the ASR and ERP system, our eyes were opened. For example, it was thought before the introduction that the shrinkage values in the distribution centres were minimal. Now, every little transaction is recorded minutely. This has created a new awareness about what logistics really costs. It is just amazing, how much information can be drawn out of these systems." The retailer NORTHY is very satisfied with the performance of its systems, as in addition to a reduction in inventory levels, the OOS rate could be dropped as well. Furthermore, the time spent by store employees in placing orders was dramatically reduced. The new spare time is used for a better control of inventory records and to provide greater service to the customers. In addition, the part-time contract with some employees has been reduced. Yet, an interviewed logistics manager stressed that after the implementation nobody was forced to quit the job. Another side benefit was that with the new system NORTHY's stores tend to stick more to the planograms and assortment decisions made by central office. The new IT systems improved the control of the stores due to increased inventory visibility. Consequently, new KPIs and mechanisms to control and measure the flow of products are in pilot, for example, a system that detects OOS incidents by analysing sales data. This system produces at the moment very crude estimations and is not currently working for slow-movers. More side benefits are expected from new electronically available information that is obtainable now for the first time. Based on this data, the next project will be to improve the amount of shelf space allocated to each SKU. As product managers are now responsible for the parameters and hence in closer contact with each store, they are now able to determine the shelf space based on facts. As successful as the implementation of ASR ordering was for packaged goods, the results from the automation of perishable goods are still not satisfactory. The employees are rather unhappy with the system, and this is most probably not going

160

6. Field Research and Managerial Implications

to change in the near future. Therefore, the management is thinking about returning to the old manual system. A possible reason for this underperformance might be the direct deliveries to the stores. Fruit and vegetables are delivered mostly by small suppliers, and this group is known to have major problems fulfilling EAN barcodes standards. For DRUCKS, the reduction of costs was in the opinion of the interviewee only a secondary reason for the introduction of the new ASR system. The main goal was to increase availability in the stores and to increase service to customers. The OOS rate was reduced by 7080% and is now between 1 and 1.5%. An average store had to spend between 3 and 5 hours to make an order of the complete assortment. Sometimes to reduce complexity, on certain days only part of the assortment was ordered to save time. The consequences were higher than necessary inventory levels. Today, only between 15 and 30 minutes is necessary for the ordering of the entire assortment. The time saved can now be used to increase inventory accuracy or, more importantly, to deliver better service to the customers. With the newer systems it is possible to have more SKUs per store, increasing thus the productivity of each sales m2. Also, logistics has profited, as the orders to the DS have become predictable.

6.5. Lessons Learned and Recommendations for Management

There can be no doubt that the automation of store replenishment is on the agenda of today's retailers. Four of the five biggest grocery retailers in Switzerland have implemented or at least piloted ASR systems. In a survey conducted in 2004 with more than 30 Swiss, German and Austrian retailers, around two thirds already had or were at least thinking about implementing higher ASR systems.100 The main reason why retailers had not yet started with the implementation was that a certain level of technology is necessary before the automation can start. For example, it does not make sense to have sophisticated ordering algorithms if the retailer has not even introduced a barcode system. In the following, the results from the quantitative and the qualitative analysis are merged. First, the adequate ASR level depending on the product and retailer's context is determined. Next, methods for successful ASR introduction are presented. The technical and organizational requirements are also discussed as best-practice store operations. This section ends with a cost-benefit analysis.
100

These short telephone interviews were conducted in summer 2004 by the author and involved retailers in the sectors electronics, apparel, grocery, jewellery and drugs.

6. Field Research and Managerial Implications

161

6.5.1. The Adequate Automation Level: Recommendations This thesis does not claim that every retailer should have the highest level of automation that exists. One of the aims of the thesis is to create a guide aiding retailers to decide for each product category the level of automation best suited to them (question Q4). The combination of the information gathered from the interviews and the quantitative analysis is shown in Figure 50. With such a decision tree, each retailer is able to find the adequate ASR system for each product category. As stated before, a company might opt to choose the best suited ASR level for each product category so that it may ultimately decide to run different systems in parallel, as seen with the examined retailers.
Which System to Choose for a Product Category? Which System to Choose for a Product Category?
Is electronic Is electronic inventory possible? inventory possible?

no

yes no
Is quantitative Is quantitative forecast better than forecast better than qualitative? qualitative?

yes

Stable demand, Stable demand, low value? low value?

yes

Can Can causal influences be causal influences be quantified? quantified?

yes

no

no

ASR0

ASR1

ASR2

ASR3

ASR4

Manual Manual ordering ordering

Ultra fresh (bakery, cheese counter)

Electronic Electronic inventoryinventorybased based ordering ordering

Fashion; Slow moving, high value items (jewellery)

Simple Simple ordering ordering heuristics heuristics

Low value items (clothes pegs)

Time Time seriesseriesbased based forecasts forecasts

Non-food (shoe-cream) Dairy products (milk)

Causal Causal modelmodelbased based forecasts forecasts

Drinks Barbecue Flowers

Figure 50: Decision tree for practitioners

In the following, the logic behind this decision tree is explained. The first node on the ASR level decision tree tests the basic possibility of having electronic inventory records for the product. Electronic inventory management is the basis of any automatic replenishment system, and so at this point SKUs that have to be ordered manually (ASR0) are separated from the others. For some products, electronic

162

6. Field Research and Managerial Implications

inventory management is not possible or economically feasible. The case of GROCO's served cheese counter discussed previously is an example of such problematic SKUs, and this retailer has decided to keep this part of the product assortment in the manual ordering scheme. The second node in the tree separates the products for which a forecast is sensible from those where it is not possible to make any forecast or where it does not make economic sense. An example for products where forecasts are nearly impossible is apparel products with a strong fashion influence. These products tend to have such unpredictable sales behaviour that any system would have difficulties predicting their sales from historical data. As every season new products appear or are trendy, there it is often no comparable data because the products are too different from each other. Other products where forecasts are not possible are unique SKUs that are only ordered once (e.g. in the case of promotions or special events). For the products where forecasting is not recommendable, retailers can choose between a semimanual (ASR1) and an automatic replenishment system with simple heuristics (ASR2). There are three reasons for choosing the manual method instead of the heuristics found in practice. First, some companies have very expensive products and a small assortment. An example for such retailers is a jeweller. The speed of turnover of such luxury goods is limited, hence manual ordering is not too complex. A second reason for retailers not to chose an ASR2 system is when their products have very high demand volatility or short cycle life, such as fashion articles. In the opinion of one interviewee, simple systems such as the minimum-maximum algorithms work fine until volatility reaches a certain point. The third argument for ASR1 systems is that for products where the quality strongly depends on the season or on the supplier, it is desirable to have the full control over ordering. In the field study, several retailers still ordered fruits such as strawberries manually. A human planner was still responsible for these products, and it was his or her task to decide when to order which quantity from which supplier.101 Although these products are ordered practically manually like ASR0 products, it still makes sense in many cases to have stock levels stored electronically. A high inventory visibility is reached with such electronic inventory records; these systems allow all direct and indirect benefits described in the previous chapter.

101

These statements only hold true for products that are delivered directly to the store. In a central distribution system it is possible to have part automation. The store orders automatically at the central office (ASR2 and higher) in the stores, there a manual planner determines the supplier and quantity to be ordered.

6. Field Research and Managerial Implications

163

The second group of products for which forecasts do not make economic sense can be automated by having simple ordering algorithms (ASR2). As one could see in the quantitative part of this thesis, many of the products on the ASR2 level had good performance compared to the manual products, although the underlying logic was rather simple. When the parameters are set well and stock level is not the predominant issue, these systems deliver reasonable performance with a low replenishment system effort. Products in these categories are all products that have a low economical importance for the retailer (such as some non-food household articles for a grocery retailer). Non-expensive articles with a long shelf life where a reduction of the stock level is not mandatory (such as clothes pegs) can profit from this level. Such simple rules can be beneficial also for products with very predictable sales behaviour.102 As the ordering logic is very basic at this level, retailers with low-performance IT systems might choose this level. Consequently, the last two ASR levels only make sense for products where there is a benefit from increased effort put into forecasting. The difference between ASR3 and ASR4 is the sophistication of the forecasts. ASR4 systems seek to increase the quality of forecasts using causal models. If the causal variables cannot be quantified or are ambiguous, the ASR3 level might be the better option. For products where the inclusion of simple casual variables (such as price) into the forecast is relatively easy or for products that are very important for the retailer (due to marketing or financial reasons) the effort made to improve forecasting might pay of. An interviewee from GROCO recognized that on hot days the availability of drinks suffered. Because at the same time, the sight of empty shelves had a negative impact on customers, the company regarded the improvement of availability as paramount. As water crates are very bulky, increasing the stock was not a feasible solution. Therefore, the inclusion of the weather in the system's forecasts is a possibility that was taken into account. This decision tree was based on empirical research carried out with dozens of retailers. The problem again is that the decisions made in the nodes are based on cost-benefit decisions. The case of DRUCKS showed, for example, that forecasts are not very expensive in terms of required IT-power. Consequently, DRUCKS makes a forecast for all its products. As it would be too time-costly to optimize every forecast, a system was chosen that dynamically optimizes its own parameters. Other retailers, for example MYFOOD, have for the majority of its automated products an ASR2

102

As can be seen in Figure 19, the ASR2 system at MYFOOD worked quite well for products with a demand volatility up to a sales coefficient of variance of 250%. For products with a higher demand volatility, ASR2 systems might not work efficiently anymore.

164

6. Field Research and Managerial Implications

system, although, according to the decision tree, ASR3 would also be adequate. To assess a change of these products towards the new system really is economically sensible a cost-benefit comparison between both systems would be necessary taking into account the context of the retailer. Overall the recommendation for retailers from the empirical analyses is that only for a few products are completely manual systems recommendable. If there are no restrictions regarding IT systems, the costs for automatic systems or other contextual restrictions, the overall recommendation is to implement higher ASR level systems. Automatic systems have the great advantage of levelling the influences of products on replenishment performance, as shown in the quantitative analysis. For the manually ordered products, the OOS rate was influenced by all the product characteristics measured, namely sales variance, speed of turnover, the price, CU/TU, product size and shelf life. On the other hand, ASR2 and ASR3 systems showed much more stable behaviour, their OOS rate was in most cases not influenced by product characteristics. Another result that was unambiguous is the importance of the right setting of replenishment parameters. The moderate results of MYFOOD's ASR2* systems show that wrongly or crudely parameterised systems will not give the results expected. The quantitative analysis showed another effect of higher ASR systems: performance differences between stores are reduced. There is the danger that excellent stores will worsen their performance by such automatic systems, as even the best IT system is not able to outperform a good store manager who has many years experience in the business and who knows exactly, how his or her customers behave.103 But in the final analysis it can be seen that automatic systems tend to improve the average performance of stores making them thus an attractive option for retailers.

6.5.2. ASR Introduction The introduction of ASR systems was for all four retailers in the field study an internal company decision and was mostly realized without the help of suppliers. The only necessity for suppliers is to have a certain technology standard (e.g. EAN barcodes, electronic dispatch advice) and a reliable service, as automatic ordering systems are more sensitive to delivery errors than manual systems. In half of the cases, the ASR systems were introduced together with new ERP systems and with major changes in

103

This is the opinion expressed by most practitioners interviewed. Yet there is no firm evidence for this statement.

6. Field Research and Managerial Implications

165

distribution logistics. In the other two cases, implementation was nevertheless a huge challenge as well because millions of parameters had to be set correctly. The setting of the parameters is, where possible, automated. One method would be to copy parameters for a new store from an existing comparable store. After the ASR system has run for a certain time, the fine tuning of the parameters begins. Another major effect that has to be considered is the impact of new ASR systems on personnel, as a fundamental duty of the store employees is cut. In three out of four cases, comprehensive training was organized to teach the right usage of the new systems. MYFOOD, GROCO and NORTHY created new organizational units as a consequence of the introduction of ASR. All retailers began by testing the systems with a small pilot. First, a single store with pilot product categories is observed during a certain time. The starting categories contain non-critical goods such as products with a stable demand, long shelf life or slow speed of turnover. Then, these categories are introduced nationwide. More complex categories are then automated on a rolling basis until the desired level of automation is reached. The retailers in this study have achieved good results with this procedure, and it is recommended to any retailer introducing ASR systems.

6.5.3. Technical and Organizational Requirements In this section, the technical and organizational requirements are discussed by automation level (research question Q5). Companies have the opportunity to implement ASR1 systems as soon as an IT-based inventory management system is available. The IT needs the capability of handling enormous amounts of data, as daily stock levels of all SKUs in each store have to be stored. New inventory systems also require a change in retailer processes and equipment. Because consumption has to be measured in order to update inventory levels, sales have to be tracked. For major grocery retailers it is not feasible to make an inventory check every evening to track all the transactions made during the day, in the same way as a number of small bakeries that were observed. In order to do that, identification systems have to be implemented that store electronically every sale at the cash desk. The most popular identification technology method used nowadays is the barcode. In future it might be RFID tags. To be able to calculate the inventory level in the store, it is obviously also necessary to track deliveries to the stores. This can either be carried out at the gate of the store, or at the gate of the supplier.

166

6. Field Research and Managerial Implications

As inventory visibility plays such an important role for automatic systems, procedures need to be put in place to ensure a high level of accuracy. Products that are not sold through the regular channel, e.g. because of shrinkage or spoilage, have to be somehow taken into account. This can be done with manual inventory checks where the items are counted and the real numbers reconciled with those in the system. In the field study, the described procedure seems to work rather well for packaged products. But there are certain products that are much harder to control, e.g. fresh products such as meat or cheese without packaging. For retailers selling such products, it is perhaps not advisable to use ASR1 as tracking of the goods is too time-consuming or too inaccurate. As noted earlier, updating to ASR2 systems is straightforward from a technical perspective. The simple heuristics do not require much IT-performance. The challenge in these systems lies in setting the parameters. A grocery retailer with 1000 stores and about 10,000 SKUs has to set in theory 10 million parameters. Furthermore, these parameters have to be checked and adapted on a regular basis, as demand in an individual store might change. In practice, companies automate this step by letting the computer decide on the parameters based on the past sales and restrictions such as lead time. For products without data history, the data from similar products is analysed. After some weeks of experience with the new system, OOS incidents and high inventory cases are manually examined, to see if the parameters have to be adjusted. ASR3 systems require a forecasting engine that continuously estimates future demand. To take again the example of the retailer before, it is now necessary to make a forecast for each product, meaning the calculation of 10 million forecasts. And in contrast to ASR2, this is not executed once but depending on the replenishment frequency up to daily! Obviously, this is a challenge for any central IT system. The example of DRUCKS showed that decentralizing a PC in each store masters this task without any problem. The forecasts are based on past datathe longer the history of the data, the higher the chance that regular variations (such as seasonal effects) and trends can be predicted correctly. On the other hand, an increase in the data requires an increase of storage capacity and computing power for the forecasts. The cost of forecasts is very difficult to quantify, therefore in Figure 51 only a schematic cost-benefit sketch is depicted.

6. Field Research and Managerial Implications

167

Increasing costs

Models that naively project past patterns

Should try to operate in this region

Models that seek underlying causes

E (total cost)

E (cost of resulting forecast errors)

E (cost of operating a procedure)

Decreasing forecast errors

Figure 51: Cost of forecasting versus cost of inaccuracy
104

The forecast method used in ASR4 systemscausal modelsis the main characteristic of this type of system. For causal models to work, it is not sufficient to see the past history of sales. Events have to be introduced into the equation as well. Basically, every event that has an influence on the demand can be included. Examples are the weather, price changes and promotions (by the retailer and by its competition), political events, holidays, among others. The challenge here is to choose the right parameters. Every added event is an opportunity to improve the quality of the forecast, but at the same time it inflicts higher costs. Most of these events cannot be easily automated, such as the information about future actions of the competition. There are few retailers with experience of this kind of system so that it remains to be seen if the additional cost of implementing the events is justified by increased forecast accuracy. To be able to interact at Level 5, companies obviously need to have information about each of the nodes of the network. Take, for example, a system designed to optimize simultaneously the replenishment of the store, the DC of the retailer and the DC of the supplier. ASR14 systems that want to develop to ASR5 need information regarding inventory levels at each site. ASR24 systems require in addition information on restrictions. ASR3 and 4 systems need the ingoing and outgoing data

104

Source: Silver, Pyke et al. (1998, p. 77).

168

6. Field Research and Managerial Implications

of each of the nodes. And for ASR4 systems to become ASR5, the events that might influence each of the nodes have to be implemented as well. A final question that will arise at any level of replenishment automation is the right choice of organization structure and power distribution. Obviously, this question is so fundamental that it can not be decided only as a consequence of an implemented ASR system. The examples in the field research showed that it is more easy to introduce ASR systems in organizations where the decision-making power is centralized. For higher systems, the power shifts even more towards central office. For ASR3 and ASR4 systems it is recommended that central planners are introduced to take care of the parameters in the forecasting engines and to intervene in exceptional cases. Yet, even with the most sophisticated systems, retailers might choose to let store employees have the final word on the replenishment system and give them full rights to alter the computer-generated orders. The results of this section are summarized in Table 49. Technical Requirements and Recommendations
- Electronic inventory management system - SKU identification system (e.g. barcode) - Electronic checkout - Data storage capacity: few daysweeks - Setting of replenishment parameters - Electronic order transmissions (e.g. EDI connection) - Electronic dispatch advice - PDAs - Forecast engine (based on historic data) - Self-optimizing forecasting engine - Data storage capacity: some weeks2 years - Technical implementation of events (such as weather) - Event and historic data-based forecast engine - High central computing power or PCs in each store - Central IT system for coordination - Electronic connection to all nodes - Highest computing and data storage capacity

Level ASR1

Recommendations on Operations and Organization

- Inventory records checks - Checkout processes - Goods entry processes - Accuracy in checkout processes - Spoilage and returns control - Regular check of parameters and inventory performance - Check of unusual sales or high inventories - Planners in central office - New products: comparison with similar products - Event system: check of unusual forecasts - Correction for promotions and special events - Past and future event data maintenance

ASR2

ASR3

ASR4

ASR5

- One organization unit responsible for entire supply chain

Table 49: Technical requirements and recommendations on operations and organization structure

6. Field Research and Managerial Implications

169

6.5.4. Store Operations: Recommended Action Independent of the final ASR level chosen, this section lays out the lessons learned from the case studies in four fields, implications of the quantitative analysis are discussed. The aim is to increase the performance of automatic systems with the right practices in different areas of store operations (see Figure 52).

Store Operations Recommendations Store Operations Recommendations

Awareness and Training
OOS impact on business Customer reaction Reshelving

OOS Measurement
Initial physical audit Regular combination of physical and electronic checks

Inventory Accuracy
Training of processes Regular physical audits Random checks Incentive programme

Promotions
Combination of qualitative and quantitative planning techniques Involvement of supplier and store managers

Figure 52: Overview of store operations recommendations

Several of the practitioners interviewed mentioned that employees have no awareness of the problem of OOS incidents. Consequently, this should be the first action taken. A physical audit of stock availability in the stores could help to create this awareness. Such an audit may show that causes for OOSs are poor inventory records accuracy and the way promotions are handled. Recommendations on these four topics are presented in the following.

Awareness and Training

Store Operations Recommendations

Knowing that a certain problem exists is the first step towards solving it. If employee awareness of the OOS problem improves, then there is a good chance of an increase of product availability. In one of their projects, Stlzle and Placzek (2004) measured the OOS rate of a German retailer over two weeks. In the first week the auditing was conducted without the employees being informed about the project, in contrast to the second week. The result was that in the second week the OOS rate dropped by 26%. Only by knowing that the OOS rate is being measured, the employees managed to increase availability dramatically. Another example of the importance of having the employees informed about the availability results from MYFOOD's OOS-project. One of the most notable results is that regarding the top 10 articles, i.e. the articles that are most sold in the stores. Two of them have an extraordinarily high OOS rate (12% and 22% respectively). Managers have to ensure that the most important goods in the stores in terms of turnover are paid the most attention by the personnel.

170

6. Field Research and Managerial Implications

Consequently, employees should be informed about which products are most critical for business and for the satisfaction of customers. Second, it is necessary to inform employees about the reactions of customers confronted with an OOS. Several retailers interviewed expressed that they were not concerned about gaps in the shelves, because, in an OOS situation customers would simply substitute the product. However, this is only true for some categories and only to a certain extant. By calculating the lost business due to such OOS incidents, the magnitude of the problem becomes clear, as made by Gruen, Corsten et al. (2002). And third, as 25% of all OOS incidents stem from reshelving practices, existing practices should be reviewed in detail. Several retailers have introduced handbooks with well-defined reshelving procedures. One English retailer has printed a bright star on the shelf floor that becomes visible as soon the product is out of stock. As ASR systems are able to relieve store employees, some of the spare time could be dedicated to accurate reshelving. As was shown in the quantitative analysis, a reduction of the backroom can be beneficial for the availability rate and should be considered by each retailer.

OOS Measurement

Store Operations Recommendations

As was depicted in the last section, a physical audit of the products' shelf availability is a good way of creating awareness and gives a one-time picture of the OOS situation in a store. But retailers are seeking a controlling to ol with which they can check availability on a regular basis without the cost of manual audits. Therefore, the logical next step after having introduced sophisticated systems to control and automate replenishment is to have a controlling instrument for availability. During the KLOG project undertaken for the MYFOOD, it was clearly showed how complicated this task is. Even when high accuracy is guaranteed, there will be always OOS situations that remain undetected. When availability suffers because shelf replenishment from the backroom has not been executed in time, it is not possible to detect this problem by just looking at the stock levels data. As the example in Figure 53 shows, the product in this example was at all times available in the store. But, because the replenishment from the backroom to the shelf did not work as expected, this SKU had a shelf OOS rate of 21% during the two weeks of the study.

6. Field Research and Managerial Implications

171

120 100 80 60 40 20 0 1 2 3 4 5 6 7 8 9 10 11 12

Inventory in store

Shelf inventory

Figure 53: Comparison of inventory on shelf and total store inventory for a glue stick

Another possibility for calculating indirectly the availability rate is by analysing the product sales data. In case of the sales of a product being unusually low for a day, an availability problem might be the underlying cause. Obviously, this method is only reliable for products with higher speeds of turnover.105 As the research on these systems is in its infancy, it might be recommendable to have a combination of physical and electronic audits to determine the OOS rate. A south-European Retailer has introduced a threefold way of looking at OOSs, as described by Smros, Angerer et al. (2004a): 1. One hour before the store is replenished, the store employees walk the store and count the items that are OOS (i.e., they count the percentage of items available before the store gets replenished). This metric is fairly conservative in that it measures availability when the likelihood of stock-outs is relatively high. The metric only measures the number of items OOS, without taking into account their selling rates. 2. A second tool is used to estimate unfulfilled demand. This metric takes into consideration the speed of turnover of products. In other words, the company not only measures the percentage of items not available in the store but also estimates the total sales that it might lose in the event that customers are not willing to substitute. One issue faced when using this tool is the difficulty in estimating the potential demand for items that are OOS. The tool basically provides statistical estimates of how many items the product could have sold had it been available. It also corrects for situations, such as wrong inventory records or defective items, where the system thinks a few units are left on the shelf but either no unit is actually
105

According to ECR France (2002) about three or more sales per day are necessary for such methods.

172

6. Field Research and Managerial Implications

there or just one unit that customers are not willing to take (as it is damaged) is available. 3. Finally, the company takes into account the attractiveness of the shelf. In order for the shelf to be attractive to the customer, it has to be filled to a given percentage of its capacity at least. The key concept behind this metric is that inventory on the shelf is not only necessary for meeting demand but also necessary to draw the consumers' attention and create demand. The company does not only want to measure the ability of the inventory position to fulfil demand but also its ability to create demand. The company thus also measures the percentage of items with inventory positions below a specified percentage of shelf capacity. By combining the information of the manual and electronic methods, the company's supply chain managers receive a very good sense of how the consumers experience the quality of store operations.

Inventory Records Accuracy

Store Operations Recommendations

The importance of inventory records accuracy was shown in the chapter about field research. For every automatic system high accuracy is beneficial. Yet for the higher ASR levels it becomes even more important as the order is often placed without a second check through human beings and the safety stocks are calculated more tightly. For this reason, employees need to be made aware of the importance of accuracy when working with ASR systems. Processes such as the check-out and the recording of spoilage, returns and stolen goods have to be defined and trained intensively. To control the results of these methods on records accuracy, the usual biannual counting of the goods is not sufficient. A continuous manual audit process is necessary that checks the accuracy of the records. This is feasible because store employees should have more free time due to the automation of the ordering. But without a defined procedure the retailer will not benefit from this spare time. One solution seen in field research, the "0123 walks," is a good example of such a defined process (see section 6.2.1). This process has many of the characteristics a good inventory auditing system should have: it is accurate (as counting errors with so few units are not probable), it is easy to understand (guarantees acceptance by personnel), it automatically concentrates on products with possible inventory inaccuracies (as that might be the reason why the inventory level is so low), and in the long run all products are checked at least once (as one can expect that all products will have a low inventory at some point). From time to time, random samples

6. Field Research and Managerial Implications

173

of products should be taken to check whether the inventory accuracy process works as expected. Several of the examined retailers run an incentive programme to promote inventory records accuracy. With a regular auditing mechanism in place, the positive effect of this incentive system should increase. Nevertheless, one should be aware that checking inventory accuracy too much might have some drawbacks as well. Employees should not forget other tasks that are also important, such as the tidiness of the sales area or customer service. To illustrate this, take the example of a grocery retailer in the UK. This company had to stop an accuracy campaign it had started a short time before; employees were requested to stop checking daily for inventory accuracy. The problem was that the hundreds of corrections made in the retailer's IT system created a commotion that distorted and confused the system, worsening the replenishment process instead of improving it.

Promotions

Store Operations Recommendations

An important lesson learned from the quantitative analysis was the role of promotions on OOS, as promotions dramatically change the inventory and sales level, thus increasing the chance for OOS situations. How much attention promotion-related OOSs receive depends very strongly on the policy of the retailer. Some retailers willingly take into account that special goods in promotions run OOS after a certain time. Other retailers, however, have a more consumer-oriented approach and want to offer the product during the entire promotion period. It is obvious that for retailers with such an approach it becomes most important to calculate the right quantity of SKUs in promotions. In the case of MYFOOD, promotions are planned completely manually. As the results are not entirely convincing and vary sharply from store to store, managers should consider a mixed approach, where the technology supports the employees in their planning decisions. A best-practice example is reported from a European grocery retailer (Smros, Angerer et al. 2004a). This example is meant to encourage retailers that are still handling promotions manually, to consider the usage of ASR systems as well: Before the promotion starts this best practice retailer uses a quite sophisticated demand-forecasting tool to predict the number of customers that will enter the store and the rise in sales per customer. The forecasts take into account variables such as the promotion characteristics, the price cut and the weather, among others. This initial forecast assumes that all stores are going to enjoy the same promotional lift. However, it is well-known that the effectiveness of the promotion depends on many store-specific variables, such as where the product is placed in the store, the number

174

6. Field Research and Managerial Implications

of facings it has been allocated, etc., and that these variables can seldom be fully controlled by the central organization. What the company does is to look at the first few days of sales at each store and then update the initial estimates literally in real time. Since this retailer benefits from a quick supply chain (both lead times from distribution centre to store and from supplier to distribution centre are quite short), the company can adjust its replenishment plans promptly. The algorithm used behaves in a way like a human planner. At the beginning of the promotion, the ASR overstocks the stores. It knows that even if the products do not sell as expected, it is likely to be able to sell them by the end of the promotion. Towards the end of the promotion, by contrast, the system is more conservative, since it acknowledges the increased risk of being stuck with the product when the promotion is over.

6.5.5. Cost-Benefit Analyses The four retailers studied into detail have drawn a positive conclusion of their ASR implementation. For the interviewees the advantages of higher ASR systems exceed the possible drawbacks. The reported benefits are: higher availability on the shelves, higher turnover, time savings due to automatic ordering, lower stock levels in the stores, smaller stores require lower delivery frequency, more stable demand at the distribution centres, and inventory and cost transparency along the retailer's supply chain. These benefits correspond to the ones reported in theoretical and practical sources dealing with systems that improved forecasting accuracy. For example, Gudehus (2002) and Duffy (2004) report improvement in operations such as decreasing inventory levels, increasing service rates, fewer OOS, more on-time deliveries, shorter order-to-deliver cycle times and more accurate KPIs for the management, among others (see also Table 4). The main initial impact of an ASR introduction is the investment necessary in new IT systems. In all cases, the new replenishment systems were received with scepticism on the part of employees. The interviewees stressed several times the importance of taking serious the wishes and concerns of the workforce. Otherwise, the implementation is likely to fail. Interestingly, in all four cases, after a couple of months the uproar settled down and employees got used to the system.

6. Field Research and Managerial Implications

175

But introducing ASR was not a success for all products. While DRUCKS, GROCO and MYFOOD report fantastic results, NORTHY is not happy with the performance of green grocery products ordered through their ASR3 system, even several after months after introduction. A return to the old system is being considered. The question is whether these sorts of products are not amenable to automation or whether the chosen ASR3 system was not adequately parameterised. The fact that other retailers have had good results with their systems for this kind of product speaks in favour of the latter cause. NORTHY has some problems with the barcodes and dispatch advices for these products, as they come mainly from small suppliers and this is a possible reason for the suboptimal results. As seen with NORTHY's case and the ASR2* systems at MYFOOD, there are many pitfalls in the implementation of such systems; detailed planning is necessary to ensure the success of the change. A major point of criticism among opponents of ASR systems is that the most sophisticated system will never cope with the best, long experienced store personnel. And they might be right, as all the implicit information earned by good store managers during a long life can never be adequately replaced by a computer system. Unfortunately, the ordering skills of store staff vary a great deal. Therefore, with the introduction of ASR systems, retailers aim to achieve constant ordering performance at a reasonably high level that will be better than the average manual ordering performance. The decision as to which level to choose for each product should be based on a costbenefit analysis. In theory, the calculation of such an analysis is simple. On the one hand, there are the benefits of having an automatic system as described in the last section. This line can be interpreted as a negative cost line, as the costs, for example OOS costs, are reduced (line CE in Figure 54). On the other hand, the more complex the replenishment system, the higher the costs for the replenishment will be (CR line). The optimum would be the point where the total cost line (TC) has its minimum.

176

6. Field Research and Managerial Implications

Increasing costs
TC: Total costs

CE: cost of errors of replenishment system

CR: cost of replenishment system

minimun ASR0 ASR1 ASR2

106

ASR3

ASR4

Automation level

Figure 54: Costs in relation to replenishment level

It is very difficult to transfer these rather theoretical thoughts to a real life cost analysis. The benefits and costs connected to such an implementation are summarized in Table 50. Possible Costs
IT investments IT running costs Personnel education Small, frequent deliveries towards stores Product handling costs in store Additional personnel at headquarters

Possible Benefits
Fewer OOS incidents Higher product turnover in stores Fresher goods Less frequent store deliveries Lower inventory costs at store level Less personnel costs for order setting Demand in DC easier to schedule Better service to the customers Inventory visibility

Table 50: Overview of possible benefits and costs following an ASR system introduction

The first difficulty is to quantify the benefits. As Fisher, Raman et al. (2000) state, most retailers are not able to quantify the value of e.g. reduced OOS incidents and markdowns. A rough estimate of the benefits of reducing OOSs is given by Gruen, Corsten et al. (Gruen, Corsten et al.). In their estimation, for each percentage point

106

Source: adapted from Silver, Pyke et al. (1998, p. 77).

6. Field Research and Managerial Implications

177

OOS reduction, an increase in turnover of 0.5 percentage points is to be expected. Some companies have in lavish calculations evaluated the benefits of the reduction of delivery frequency. Nestl Germany, for instance, has found out that a change from delivering every 24 hours to every 48 hours means a reduction in the costs by the factor 56 (Weder 2005). Therefore, MYFOOD's initiative to reduce the replenishing frequency has a major benefit potential. Other indirect benefits, such as the new reached level of inventory visibility across retailer's storing places are much more difficult to quantify. As difficult to calculate are the cost side of such projects. The technology costs will strongly depend on the technology already available on site as well as on the organizational structure. It is hence impossible to make a general quantification of costs. A centralized retailer changing from an ASR0 system to an automatic level which needs to first implement barcodes and scanners may find here the biggest cost block, making the change very expensive. As this retailer is already a centralized organized, adapting the organization might not be necessary. On the other hand, a decentralized retailer that already has a working ERP system might with just a software upgrade change from an ASR2 system to an ASR4 system at negligible IT costs. But this retailer might face strong opposition from the independent stores, making it very costly to adapt the organization and train the personnel. The effect of ASR on store delivery frequency and handling costs is not quite clear. As seen before, MYFOOD sees the possibility to reduce deliveries to its stores. GROCO logisticians reported frequent, low quantity orders created by an old ASR2 system. This not only increases the transport cost, but at the same time store reshelving costs. Obviously, it will depend on the parameterisation of the ASR systems whether ASR will reduce or increase the logistics costs. As stated before, many retailers combine the introduction of higher ASR systems with an introduction of or change in ERP systems. This makes a final assessment of the true benefits and costs very difficult. Instead of making a very imprecise estimate, in the following the results from a comprehensive cost-benefit study of MYFOOD is presented. This company changed its ASR system, its ERP system and its distribution network all at the same time. Although this makes it difficult to attribute real benefits to any single change, this case is a good example of how fast the restructuring of replenishment systems can be amortised. When ASR2 systems were introduced, MYFOOD lived simultaneously through a very radical change of its organization. Its entire distribution strategy was changed, namely from a decentral to a more central logistics system. Plans for the change of

178

6. Field Research and Managerial Implications

the distribution system covered a 10-year time period. After five years, MYFOOD drew its first cost-benefit conclusion. The goal of this analysis was to quantify the benefits of the change from a decentral, manual system to a central system with automatic replenishment. This task was described by one interviewee as "back-breaking work," as a homogeneous definition on how to value costs and benefits was necessary through the entire organization. The biggest benefit was an increase in turnover (due to better, centralized category management) followed by a reduction in purchasing costs and in stock. The increase in turnover adjusted by price inflation, sales area increase and marketing was calculated at 0.7%. This increase includes the effect of higher availability, fresher goods and a more attractive assortment due to central category planners. The biggest benefit was seen for seasonal, non-food products. The costs were split into three parts: IT, marketing and logistics. Surprisingly, the IT costs in the long run did not increase. During the implementation phase, there were indeed increased costs due to investments in equipment and personnel education. But after 23 years, the annual IT costs were again on the level of the former, decentral system. In logistics, a similar situation happened. During the ramp-up phase of the system, inventory levels did increase significantly, the central warehouse was loaded nearly to maximum capacity, and the distribution running costs were very high. This was partly a consequence of a security plan to ensure complete availability during the whole period. Yet after a very short time, inventory stocks could be reduced. Two years after the introduction, stocks were lower than before the change, although the turnover had increased. Overall, this special retailer is more than happy with the restructuring and implementation of the new logistics concept. As early as two years after the introduction of the central system, the accumulated benefits were higher than the costs. The yearly benefit for the retailer is a nine-digit euro number. Regarded from the viewpoint of the customers, the benefits are, according to the interviewee, a better assortment that is homogenized across the stores, a higher availability, and better service due to the free time of the personnel. The cost reduction for the firm will finally result in lower prices for the customers. The conclusion of this section is that in a cost-benefit analysis it is very hard to isolate the benefits from implementing a higher ASR, as only in a few cases the automation of the replenishment was realized independently from the introduction of a new ERP system or a change in logistics. Nevertheless, the quantitative and qualitative examinations conducted in this thesis show what kinds of benefits one can expect. The example of MYFOOD demonstrated how major investments in IT and logistics have paid off after only two years. Retailers have to quantify as far as possible the

6. Field Research and Managerial Implications

179

benefits and costs presented in this section before the final judgement on the financial benefit of an ASR introduction can be pronounced.

180

7. Conclusion

7. Conclusion
In this final chapter, the contributions for theoretical research and practitioners are summarized, before further research fields are depicted.

7.1. Theoretical Contributions

This thesis has some theoretical implications that are described in the following along the four theoretical perspectives used in this thesis (see Figure 55).
Theoretical Theoretical foundation for foundation for ASR research ASR research

ASR benefits ASR benefits Store processes Store processes

ASR and ASR and human agency human agency

Explanatory model Explanatory model

Inventory management Inventory management research research

Logistics & operations Logistics & operations management management

Business information Business information systems research systems research

Contingency theory Contingency theory

Figure 55: Theoretical contribution of thesis

The traditional inventory management research has concentrated on the development of mathematical models of replenishment systems (e.g. Galliher, Morse et al. 1959; Bassok 1999). This research stream is criticized by Wagner (2002) as too theoretical and too far away from the challenges of real life. The main contribution of this thesis is that a topic was chosen that is, on the one hand relevant for the challenges retailers experience every day and, on the other hand, a field that has not been in the focus of inventory research up to now. The novelty of this research field demanded first of all the creation of a descriptive model. In chapter 4 such a model was developed based on the traditional inventory holding models as described by Silver, Pyke et al. (1998) and Gudehus (2002). The descriptive model was the basis for the creation of the six-step classification system that is valuable for understanding the main differences between manual and automatic systems (see Figure 13). The classification of ASR systems facilitates the creation of hypotheses concerning replenishment performance by each different system. A quantitative analysis proved the significant performance differences between ASR levels. Consequently, the enhancement of the existing inventory management research stream by the creation of such an ASR classification system is appropriate. Another contribution of this thesis concerns the theoretical sources that examine out-of-stocks in retail. Most of OOS research has concentrated, on the one hand, on the extent of the OOS rate and, on the other hand, on the consumer reaction. In contrast, this thesis has taken a third aspect into focus, namely the influence of product and store characteristics on

7. Conclusion

181

the OOS rate. Again, the quantitative analysis showed that there are indeed several significant relationships between these characteristics and inventory performance that have not yet been studied in detail. A special field of logistics and operations management research examines the performance of automatic replenishment programmes. As predicted by Ellinger, Taylor et al. (1999) and Gtz (1999), the automatic replenishment systems reduced OOS incidents as well as inventory holdings. For the first time in an academic analysis, the size of this reduction was quantified in practice. Another most noteworthy outcome is that the influence of product characteristics on replenishment performance can be significantly reduced when ordering is automated. Sales variance, speed of turnover, price, case pack size and product size are significant influences on the OOS rate in the case of manual systems, but not for the automatic systems. Similar results are obtained regarding the influencing factors of inventory levels. A second contribution of this thesis regards store operations. A novelty is the description of store operations in the context of ASR systems, something that very few researchers have examined up to now.107 This thesis confirms some of the implications for business and human agency as stated by theoretical business information sources. For example, several papers predict the change of human working patterns due to the introduction of new IT-based replenishment systems. The employees in the retailers studied clearly had to radically change their working behaviour. The innovation in this thesis is to transfer the statements about ERP systems to the context of ASR systems. As all ASR systems examined in the sample are an integral part of superordinated ERP systems, this procedure can be considered appropriate. The result is, as expected by researchers such as Kallinikos (2004) that human agency is indeed deeply influenced by new ASR systems. The way orders are executed within the retailers examined is much more structured and transparent than it used to be before the introduction of higher ASR systems. Personal communication and typical human behaviour such as improvisation is strongly limited by the new systems. As recommended by Hong and Kim (2002), the aspect of organizational change in the context of ASR implementation was examined. The disruptive organizational change predicted by Soh, Kien et al. (2000) could be partially seen for some of the retailers, where new central organizational structures were created and a decision power shift occurred from the stores to the central office.

107

One of the few exceptions is the work of Broekmeulen, van Donselaar et al. (2004b; 2005).

182

7. Conclusion

The last theoretical contribution concerns contingency theory. This theory states that the effective level of some endogenous variable in a system depends on the level of other environmental variables. This theory has been applied in many fields, ranging from leadership style (Fiedler 1967) to organization structure (Lawrence and Lorsch 1967; Staehle 1991) to logistics (Pfohl and Zllner 1987; Chow, Heaver et al. 1994). This thesis applies the contingency aspect to retailers' replenishment systems. In the field study, the statements of Zomerdijk and Vries (2002) concerning the vast influence of retailers' context on the performance of replenishment systems were confirmed. The necessity of having the right organization (task allocation design and decision-making structure) and the influence of employees (the communication processes and behaviour) on the performance of the system was confirmed by many of the interviewees. The quantitative examination also showed the expected result. Although all stores examined had the same logistical restrictions and used the same replenishment system, their performance differed substantially from each other. The data confirms that the influence of store managers, product characteristics and store characteristics on replenishment performance are substantial. The typical answer of contingency theory to the question of which ASR system to choose is "it depends." This contingency aspect is reflected by the decision tree developed in this thesis, as there the choice of the right replenishment system for a certain retailer will mainly depend on product characteristics.

7.2. Contribution for Practitioners

The overall contribution to practice is that, with this thesis, managers have, for the first time, a systematic analysis of automatic replenishment systems. The analyses in this thesis were mainly based on grocery retailers, yet it is possible to transfer the results to any make-to-stock industry where availability plays a crucial role and orders are still set manually. This work is intended as a practical aid for managers to assess and improve their replenishment systems. An overview of practitioners' questions in the context of ASR systems answered in this work can be seen in Figure 56.
Which products and Which products and stores are prone to stores are prone to under-perform? under-perform? Do II have Do have the right system the right system for my products? for my products?

Which types of ASR Which types of ASR do exist? do exist?

Which benefits Which benefits can II expect? can expect?

How do II implement How do implement ASR systems? ASR systems?

ASR classification ASR classification

Quantitative analysis Quantitative analysis

Explanation model Explanation model

Decision tree Decision tree

Case study examples Case study examples

Figure 56: Contribution for practitioners

7. Conclusion

183

First of all, with the help of the developed classification system, retailers are able to gain an overview of the existing automatic systems. There are six different ASR system types, the main difference between them being the ordering logic. While basic systems calculate the order based on simple algorithms (e.g. minimum-maximum logic), the most sophisticated ones use causal forecasting techniques. Practitioners need to know the possible benefits of ASR systems. The analysis of a grocery retailers' data showed the various advantages of automatic systems. The retailer examined managed to reduce its inventory levels by 33% for non-food products and by 16% for dairy products compared to a random control group. In addition, the inventory levels of the dairy products were more stable when the products were ordered automatically by ASR3 systems. A major benefit is also the reduction of the number of OOS incidents. Products that were ordered manually had on average an almost 8-times higher OOS rate than products ordered by ASR2 systems. An additional benefit is that automatic systems show robust results. While the performance of manual replenishment is strongly influenced by product characteristics such as price, speed of turnover, etc., automatic systems exhibit much more consistent results. The time savings for each store are estimated between 0.5 and 5 man-hours, depending on the size of the store. Finally, the examined retailers report secondary benefits of ASR implementation such as the possibility to reduce logistics costs, for instance, by lowering store replenishment frequency. Nevertheless, the success of ASR systems will depend on many factors, as the explanatory model has shown. Understanding the influence of store and product characteristics on inventory performance can enable retailers to identify products and stores where the replenishment performance is probably not as desired. For instance, it was demonstrated that stores having the biggest backrooms tend to have at the same time the highest OOS rates. With this knowledge, necessary counteractions can be initiated. Practitioners want to know whether they have chosen the adequate system for their products. For this reason, the fourth contribution of this thesis is the development of a decision tree that helps managers to identify the right ASR level for each product category. There are still many products where manual ordering (ASR0) is appropriate. For instance, for very fresh products such as bread rolls, electronic inventory records are not feasible. For other products that show very irregular demand (e.g. fashion goods) or that are very expensive (e.g. jewellery) semi-automation might not be economically sensible (ASR1). However, for most articles of a typical grocery retailer automation should be strongly consideration.

184

7. Conclusion

Products with a stable demand or low value can achieve good results with simple ASR2 systems. All other products might benefit most from automation based on a forecasting engine, taking into consideration that computing power is nowadays not the decisive constraint. Retailers that have decided to change their replenishment system should be aware of the wider consequences of their actions. The implications for the companies go far beyond a mere improvement of the replenishment system. Retailers' technology, organizational structure and processes have to be adapted to successfully introduce ASR systems. Depending on the ASR level, different technical systems have to be introduced, such as an electronic inventory records system, electronic data interchange, forecasting engines, etc. However, retailers have seen the ASR introduction as an opportunity to harmonise their entire IT and other technologies in use, e.g. by introducing ERP systems or by installing the same type of tills in all stores. Another change concerns the organization of some retailers, where a shift in the decision-making power from the stores to the headquarters occurred. New central planning functions have been created that control the flow of stores replenishment and take care of the ASR parameters. Furthermore, ASR systems have dramatically changed the working behaviour of many employees. Not every employee welcomes the automation as a relief from the burden of placing orders, as this task is regarded by many as an essential part of their work. Successful retailers have created an acceptance among employees by showing them the benefits of the new systems. IT-courses have dispelled the fear of many store employees concerning the new technique. Other courses were used to show employees the new processes required to support ASR effectiveness. One example of such new processes are the ones that aim at increasing inventory records accuracy. As ASR systems are highly dependent on records accuracy, definition and training of processes such as physical counts, the check-out, the handling of shrinkage, among others, is crucial. Overall, the implementation of ASR systems is a great opportunity for practitioners. However, this thesis not only highlighted the benefits but also the risks of automating replenishment. Practitioners have to be careful, because the introduction of wrongly parameterised ASR systems will be counterproductive. At the end of the day, managers' final question will be: Is introducing ASR worth all the effort? This is a difficult question, and the answer needs to be decided by every practitioner in each individual context. The change from a push to a pull system requires a significant effort. The case study of MYFOOD has shown that only two years after the introduction of the new ASR2 system the investment had been fully amortised. Depending on the retailer's starting point, this period could be even shorter. The four

7. Conclusion

185

case studies of retailers which strongly benefit from ASR systems can be a major help for any practitioner planning to automate store replenishment.

7.3. Further Research Fields

Considering the novelty of this research field, the present thesis can only be the beginning of research projects on this topic. Six research fields have been identified that could be examined in future research to gain more detailed insights into the ASR topic. Further Research Fields
Quantitative analysis with larger empirical basis (several retailers) Research on non-grocery retailers Research on ASR4 systems ASR5: multi-tier systems ASR influence on human agency Overall replenishment cost function
Table 51: Overview of further research opportunities

First of all, the quantitative examination in this thesis concentrates on a single grocery retailer. It could be proven that the introduction of the replenishment systems was indeed beneficial for this retailer, as long as the replenishment logic was set correctly. The next logical step would be to have another empirical examination with several retailers to prove that this was not just an exception but that indeed all retailers can benefit from this approach. However, such a detailed examination as was made for this thesis may not be possible when looking at several retailers at once. Moreover, this approach would examine a broad sample of retailers, e.g. with the help of surveys, to create a statistical proof of the benefits of ASR systems. Second, the present research focused on grocery retailers and chemists. Yet from several interviews it is known that this topic is also relevant for other types of retailer. Further research could hence concentrate on the transfer of the present results to other types of products. Further studies are required in particular for articles with a high fashion influence and with long lead times (such as in the apparel industry). Third, in this thesis only the change from manual to ASR2 and ASR3 systems could be statistically tested. A most interesting question for retailers is whether an upgrade in ASR system is worth the effort. This knowledge is necessary to make a

186

7. Conclusion

cost-benefit analysis. Statistical examinations of retailers that have made the jump from ASR2 or ASR3 to an ASR4 system would be particularly useful to practitioners. A fourth possibility for further research is the examination of ASR systems from an ECR perspective. The main idea of ECR is collaboration along the supply chain in order to obtain better results. The question of the contribution of suppliers to the success of ASR systems is still to be answered, as most of the retailers examined in this thesis carried the implementation on their own. The models developed in this work only regard the last tiers of the supply chain. The final aim of such collaborative supply chains would be the hypothesised ASR5 system. The development of systems that optimize replenishment along the entire supply chain would be one of the biggest challenges in this field for future research. The next topic which could only be examined on the surface is the influence of ASR introduction for human agency. One possible approach would be a socialpsychological one. It would be interesting to see how the satisfaction of the individual store employee changes after the introduction of ASR systems and how long it takes to return to its former level. Another focus would be research on the ordering behaviour of store employees confronted with ASR systems, as conducted by Broekmeulen, van Donselaar (2005), and on the optimal degree of influence store personnel should have on the automatically generated orders. Finally, a topic that also requires further research is the one of the multiple tradeoffs a retailer is facing when dealing with the setting of the replenishment system. It was a remarkable result to prove that ASR systems are able to reduce OOS rate and inventory level at the same time. Yet, it is not clear what exactly the relationship is between the two. At some point, the store manager will have to decide on a tradeoff: further reducing inventory will raise the OOS level and vice versa. Besides, there are many other parameters that have influence on the total replenishment costs and the handling workload in the store, such as shelf space, case order size and data quality. The role of inventory records inaccuracy has to be examined in detail, as the retailer MYFOOD in this thesis had good performance results despite major inventory records inaccuracies. Consequently, further research should take a closer look at creating an overall cost function, where all the possible costs of store replenishment are taken into account alongside their influence on performance. This thesis provided a small contribution for the realization of this ambitious task.

8. Appendix and References

187

8. Appendix and References

8.1. Statistical Appendix
8.1.1. Calculation of the Inventory Level The difficulty with the data extracted from MYFOOD's ERP system was that the system does not record any absolute values of the stock level. This means that it is not possible to determine ex post the quantity of goods there were for a certain product on a specific day. In addition, MYFOOD interviewees stressed the inaccuracy of the inventory records. This statement is supported by physical audits (see Figure 49) and by other researchers (e.g. Broekmeulen, van Donselaar et al. 2005). The only available data was relative, i.e. showing how many units were delivered and how many units left the store on a certain day. With this data it is possible to determine the stock level curve, but without knowing at which height on the y-axis the curve is (see Figure 57) .

Calculated Inventory without zero line

supposed zero stock-point

16.08.2004

27.09.2004

Figure 57: Relative inventory level curve without zero line

Nevertheless, additional information makes it possible to determine the zero line. It is evident that the inventory levels in the store can never be negative. Consequently, the relative stock curve can be set in such a way that its minimum equals zero. This method seems adequate for several reasons. First, the only problem with this method would be with products that are never OOS. Nevertheless, the chances of this

25.10.2004

30.08.2004

13.09.2004

19.07.2004

02.08.2004

11.10.2004

188

8. Appendix and References

happening is very rare, because in the sample a data period of 90 days is covered.108 The chances of a product of being always in stock taken an OOS rate of 5% equals 0.9590= 0.1%. Second, even if the product was never OOS, the minimum value is likely to be close to zero, thus still delivering a good approximation. For 15 of the 100 articles in the sample, the actual inventory levels were measured in the store during two weeks. Comparing both lines, as can be seen in Figure 58, show the accuracy of this method.109 Third, the variance of the stock is independent of the height of the level, therefore this KPI is not affected by this procedure.

50 40 30 20 10 0 19.07.2004 02.08.2004 16.08.2004 30.08.2004 27.09.2004 11.10.2004 25.10.2004 13.09.2004 Calculated inventory after correction Manual counted inventory

Figure 58: Absolute inventory level curve after the correction

This method is not 100% correct. Especially for products with very low OOS rates, the inventory levels might be underestimated. However, greater accuracy is not necessary, as the main goal of this thesis was to compare different ASR methods with each other. As the possible inaccuracy of this method impacts equally both manually- and automatically-ordered goods, the comparison is still valid.

108

Therefore the analyses on inventory levels done with the two-weeks data (Dataset1) have to be viewed with caution. Yet, in those analyses it was not important to show the absolute level of the inventory, but more how the impact of an external variable (e.g. price) influenced the inventory level. For this purpose, this method is appropriate. 109 The still existing difference between both lines can be explained by the fact that the inventories were counted during the day. Sales happening after the count make the manual line stand above the calculated line. Another possible explanation for the small gap could be that the minimum value on August 14th was not zero but a little higher (around 2 units).

8. Appendix and References

189

8.1.2. ANOVA Considerations and Prerequisites The main task of ANOVAS is to prove the existence of a correlation between the variables, e.g. that for one group the OOS rate is higher than for another. Consequently, one has to be cautious about interpreting the size of this correlation (Backhaus, Erichson et al. 2003). The ANOVAs only prove that one group is statistically different from another, yet the direction, i.e. if the one group is higher or lower than the other, has to be seen from the context. Normally this is not a problem, as the descriptive statistics clearly show which group had the higher value. In the following, the data conditions that have to be fulfilled for an ANOVA to be applied are discussed (cf. Backhaus, Erichson et al. 2003, pp. 150152). First, the dependent variable has to be a metric scaled variable. In this thesis either the OOS rate or some stock-level-related variables are tested, this is fulfilled with the present data.110 Second, ANOVA designs normally demand that the sample taken is representative for the entire basic population. This can be assured, e.g. through a random sample. In this study the products and stores were not randomly chosen. Yet in the choosing processas described in the section 5.1it was taken care that representatives of each product category or store type that are relevant for the analysis of the performance were present. Therefore, no distortion of the analysis is expected. Third, the values in the population should have a normal distribution (Rao 1998) and fourth, the population variances have to be equal. But the last two restrictions are often relaxed in practice and nevertheless the results are significant ANOVAs (Backhaus, Erichson et al. 2003). In the analysis of this thesis, to compare the effects with each other, a Tamhane's T2 post hoc analysis is made where the assumption of equal variance is not required.

110

To be precise, most of the data presented in this case is censored, as e.g. the OOS rate cannot be lower than 0% or higher than 100%. Yet, as long as the data is not very close to these limits, one does not expect significant impacts on the results. In the marketing literature, this procedure is most common (cf. Katsikea and Skarmeas 2003).

190

8. Appendix and References

8. Appendix and References

191

8.2. References
Aalto-Setl, V. (2001). "The effect of concentration and market power on food prices: evidence from Finland." Journal of Retailing 78(3): 207-216. Accenture (2000). Science retailing: bringing science to the art of retail. White Paper. Achabal, D. D., S. H. McIntyre, et al. (2000). "A decision support system for vendor managed inventory." Journal of Retailing 76(4): 430454. Ailawadi, K. L., N. Borin, et al. (1995). "Market power and performance: a crossindustry analysis of manufacturers and retailers." Journal of Retailing 71(3): 211-248. Akkermans, H. A., P. Bogerd, et al. (1999). The impact of ERP on supply chain management. Fontainebleau, INSEAD. Alicke, K. (2003). Planung und Betrieb von Logistiknetzwerken. Berlin, Springer. Andersen Consulting (1996). Where to look for incremental sales gain. The retail problem of out-of-stock. Atlanta, The Coca-Cola Research Council. Anderson, B. V. (2004). Best practicesintegrated retail replenishment. The Post. 20. Angerer, A. (2004). "Out of Stock: Ausma, Ursachen und Lsungen." Logistik Inside 7(2): 16. Angerer, A. and D. Corsten (2004). Zum Nutzen der Kunden. Logistik als Motor des Einzelhandels. Frankfurter Allgemeine Zeitung. Frankfurt. Anon. (2004). "Organize for efficiency." Supervision 65(10): 2526. Anupindi, R. B., Yehuda (1999). "Centralization of stocks: retailers vs. manufacturer." Management Science 45(2): 178-191. Arnold, D., H. Isermann, et al. (2004). Handbuch Logistik. Berlin, Springer. Arthur Andersen LLP (1997). Small store survival: success strategies for retailers. New York, John Wiley & Sons. Backhaus, K., B. Erichson, et al. (2003). Multivariate Analysemethoden. Berlin, Springer. Balachander, S. and P. H. Farquhar (1994). "Gaining more by stocking less: a competitive analysis of product availability." Marketing Science 13(1): 322. Bancroft, N. H., H. Seip, et al. (1996). Implementing SAP R/3. Greenwich, Manning. Bansal, P. and K. Roth (2000). "Why companies go green: a model of ecological responsiveness." Academy of Management Journal 43(4): 717736.

192

8. Appendix and References

Bassok, Y. A., Ravi (1999). "Single-period multiproduct inventory models with substitution." Operations Research 47(4): 632-642. Baumard, P. and J. Ibert (2001). What approach with which data? In Doing management research. R.-A. Thitart. London, SAGE. Baumgarten, H. and J. Thoms (2002). Trends und Strategien in der Logistik Supply Chains im Wandel. Berlin, Technische Universitt Berlin. Baumgarten, H. and S. Wolf (1993). Perspektiven der Logistik: Trend-Analysen und Unternehmensstrategien. Berlin. Bearing Point (2003). "Inventory Management 2003: Communication and technology bear renewed promise." Chain Store Age(Section 2). Bell, R., R. Davies, et al. (1997). "The changing structure of food retailing in Europe: the implications for strategy." Long Range Planning 30(6): 853861. Bergeron, F., L. Raymond, et al. (1998). The contribution of information technology to the performance of SMEs: alignment of critical dimensions. Proceedings of the 6th European conference on information systems, Aix-en-Provence. Bergeron, F., L. Raymond, et al. (2001). "Fit in strategic information technology management research: an empirical comparison of perspectives." Omega 29(2): 125142. Beringe, A. v. (2002). The integrated supply and demand chain. Speech at Sapphire, Lisbon. Bernard, P. (1999). Integrated inventory management. New York, John Wiley & Sons. Bobitt, H. R. and J. D. Ford (1980). "Decision-maker-choice as a determinant of organizational structure." Academy of Management Review 5(1): 1323. Broekmeulen, R., K. van Donselaar, et al. (2005). An empirical study of ordering behavior of retail stores. Working paper of the School of Technology Management, Eindhoven. Broekmeulen, R., K. van Donselaar, et al. (2004a). Automated Store Ordering: an enabler for new inventory replenishment strategies in supermarkets. Report of the Faculteit Technologie Management, Eindhoven. Broekmeulen, R., K. van Donselaar, et al. (2004b). Excess shelf space in retail stores: an analytical model and empirical assessment. Beta working paper series, Eindhoven.

8. Appendix and References

193

Brynjolfsson, E. and L. M. Hitt (1995). "Computers as a factor of production: the role of differences among firms." Economics of Innovation and New Technology 3(3-4): 183199. Brynjolfsson, E. and L. M. Hitt (1998). "Beyond the productivity paradox." Communications of the ACM 41(8): 4955. Cachon, G. (2001). "Managing a retailer's shelf space, inventory and transportation." Manufacturing & Service Operations Management 3(3): 211219. Cachon, G. and M. Fisher (2000). "Supply chain inventory management and the value of shared information." Management Science 46(8): 10321048. Cadeaux, J. M. (1994). "Flexibility and performance of branch store stock plans for a manufacturer's product line." Journal of Business Research 29(3): 207218. Campbell, D. T. and D. W. Fiske (1959). "Convergent and discriminant validation by the multitrait-multimethod matrix." Psychological Bulletin 56: 81-105. Carter, M. W. and C. C. Price (2001). Operations research: a practical introduction. Boca Raton, CRC Press. Chagu, V. (1996). "Grer la technologie dans les PME." Direction Gestion Enterprises 1996(157): 1321. Chan, Y. and S. Huff (1993). Investigating information systems strategic alignment. Proceedings of the 14th international conference on information systems, Orlando. Chang, Y. S. (1967). "Reorder point models with non-decreasing stock-out costs." Journal of Industrial Engineering 18(6): 365374. Chase, C. (1995). "The realities of business forecasting." Journal of Business Forecasting 14(1): 2 and 26. Chen, F., Z. Drezner, et al. (2000). "Quantifying the bullwhip effect in a supply chain: the impact of forecasting, lead time and information." Management Science 46(3): 436443. Chircu, A. M. and R. J. Kauffman (2000). "Limits to value in electronic commercerelated IT investments." Journal of Management Information Services 17(2): 5980. Chow, G., T. D. Heaver, et al. (1994). "Logistics performance: definition and measurement." International Journal of Physical Distribution & Logistics Management 24(1): 1728.

194

8. Appendix and References

Chow, G. and L. E. Henriksson (1993). A critique of survey research in logistics. Working paper 93, Faculty of Commerce and Business Administration, University of British Columbia. Christopher, M. (1998). Logistics and supply chain management. London, Pitman Publishing. Ciborra, C. U. (2000). From control to drift. Oxford, Oxford University Press. Closs, D. J., A. S. Roath, et al. (1998). "An empirical comparison of anticipatory and response-based supply chain strategies." International Journal of Logistics Management 9(2): 2134. Coca-Cola Retailing Research Council (1994). "Measured marketing: a tool to shape food store strategy." Connell, J. (2001). "Influence of firm size on organizational culture and employee morale." Journal of Management Research 1(4): 220232. Corsten, D. (2002). Efficient Consumer Response adoption: theory, model and empirical results. Habilitation, University of St. Gallen. Corsten, D. and T. Gruen (2004). "Stock-outs cause walkouts." Harvard Business Review 82(5): 2627. Corstjens, M. and P. Doyle (1981). "A model for optimizing retail space allocations." Management Science 27(7): 822833. Cottrill, K. (1997). "Reforging the supply chain." The Journal of Business Strategy 18(6): 3539. Coughlan, P. and D. Coghlan (2002). "Action research for operations management." International Journal of Operations & Production Management 22(2): 220240. Cox, K. K. (1970). "The effect of shelf space upon sales of branded products." Journal of Marketing Research 7(February): 5558. Curhan, R. (1972). "The relationship between shelf space and unit sales." Journal of Marketing Research 9(November): 406412. Dagnoli, J. (1987). "Impulse governs shoppers." Advertising Age(11): 93. Dalrymple, D. J. (1964). "Controlling retail inventories." Journal of Retailing 40(1): 9 16. Daugherty, P. J., M. B. Myers, et al. (1999). "Automatic replenishment programs: an empirical examination." Journal of Business Logistics 20(2): 6372. Davenport, T. (2004). "Decision evolution; Automated systems are helping businesses make decisions more productively and consistently." CIO 18(1): 1.

8. Appendix and References

195

Davis, T. (1993). "Effective supply chain management." Sloan Management Review 4(34): 3547. DeHoratius, N. and A. Raman (2002). Inventory record inaccuracy: an empirical analysis. Working paper, Harvard Business School. DeHoratius, N. and A. Raman (2005). "Retail performance improvement through appropriate store manager incentive design: an empirical analysis." Unpublished paper, University of Chicago. Densley, B. (1999). The magnificent seven: getting the biggest bang from the ERP buck. In Proceedings of the first international workshop EMRPS99. J. Eder, N. Maiden and M. Missikoff. Rome, Istituto de Analisi dei Sistemi ed Informatica: 5965. Denzin, N. K. (1978). The research act. A theoretical introduction to sociological methods. New York, McGraw-Hill Company. Denzin, N. K. (1989). The research act: A theoretical introduction to sociological methods. Englewoods Cliffs, NJ, Prentice Hall. Dewan, S. and C. Min (1997). "The substitution of IT for other factors of production: a firm-level analysis." Management Science 43(12): 16601675. Dey, I. (1993). Qualitative data analysis: an user-friendly guide for social scientists. London, Routledge. Drazin, R. and A. H. van de Ven (1985). "Alternative forms of fit in contingency theory." Administrative Science Quarterly 30(4): 514539. Dreyfus, H. and S. Dreyfus (1986). Mind over machine. New York, Free Press. Drze, X., S. J. Hoch, et al. (1994). "Shelf management and space elasticity." Journal of Retailing 70(4): 301326. Dubelaar, C., G. Chow, et al. (2001). "Relationships between inventory, sales and service in a retail chain store operation." International Journal of Physical Distribution & Logistics Management. 31(2): 96-108. Duffy, M. (2004). "How Gillette cleaned up its supply chain." Supply Chain Management Review 8(3): 2027. ECR Europe (2000). Day-to-day category management. Facilitated by Andersen Consulting, Brussels. ECR France (2002). Optimal shelves availability. Presentation by the ECR group Taux de service en linaire. Glasgow.

196

8. Appendix and References

Eichen, B. F. and S. F. Eichen (2004). dm-drogerie markt: Vendor Managed Inventory. In Supply Chain Management erfolgreich umsetzen. Grundlagen, Realisierung und Fallstudien. D. Corsten and C. Gabriel. Berlin, Springer: 175195. Eisenhardt, K. M. (1989). "Building theories from case study research." Academy of Management Review 14(4): 532550. Ellinger, A. E., J. C. Taylor, et al. (1999). "Automatic replenishment programs and level of involvement: performance implications." International Journal of Logistics Management 10(1): 2536. Emmelhainz, M. A., J. R. Stock, et al. (1991). "Consumer responses to retail stockouts." Journal of Retailing 67(2): 138147. European Logistics Association and A.T. Kearney (1999). Insight to impact: results of the fourth quinquennial European logistics study. Brussels. Fafchamps, M., J. W. Gunning, et al. (2000). "Inventories, liquidity and contractual risk in African manufacturing." The Economic Journal 110(466): 861893. Fahrni, F. (1998). TQM in einem technischen Konzern: Unternehmenskultur im Wandel. In Qualittsmanagement an der Schwelle zum 21. Jahrhundert. R. Boutellier and W. Masing. Mnchen, Hanser. Fairris, D. (2004). "Towards a theory of work intensity." Eastern Economic Journal 30(4): 587602. Falck, M. (2005). "Automated store-ordering dream can come true." Supply Chain Technology Software(July/August): 1012. Feitzinger, E. and H. L. Lee (1997). "Mass customization art." Harvard Business Review 75(1): 116121. Fiedler, F. E. (1967). A theory of leadership effectiveness. New York, McGraw Hill. Fiorito, S. S., E. G. May, et al. (1995). "Quick response in retailing: components and implementation." International Journal of Retail & Distribution Management 23(5): 1221. Fischhoff, B. (1994). "What forecasts (seem to) mean." International Journal of Forecasting 10(3): 387404. Fisher, M. (2004). "To me it's a factory. To you it's a store." ECR Journal. International Commerce Review 4(2): 818. Fisher, M. L., J. H. Hammond, et al. (1994). "Making supply meet demand in an uncertain world." Harvard Business Review 72(3): 8393.

8. Appendix and References

197

Fisher, M. L., A. Raman, et al. (2000). "Rocket science retailing is almost hereare you ready?" Harvard Business Review 78(4): 115124. Fleck, J. (1994). "Learning by trying: the implementation of configurational technology." Research Policy 23(6): 632652. Fleisch, E. and C. Tellkamp (2004). "Inventory inaccuracy and supply chain performance: a simulation study of a retail supply chain." International Journal of Production Economics 95(3): 373385. Friedli, T., S. Billinger, et al. (2005). "Combining action research, grounded theory and case study research with long term projects." Unpublished paper, University of St.Gallen. Galliher, H. P., P. M. Morse, et al. (1959). "Dynamics of two classes of continuousreview inventory." Operations Research 7(3): 362384. Gassmann, O. (1999). "Praxisnhe mit Fallstudienforschung." Wissenschaftsmanagement 6(3): 1116. Gausemeier, J. and G. Gehnen (1998). "Intelligent material flow by decentralized control networks." Journal of Intelligent Manufacturing 9(2): 141146. Glaser, B. and A. Strauss (1967). "The discovery of grounded theory: strategies of qualitative research." 3(2): 269270. Gtz, N. D. (1999). Automatische Dispositionssysteme fr den Handel. Kln, Josef Eul. Grocery Manufacturers of America and Roland Berger (2002). "Full-shelf satisfaction. Reducing out-of-stocks in the grocery channel. An in-depth look at DSD categories." Gronau, N. (2004). Enterprise resource planning und supply chain management. Mnchen, Oldenbourg Wissenschaftsverlag. Groote, X. d. (1994). Inventory theory: a road map. Unpublished INSEAD teaching note. Gruen, T., D. Corsten, et al. (2002). Retail out-of-stock. Research Report, Grocery Manufacturers of America. Gudehus, T. (2001). "Optimaler Nachschub in Versorgungsnetzen." Logistik Management 3(2/3). Gudehus, T. (2002). Dynamische Disposition. Berlin, Springer. Gudehus, T. (2004). Logistik Grundlagen, Strategien, Anwendungen. Heidelberg, Springer Verlag.

198

8. Appendix and References

Guptill, A. and J. L. J.L. Wilkens (2002). "Buying into the food system: trends in food retailing in the US and implications for local foods." Agriculture and Human Values 19(1): 3951. Gurbaxani, V. and S. Whang (1991). "The impact of information systems on organizations and markets." Communications of the ACM 34(1): 5973. Hanke, J. E., D. W. Wichern, et al. (2005). Business forecasting. Upper Saddle River, Prentice Hall. Hines, P., M. Holweg, et al. (2000). "Waves, beaches, breakwaters and rip currents a three-dimensional view of supply chain dynamics." International Journal of Physical Distribution & Logistics Management 30(10): 827846. Hollinger, R. C. and J. L. Davis (2002). National retail security survey 2001. University of Florida, Department of Sociology and the Center for Studies in Criminology and Law. Hong, K. K. and Y. G. Kim (2002). "The critical success factors for ERP implementation: an organizational fit perspective." Information & Management 40(1): 2540. Hopp, A. and G. Arminger (2005). Kundenorientiertes Demand Chain Management bei Metro: Nutzung von Verkaufsdaten zur Steuerung der Hersteller. Presentation at the seminar Innovatives Replenishment in der KonsumgterDistribution, Horn. Horst, F. (2005). "Weniger Verluste in 2004." Retail Technology Journal(3). Hoyer, W. D. (1984). "An examination of consumer decision making for a common repeat purchase product." Journal of Consumer Research 11(3): 822831. Huber, G. P. (1984). "The nature and design of the post-industrial organization." Management Science 30(8): 928951. Hui, C., S. Lam, et al. (2001). "Can good citizens lead the way in providing quality service? A field quasi experiment." Academy of Management Journal 44(5): 988995. Inderfurth, K. and S. Minner (1998). "Safety stocks in multi-stage inventory systems under different service measures." European Journal of Operations Research 106(1): 5773. Jain, C. L. (2003). "Forecasting errors in the consumers products industry." The Journal of Business Forecasting Methods & Systems 22(2): 24.

8. Appendix and References

199

Jick, T. D. (1979). "Mixing qualitative and quantitative methods: triangulation in action." Administrative Science Quarterly 24(4): 602-611. Johansson, U. (2002). "Information technology in supplier-retailer relationships." Working paper, Institute of economic research, Lund University. Kahn, K. B. and J. T. Mentzer (1996). "Logistics and interdepartmental integration." International Journal of Physical Distribution & Logistics Management 26(8): 614. Kallinikos, J. (2004). "Deconstructing information packages. Organizational and behavioural implications of ERP systems." Information Technology & People 17(1): 830. Karrer, M., T. Placzek, et al. (2004). "Einsatz strategieorientierter Steuerungsinstrumente in der Logistik. Die Logistik-BSC am Beispiel groer Handelsunternehmen." Controlling 16(8-9): 503510. Katsikea, E. S. and D. A. Skarmeas (2003). "Organisational and managerial drivers of effective export sales organisations. An empirical investigation." European Journal of Marketing 37(11/12): 17231745. Keh, H. T. and S. Y. Park (1997). "To market, to market: the changing face of grocery retailing." Long Range Planning 30(6): 836846. Ketokivi, M. A. and R. G. Schroeder (2004). "Strategic, structural contingency and institutional explanations in the adoption of innovative manufacturing practices." Journal of Operations Management 22(1): 6389. Ketzenberg, M., R. Metters, et al. (2000). "Inventory police for dense retail outlets." Journal of Operations Management 18(3): 303316. Kieser, A. (2002). Organisationstheorien. Stuttgart, W. Kohlhammer. Klaus, P. (2003). Die wahren Kosten der Logistik. URL: http://www.logistikinside.de/sixcms/list.php?page=pg_archiv_heftuebersicht&skip=20, accessed 01.09.2005. Kling, R. (1996). Computerization and controversy. San Diego, Academic Press. Krber, H. J. (2003). Die Zukunft des Handels. Presentation at the 20th BVL Congress, Berlin. Kotler, P. and F. Biemel (1999). Marketing-Management. Analyse, Planung, Umsetzung und Steuerung. Stuttgart, Schffer-Poeschel. Kotzan, J. A. and R. U. Evanson (1969). "Responsiveness of drug store sales to shelf space allocations." Journal of Marketing Research 6(November): 465469.

200

8. Appendix and References

Kowalski, M. (1992). Qualitt in der Logistik. Wiesbaden. Krafcik, J. (1988). "Triumph of the lean production system." Sloan Management Review 30(1): 4152. Krane, S. D. (1994). "The distinction between inventory holding and stockout costs: implications for target inventories, asymmetric adjustment and the effect of aggregation on production smoothing." International Economic Review 35(1): 117136. Kromrey, H. (2002). Empirische Sozialforschung Modelle und Methoden der Datenerhebung und Auswertung. Wiesbaden, Opladen: Leske und Budrich. Krueckenberg, H. F. (1969). The significance of consumer response to display space reallocation. Proceedings of the Fall Conference, American Marketing Association. Krumbholz, M., J. Galliers, et al. (2000). "Implementing enterprise resource planning packages in different corporate and national cultures." Journal of Information Technology 15(4): 267280. Kuk, G. (2004). "Effectiveness of vendor-managed inventory in the electronics industry: determinants and outcomes." Information & Management 41(5): 645 654. Kumar, V., B. Maheshwari, et al. (2001). "An investigation of critical management issues in ERP implementation: empirical evidence from Canadian organizations." Technovation 23(10): 793807. Kurt Salmon Associates (1993). Efficient Consumer Responseenhancing consumer value in the grocery industry. Washington. Lamnek, S. (1993). Qualitative Sozialforschung. Weinheim, Beltz PsychologieVerlags Union. Lawrence, P. R. and J. W. Lorsch (1967). Organization and environment. Cambridge, Homewood. Lederer, A. L. and A. L. Mendelow (1990). "The impact of the environment on the management of information systems." Information Systems Research 1(2): 205222. Lee, H. L. (2001). Ultimate enterprise value creation using demand based management. Stanford Global Supply Chain Management Forum. Lee, H. L. and C. Billington (1992). "Managing supply chain inventory: pitfalls and opportunities." Sloan Management Review 33(3): 6573.

8. Appendix and References

201

Lee, H. L. and C. Billington (1993). "Material management in decentralized supply chains." Operations Research 41(5): 835847. Lee, H. L., V. Padmanabhan, et al. (1997a). "The bullwhip effect in supply chains." Sloan Management Review 38(3): 93102. Lee, H. L., V. Padmanabhan, et al. (1997b). "Information distortion in a supply chain: the bullwhip effect." Management Science 43(4): 546558. Leidner, D. and S. Jarvenpaa (1995). "The use of information technology to enhance management school education: a theoretical view." MIS Quarterly 19(3): 265 291. Lindman, R. H. (1992). Analysis of variance in experimental design. New York, Springer. Malone, T. (1997). "Is empowerment just a fad?" Sloan Management Review 38(2): 4553. March, J. G. and J. P. Olsen (1989). Rediscovering institutions. London, Free Press. Markus, L. M., S. Axline, et al. (2000). "Learning from adopter's experiences with ERP: Problems encountered and success achieved." Journal of Information Technology 15(4): 245265. Markus, M. L. and T. Cornelius (2000). The enterprise systems experiencefrom adoption to success. In Framing the domains of IT research: glimpsing the future through the past. R. W. Zmud. Cincinnati, Pinnaflex Educational Resources. Marshall, C. and G. B. Rossmann (1995). Designing qualitative research. Newbury Park, SAGE. Mau, M. (2000). Supply Chain durch Management. Realisierung von

Wertschpfungspotentialen

ECR-Kooperation

zwischen

mittelstndischer Industrie und Handel im Lebensmittelsektor. Frankfurt (Main), Fachverlag Moderner Wirtschaft. McGrath, J. (1982). Dilemmatics: the study of research choices and dilemmas. In Judgement calls in research. J. E. McGrath, J. Martin and R. A. Kulka. Newbury Park, Sage: 69102. Mentzer, J. T. and K. B. Kahn (1995). "Forecasting technique familiarity, satisfaction, usage, application." Journal of Forecasting 16(1): 465476. Merkel, H. (2003). "Standard-Warenwirtschaftssysteme fr den mittelstndischen Einzelhandel." Lebensmittelzeitung 55(25): 95-100.

202

8. Appendix and References

Metaxiotis, K. S., J. E. Psarras, et al. (2003). "Production scheduling in ERP systems. An AI-based approach to face the gap." Business Process Management Journal 9(2): 221247. Millet, I. (1994). "A novena to Saint Anthony, or how to find inventory by not looking." Interfaces 24(4): 6975. Moon, M. A., J. T. Mentzer, et al. (1998). "Seven keys to better forecasting." Business Horizons 41(5): 4453. Mumford, L. (1934). Technics and civilization. London, Harvest/HBJ. Myers, M. B., P. J. Daugherty, et al. (2000). "The effectiveness of automatic inventory replenishment in supply chain operations: antecedents and outcomes." Journal of Retailing 76(4): 455-481. Noble, D. (1984). Forces of production: a social history of industrial automation. New York, Alfred A. Knopf. O'Leary, D. E. (2000). Enterprise Resource Planning systems: systems, life cycle, electronic commerce and risk. Cambridge, University Press. Perrow, C. (1967). "A framework of comparative analysis of organizations." American Sociological Review 32(2): 194208. Petrak, L. (2005). "Dating Game." National Provisioner 219(6): 7852. Pfeffer, J. and H. Leblebici (1971). "Information technology and organizational structure." Pacific Sociological Review 20(2): 241261. Pfohl, H. C. (2004). Logistiksysteme. Berlin, Springer. Pfohl, H.-C. and W. Zllner (1987). "Organization for logistics: the contingency approach." International Journal of Physical Distribution & Materials Management 17(1): 316. Progressive Grocer (1968). "The out-of-stock study." 116. Ptak, C. A. and E. Schragenheim (2000). ERP: tools, techniques and applications for integrating the supply chain. London, St. Lucie Press/APICS. Punch, K. F. (1998). Introduction to social researchquantitative and qualitative approaches. London, SAGE Publications. Raman, A. (2000). "Retail-data quality: evidence, causes, costs and fixes." Technology in Society 22(1): 97109. Raman, A., DeHoratius, et al. (2001a). "The missing link in retail operations." California Management Review 43(3): 136152.

8. Appendix and References

203

Raman, A., N. DeHoratius, et al. (2001b). "The Achilles' heel of supply chain management." Harvard Business Review 79(5): 25-28. Rao, P. V. (1998). Statistical research methods in the life sciences. Pacific Grove, Brooks/Cole Publishing Company. Ritzman, L. P. and B. E. King (1993). "The relative significance of forecast errors in multistage manufacturing." Journal of Operations Management 11(1): 5163. Roland Berger (2003a). 2003 GMA Logistics Study. GMA publication. Roland Berger (2003b). ECROptimal shelf availability. ECR Europe publication. Sabath, R. E., C. W. Autry, et al. (2001). "Automatic replenishment programs: the impact of organizational structure." Journal of Business Logistics 22(1): 91105. Sabherwal, R. and L. Vijayasarathy (1994). "An empirical investigation of the antecedents of telecommunication-based interorganizational systems." European Journal of Information Systems 3(4): 268284. Sackett, P. R. and J. R. Larson (1990). Research strategies and tactics in industrial and organizational psychology. In Handbook of industrial and organizational psychology. M. D. Dunnette and L. M. Houghs. Palo Alto, Consulting Psychology Press. Sarpola, S. (2003). Enterprise Resource Planning (ERP) software selection and success of acquisition process in wholesale companies. Working paper, Helsinki School of Economics. Sawyer, S. and R. Southwick (2002). "Temporal issues in information and communication technology enables organizational change: evidence from an enterprise systems implementation." Information Society 18(4): 263280. Scandura, T. A. and E. A. Williams (2000). "Research methodology in management: current practices, trends and implications for future research." Academy of Management Journal 43(6): 12481264. Schneckenburger, T. (2000). Prognosen und Segmentierung in der Supply Chain. Ein Vorgehensmodell zur Reduktion der Unsicherheit. Dissertation, University of St.Gallen. Schonberger, R. I. (1982). Japanese Manufacturing Techniques. New York, The Free Press.

204

8. Appendix and References

Schoonhoven, C. B. (1981). "Problems with contingency theory: testing assumptions hidden within the language of contingency 'theory'." Administrative Science Quarterly 26(3): 672693. Schumpeter, J. A. (1935). Theorie der wirtschaftlichen Entwicklung: eine

Untersuchung ber Unternehmergewinn, Kapital, Kredit, Zins und den Konjunkturzyklus. Mnchen, Duncker & Humblot. Shalla, G. (2005). Supply Chain- und Beschaffungsmanagement im Handel am Beispiel der Globus-Gruppe (WND). Presentation at the University of St.Gallen. Sheeber, L., E. Sorensen, et al. (1996). "Data analytic techniques for treatment outcome studies with pretest/posttest measurements: an extensive primer." Journal of Psychiatric Research 30(3): 185199. Sheer, A. W. and F. Habermann (2000). "Enterprise Resource Planning: Making ERP a success." Communications of the ACM 43(4): 5761. Silver, E., D. Pyke, et al. (1998). Inventory management and production planning and scheduling. New York, John Wiley & Sons. Sloot, L., P. C. Verhoef, et al. (2002). The impact of brand and category characteristics on consumer stock-out reactions. Report, Erasmus Research Institute of Management. Smros, J., A. Angerer, et al. (2004a). Logistics processes of European grocery retailers. Unpublished research project report. Smros, J., A. Angerer, et al. (2004b). Retailer views on forecasting collaboration. Logistics Research Network (LRN), Dublin. Smith, H. W. (1975). Strategies of social research: the methodological imagination. Englewood Cliffs, Prentice-Hall. Soh, K., S. S. Kien, et al. (2000). "Cultural fits and misfits: is ERP an universal solution?" Communications of the ACM 43(4): 4751. SPSS Inc. (2003). SPSS Advance Models 12.0, SPSS Handbook. Staehle, W. H. (1991). Management. Mnchen, Franz Vahlen. Stank, T. P., P. J. Daugherty, et al. (1999). "Collaborative planning: supporting automatic replenishment programs." Supply Chain Management 4(2): 7785. Sterns, K. M., L. S. Unger, et al. (1981). Intervening variables between satisfaction/dissatisfaction and retail re-patronage intention. In An assessment

8. Appendix and References

205

of marketing thought and practice: 1982 Educators' Conference Proceedings. B. J. Walker. Chicago, American Marketing Association: 179182. Stlzle, W. (2004). "Supply Chain Controlling und Performance Management Konzeptionelle Herausforderungen fr das Supply Chain Management." Logistik Management 4(3): 1021. Stlzle, W., K. F. Heusler, et al. (2001). "Die Integration der Balanced Scorecard in das Supply Chain Management- Konzept (BSCM)." Logistik Management 3(2 3): 7385. Stlzle, W., K. F. Heusler, et al. (2004). Erfolgsfaktor Bestandsmanagement. Konzept Anwendung Perspektiven. Zrich, Versus. Stlzle, W. and T. Placzek (2004). Umsetzung von Optimal Shelf Availability Messkonzepte und Standardisierungspotenziale. Presentation at the BVL congress. Stratopoulos, T. and B. Dehning (2000). "Does successful investment in information technology solve the productivity paradox?" Information & Management 38(2): 103117. Taylor, J. C. and S. E. Fawcett (2001). "Retail on shelf performance of advertised items: an assessment of supply chain effectiveness at the point of purchase." Journal of Business Logistics 22(1): 7389. Tellkamp, C., A. Angerer, et al. (2004). "Improving data quality and supply chain performance with RFID: evidence from the retail industry." IT Cutter 17(9): 19 24. Tidd, J. (1993). "Technological innovation, organizational linkages and strategic degree of freedom." Technology Analysis & Strategic Management 5(3): 273 284. Toporowski, W. and B. Herrmann (2003). "Neue Entwicklungen in der Logistik Konzepte der Praxis und Beitrge der Wissenschaft." Handel im Fokus, EHI Newsletter(II). Ulrich, H. (1981). Die Betriebswirtschaftslehre als anwendungsorientierte Sozialwissenschaft. In Die Fhrung des Betriebes. M. N. Geist, R. Khler and C. Sandig. Stuttgart, Poeschel: 126. Ulrich, H. (2001). Systemorientiertes Management Das Werk von Hans Ulrich. Bern, Paul Haupt.

206

8. Appendix and References

van Donselaar, K., T. van Woensel, et al. (2004). Improvement opportunities in retail logistics. In Consumer driven electronic transformation: applying new technologies to enthuse consumers. G. I. Doukidis and A. P. Vrechopolous. Berlin, Springer. Vessey, I. (1991). "Cognitive fit: a theory based analysis of the graphs versus tables literature." Decision Science 22(2): 219240. Vessey, I. and D. Galleta (1991). "Cognitive fit: an empirical study of information acquisition." Information Systems Research 2(1): 6384. Vollmann, T. E., W. L. Berry, et al. (1992). Manufacturing planning and control systems. Homewood, Irwin. Vossen, M. (2000). Warenwirtschaft. Meilenweit vom Ziel entfernt. Der Handel. 05: 36. Wacker, J. G. and R. R. Lummus (2002). "Sales forecasting for strategic resource planning." International Journal of Operations & Production Management 22(9/10): 10141031. Wagner, H. M. (2002). "And then there were none." Operations Research 50(1): 217 226. Watson, E. E. and H. Schneider (1999). "Using ERP in education." Communications of the AIS 1(9): 1-48. Webb, E. J., D. T. Campbell, et al. (1966). Unobtrusive measures: nonreactive research in the social sciences. Chicago, Rand McNally. Weder, M. (2005). Die Logistik der Nestl Deutschland. Effiziente ReplenishmentStrategien zur Erfllung der Kunden- und Produktanforderungen. Speech at the conference "Innovatives Replenishment in der Konsumgter-Distribution", Bad Horn. Wegener, D. (2002). On-shelf-availability. Extent, causes and consequences. Unpublished diploma thesis, University of St.Gallen. Womack, J. P. and D. T. Jones (1996). Lean thinking. New York, Simon and Schuster. Woods, J. (2002). "SCM5 will drive the next wave of supply chain advantage." Gartner research note. Yin, R. K. (1988). Case study research: design and methods. London, Newbury Park. Zinn, W. and P. C. Liu (2001). "Consumer response to retail stockouts." Journal of Business Logistics 22(1): 4971.

8. Appendix and References

207

Zinn, W., J. T. Mentzer, et al. (2002). "Customer-based measures of inventory availability." Journal of Business Logistics 23(2): 1944. Zomerdijk, L. G. and J. d. Vries (2002). "An organizational perspective on inventory control: theory and a case study." International Journal of Production Economics 81/82(3): 173183. Zuboff, S. (1988). In the year of the smart machine: the future of work and power. New York, Basic Books.

208

8. Appendix and References

8.3. List of Interviews111

Company Athleticum Sportmarkets Athleticum Sportmarkets Athleticum Sportmarkets BonAppetitGroup Camion Tranport Wil COOP COOP COOP Name Markus Scheffler Rolf Engler Urs Wettstein Dr. Christian Kubik Hanspeter Imbach Martin Badertscher Hanspeter Drig Stefan Gchter Position Director logistics Vice-store manager Product manager Director supplier relationship management. Director logistics Store manager Director category management Director DC-Gossau Date Place

23.06.2003 Hnenberg 01.07.2003 St.Gallen 08.07.2003 Hnenberg 15.09.2003 Winterthur 13.12.2002 Wil 26.08.2003 Oberriet 12.09.2003 Gossau 17.02.2004 Gossau 17.06.2004 St.Gallen

Denner Denner Denner Gebrder Weiss Gebrder Weiss Interdiscount Interdiscount Interdiscount Intersport Germany Intersport International Intersport SFS Kaisers Tengelmann Kaisers Tengelmann Kaisers Tengelmann M-Electronics

Arthur Mathys Arthur Mathys Daniel Bucher Wolfgang Brunner Jrgen Glassner Hansjrg Kohli Urs Wilhelm Markus Fitzi Klaus Bader Raider Magnus

Director logistics Director logistics Store manager Information systems Head of purchasing, central Switzerland Director item management Director communications/PR Store manager Director logistics Director logistics Director/store manager Director/store manager Logistics manager Logistics manager Store manager Store manager

27.02.2004 Zurich 04.08.2004 Zurich 07.08.2004 Walenstadt 12.12.2002 Altenrhein 12.12.2002 Altenrhein 26.09.2003 Jegenstdorf 26.09.2003 Jegenstdorf 04.10.2003 St.Gallen 07.07.2003 Heilbronn 14.07.2003 Ostermundigen 09.07.2003 Kempten 01.07.2003 Widnau 04.12.2003 Mnchen 04.12.2003 Mnchen 04.12.2003 Mnchen 25.09.2003 St.Gallen

Intersport Magdon Friedrich Magdon Ernst Enz H. Warzecha H. Ushold H. Trkyilmaz Oliver Muggli

111

Some of the interviews with representatives from GROCO, MYFOOD, NORTHY and DRUCKS are omitted for confidentiality reasons.

8. Appendix and References

209

Metro Group Migros Migros Migros Migros PKZ group PKZ group PKZ group PKZ group SAF AG SAP AG

H. Orbach Ren Meyer Roland Studer Guido Federspiel Robert Vuilleumier Friedhelm Lange Irne Manser Moreno Papes Konrad Niederhusern Prof. Gerhard Arminger Alexander Kunkel

Logistics manager Store manager Director logistics Project-leader logistics Director logistics (Supply) Supply manager hardware Director logistics Store manager Director purchasing Store manager Scientific advisor Consultant Head of purchasing/ material planning Director logistics Store manager Regional director IT Director sourcing Switzerland Head of material planning Store manager DC director DC vice-director

11.03.2004 Dsseldorf 04.11.2003 Tuttlingen 05.09.2003 Zurich 22.09.2003 02.07.2004 Zurich

Metro Group/Extra H. Rapp

18.03.2004 Zurich 22.09.2004 Zurich 28.07.2003 Urdorf 12.08.2003 St.Gallen 20.08.2003 Urdorf 02.09.2003 Zurich 26.02.2004 Tgerwilen 09.03.2004 Suhr 18.03.2004 Regensdorf

Schild Spar Schweiz Spar Schweiz Spar Schweiz Valora Group Vgele Schweiz Vgele Schweiz Volg AG Volg AG

Thomas Herbert Egon Baumgrtner Kurt Walser Wolfgang Mhr Willy Schmid Bernhard Westphal Stefan Widmer Vogler S. Nf

29.08.2003 Luzern 11.09.2003 St.Gallen 15.09.2003 St.Gallen 16.02.2004 Gossau 17.09.2003 Muttenz 25.07.2003 Pfffikon 18.08.2003 St.Gallen 03.03.2004 Oberwinterthur 05.03.2004 Oberwinterthur

210

8. Appendix and References

8.4. Curriculum Vitae

Name Born Alfred Angerer 19th September, 1974 in Puebla (Mexico)

Education
Oct. 2002 Dec. 2005 Oct. 1995 Aug. 2001 University of St.Gallen (Switzerland) Doctoral student Karlsruhe University of Technology (Germany) Diplom Wirtschaftsingenieurwesen (Master in Industrial Engineering and Business Science) University of Edinburgh (Scotland) Honour courses in Department of Economics Bodensee Gymnasium, Lindau (Germany) Abitur (equivalent to A-levels)

Oct. 1999 June 2000 Sept. 1986 June 1995

Employment
Since Jan. 2006 Oct. 2002 Dec. 2005 Jan. 2002 Aug. 2002 July 2000 Sept. 2000 Oct. 1996 Dec. 1999 McKinsey & Company, Munich (Germany) Supply Chain Management consultant University of St.Gallen (Switzerland) Research associate Nestl Germany, Biessenhofen (Germany) Supply Chain Management trainee Lycos Europe-Bertelsmann, Guetersloh (Germany) Internship in the Business Development department WebDesign Angerer & Dolling GbR mbH, Karlsruhe (Germany) Independent management of an own internet company Broadgate Technology Services, Madras (India) Internship at software and internet-service provider Siemens AG, Karlsruhe (Germany) Internship in the department "Sales of Communication Systems and Networks"

May 1999 June 1999 July 1997 Sept. 1997