Post Graduate Diploma in Management

Future Managers Group Programme

(PGDM: FMG-26) 2017-19 Batch Term: III (Jan-Mar 2018)

Operations Management
Minor Project Weightage: 10%

Predictive Analytics model for Improving

Backorder Recovery
(Toolware CD Included)

OM
Minor Project

1
 ABSTRACT

Here in this project we have tried to understand the term backorders, their reasons, how they
affect the organization’s reputation and the organization’s profit generating potential. We have
conducted the literature review on the recent developments and publications in the field of
backorder management, involving the 3D printing technique in automobile sector and the use of
Machine Learning in reducing the number of backorders. We have further bifurcated this project
in two parts: in the first part we have tried to explain a case study of how Volvo is using 3D printing
to tackle the problem of the backorders and in the second part we have used in python to analyze
a data set of a hypothetical company which wants to accurately predict the backorders.

2
 TABLE OF CONTENTS

Sr. No. Topic Page No.

1 Introduction 4
2 Literature Review 5
3 Project Objectives 9
4 Framework Development 10
5 Issues in Organization 11
6 Framework Applied 12
7 Toolware Design 13
8 Conclusions and Recommendations 17
9 References 18

3
 INTRODUCTION

A backorder is an order for a good or a service which due to insufficient inventory supplies cannot
be met in the current time. Thus, a backorder represents an unfilled customer order or
commitment and is an out-of-stock product ordered, promised to be shipped when the product
becomes available. In the case of a backorder, the item might not be held in the company's
available inventory but could still be in production. Alternatively, the company might need to still
manufacture more of the product. The nature of the backorder and the number of items on
backorder will affect the amount of time it will take before the customer can eventually receive
the ordered product. Greater the number of items backordered, greater is the demand for the
item.
Backorder management forms an important subset of the inventory management analysis. The
percentage of items backordered and the number of backorder days are important measures of
the quality of a company's customer service and the effectiveness of its inventory management. A
company that consistently sees items in backorder, might be running too lean. Such a company
might be losing out on business by not providing the products demanded by its customers. This is
because when an item is on backorder, a customer might look somewhere else for a substitute
product, especially if the expected waiting time until the product becomes available is
considerably long. This can provide an opportunity for once loyal customers to try other
companies' products and potentially switch. Thus, it is a huge disadvantage in a highly competitive
industry, where the number of alternatives available are huge or in an industry where brand
loyalty is very low. Difficulties in proper inventory management leading to enormous back orders,
can eventually lead to loss of market share as customers become frustrated with the company's
lack of product availability. Moreover, customers waiting for their backordered products must
receive shipping notifications on the shipping date they’ve been provided. This means that
customers must be kept in the loop in case of any delays or changes, to avoid a storm of
complaints and help maintain a good reputation for the company.
A backorder cost is a cost incurred by a business when it is unable to fulfil an order and therefore,
must complete it later. It can be in discrete terms, as in the cost to replace a specific piece of
inventory, or intangible, such as the effects of poor customer service. Backorder costs are usually
computed and displayed on a per-unit basis. It is essential for companies to monitor the backorder
costs, as the relationship between storage costs of inventory and backorder costs will determine
whether a company should over- or under-produce. If the carrying cost of inventory is less than
Backorder costs (which is true in most of the cases), the company should over-produce and keep
an inventory.
In the automobile sector, the design and operation of spare part management systems is very
crucial to determine optimal inventory level in order to reduce costs. Fluctuating inventory levels
can dramatically alter the supply chain’s efficiency and responsiveness because inventory is an
important cross functional driver of supply chain performance. Further, efficient backorder
management ensures that the products are ready and available when the customer wants them
thereby reducing the customer waiting time in the service sector. It has been seen that in the
automobile sector, the costs incurred due to shortage or backorders are higher than the cost for
inventory and the regular order, since the failure has already taken place.

4
Thus, the goal is to minimize the sum of the expected backorders of all service parts at all local
bases by tracking both inventory levels and item velocity, making precise predictions about Low
Stock and Out of Stock levels, and using that information to trigger workflows so that inventory
never “runs dry.”

BACKORDER ANALYSIS
Backorder analysis aims to determine the reason due to which a particular stock part was not in
the bin when the customer wanted to buy it. Therefore, it seeks to explore the causes of
backorders and the remedies there might be used to minimize having backorders in the first place.
The main causes of backorders are:

1. Order not promptly placed:

Based on the vendor’s order cycle, stock orders are placed to the vendor daily, weekly, or
some other time period. The part is put onto a suggested order which then is normally
reviewed by an individual in the parts department and a decision made as to placing the
order or not. If the decision is made to delay placing a stock order, that might be the cause
of a backorder.

It is advisable that dealers never take a part number off of a suggested order. The rules for
inventory management should not be changed as changes can be an ongoing cause of a
backorder.

2. Warehouse discrepancies: A discrepancy occurs when the quantity on hand in the

computer is not the same as the quantity on hand in the bin. The first and most prominent
cause of this discrepancy is from a physical count itself.
3. Human error: The part is either in the bin location or in the warehouse in another location,
but somehow the person picking the order didn’t find it. Such an error is termed as human
error.
4. Factory shortages: Factory shortages lead to a lot of backorders and the dealer has no
control over them. This can be avoided by recognizing early when the vendor is running out
of a part. Apart from this, there is no mechanism to control the vendor performance on
stock orders.
5. Inaccurate order points: There are several considerations for this cause of a backorder. The
items making up the order point are not current or accurate. There is a seasonal upswing in
demand or the delivery cycle from the suppliers has changed and the production unit has
not adjusted their rules. All of these situations can be identified and corrected.

6. Abnormal demand: Such a situation arises if the customer places order for an item in a
quantity that does not follow the regular buying behaviour or pattern. In such a scenario,
the reason behind the abnormal demand must be figured out. Then together, between the
customer and the company’s customer service personnel, a solution can be found.

5
 LITERATURE REVIEW

BACKORDERS IN CASE OF TESLA

Tesla, Inc. (formerly Tesla Motors) is an American company, based in Palo Alto, California that
specializes in electric vehicles, energy storage and solar panel manufacturing. The company
operates multiple production and assembly plants, notably Gigafactory 1 near Reno, Nevada and
its main vehicle manufacturing facility at Tesla Factory in Fremont, California.

Tesla’s Model 3 is an ambitious car: It's the car meant to take electric cars mainstream—and Tesla
into the ranks of the big automakers. Tesla’s Model 3 was unveiled in March 2016 and had
accumulated 455,000 net reservations prior to the handover event held in July 28 2017 at Tesla’s
factory in Fremont, Calif. To reserve a Model 3, an electric vehicle with a base price of $35,000 and
220 miles of battery range, customers have to provide a $1,000 refundable deposit.

As on 3rd January 2018 it was reported that, Tesla had produced 2,425 of its new Model 3
electric cars in the fourth quarter and delivered 1,550, missing Wall Street expectations as the
company led by Elon Musk is constantly trying to solve production bottlenecks that have
forced the company to dial back its own mass manufacturing goals. The primary production
constraint, by far, is in battery module assembly. There are four "zones" in battery module
production for the Model 3, and one if the zones suffers from a severe glitch.

Despite making “major progress” addressing Model 3 production bottlenecks, the company
has pushed back its target of producing 5,000 Model 3 sedans per week to the end of the
second quarter of 2018. This was the second time that Tesla had punted on its 5,000 -per-
week milestone. Tesla said that it will have a more gradual ramp through the first quarter to
focus on quality and efficiency rather than simply pushing for the highest possible volume in
the shortest period of time. The company has been focusing on quality versus volume, which
is the right focus. The Model 3 must be right in terms of quality and thus ramping up
production levels with a flawed product will be a foolish idea.

According to an analyst: “It seems to me that people would be willing to wait since they were
never given a firm date when they would take delivery in the first place. The long wait seems
to be building anticipation.” This case shows that Backorders were working in favor of Tesla
Inc.

6
 Figure 1. Number of Reservations of Tesla’s Model 3 (Hull, 2017)

BACKORDERS AT VOLVO

The Volvo Group is a Swedish multinational manufacturing company headquartered

in Gothenburg, Sweden. While its core activity is the production, distribution and sale of trucks,
buses and construction equipment, Volvo also supplies marine and industrial drive systems and
financial services. In 2016, it was the world’s second largest manufacturer of heavy-duty trucks.

According to Carolina Corin (2016) Volvo defines a Backorder (BO) as an order for a part that
cannot be filled at the current time due to a lack of available supply. BO is a status of an order and
is defined by no instant availability in the storage shelves locations. BOs can be initiated in
different parts of the distribution chain. When the BO is initiated from the regional distribution
centers (RDCs), i.e. the article is below the order point, the BO is not considered being as critical as
the BOs that are initiated from dealers and those that are a direct result of a customer need. Volvo
has a 94% service level for delivering spare parts instantly from stock to their customers, which
means that Volvo plans for ~6% unavailability that results in either day orders or BOs.

A day order is an order which is not instantly available but can be delivered the same day, these
orders are not considered as BOs even if they are registered as such by the system. Approximately
1% of the orders become BOs. In total they have over 3000 dealers. In the distribution chain for
the aftermarket there are different distribution centers (DC); there are central distribution centers
(CDCs), RDCs and support distribution centers (SDCs).

3D PRINTING AT VOLVO
Presently, Volvo has incorporated 3-D printing in its manufacturing and production units for
customized tools. The expectations from 3-D printing include –
 Reduction on Cost of Production (by ~ 90%).

7
  Significant reduction in Lead Time.

Additionally, Volvo has successfully tested to use 3DP to solve BOs. Both tools and actual spare
parts have been printed. Within the Volvo Group there are several ongoing initiatives regarding
3DP. Today there are not many coordination activities between the different initiatives. However,
this has been noticed by the company and they have started mapping all 3D-activities within the
Volvo Group. Even so, the exact knowledge and expertise that the company contains is still not
completely mapped.

Volvo has taken huge Leaps to make 3-D printing for inventory management, a reality. Presently,
they have contacted with two global 3-D Printing Suppliers, have provided information regarding
both the current capacity and the future possibilities for the automotive aftermarket.

Volvo believes that though lead time is an issue that would require some patience to be taken
care of due to technology of 3-D printing in its infancy state, they can use this technology to take
care of the regular backorders, where lead time is not a major issue. So, though this technology
has its drawbacks for the automobile industry, continuous improvements going on every day in 3-
D printing could facilitate the process of Inventory management.

8
 PROJECT OBJECTIVES

The world has become very dynamic and the consumer demands have become dynamic too. The
demand for certain products is not based on just the specifications of the product, rather it is
dependent over several other factors like seasonality of demand, forecasts and many other factors.
For the producer or the manufacturer the major purpose is to improve the accuracy of forecasts so as
they know about certain factors which can help them to reduce costs. Further such predictions are
necessary to if we take into consideration the consumer satisfaction and demand, as the producers
or the manufacturers have to build a robust and fast supply chain.

FRAMEWORK DEVELOPMENT
It has been seen that many industries resort to global sourcing in order to reduce the overall costs.
The global sourcing generally reduces per unit costs, but the impact of global sourcing on total
costs throughout the supply chain often remains unrecognized. In particular, costs such as
purchase price and transportation are easily calculated and recorded. However, costs due to
quality issues, reverse logistics and particularly increased lead-time often are unmonitored.

Costs associated with lead-time can be difficult to quantify. In general, as lead-time grows, so does
lead-time variability, which negatively impacts forecast accuracy. As forecast accuracy worsens,
end product and component backorders tend to increase. Moreover, low forecast accuracy tends
to lead to increased system buffers in the form of inventory. The buffers due to increased lead-
time come in the form of safety stock and increased batch size to achieve economies of scale and
transportation. While the larger safety stocks and batch sizes keep companies at desired customer
service levels, the inventory and related costs grow throughout the supply chain.

Challenges with predicting backorders:

One of the first steps in developing a backorder strategy is to understand which products are at
risk of being backordered. This sounds easy but in reality is rather difficult to determine. A number
of factors can make this a challenging business problem.

Demand-event dependencies:
Demand for existing products are constantly changing, often dependent on non-business data
(holidays, weather, and other events). Consider demand for crackers or blankets at during Diwali
and winter season. The demand for blankets in Homecentre will increase if the temperature falls
very rapidly or how less the temperature is.

Small numbers (anomalies):

If backorders are very infrequent but highly important, it can be very difficult to predict the
minority class accurately because of the imbalance between backorders to non-backorders within

9
the data set. Accuracy for the model will look great, but the actual predictive quality may be very
poor.

Artificial intelligence and big data have disruptively changed the industry, as the barriers to their
implementation (cost, computing power and open-source platforms) disappear. In this context,
machine learning is applied on the design and development of predictive models which assess all
areas of management, providing essential insights for companies to understand and react to
changes in its operation.

A predictive analytics program can identify which products are most likely to experience
backorders giving the organization information and time to adjust. Machine learning can identify
patterns related to backorders before customers order. Production can then adjust to minimize
delays while customer service can provide accurate dates to keep customers informed and happy.
The predictive analytics approach enables the maximum product to get in the hands of customers
at the lowest cost to the organization. The result is a win-win: organizations increase sales while
customers get to enjoy the products they demand.

ISSUES IN THE ORGANIZATION

Case of Danish Breweries using Machine Learning for backorder management

From the Danish beer brewing industry background, the demand for craft beers is rising by years.
For producers, one of the primary purposes of supply chain collaboration is to improve the
accuracy of forecasts, especially international business. In this case, when the brewery needs to
collaborate with foreign distributors and retailers, many issues might appear due to the lack of
accurate predictions. In this case, considering customer satisfaction and demand, the producers
need to be well prepared to ensure a robust and fast supply chain. With the help of advanced
predictions, the business will have time to react before any unexpected event occurs and
therefore help the producers be more intelligent on stocking and become more proactive.

10
The above figure shows an example of the interaction of stakeholders in the supply chain. An
organisation can shift demand fulfilment to other periods by allowing back orders. That is, orders
are taken in one period, and deliveries promised for a later time. The success of this approach
depends on how willing customers are to wait for delivery. Moreover, the costs associated with
backorders can be difficult to pin down since they would include lost sales, annoyed or
disappointed customers, and perhaps additional paperwork

To avoid bullwhip effect, small brewers need to react fast and in the meantime avoid overreacting
when the market changes. By having a more advanced forecasting too, business can identify the
market more precisely where the predictions providing more evidence about what’s going to
happen.

They have relied on some variables that we will describe in the Framework Applied and upon
some observations many such companies are using similar variables in their inventory
management practices. So we have generalized the model used by Danish Breweries to a
hypothetical data set of a manufacturer and then we analyzed and discussed the deductions.

11
 FRAMEWORK APPLIED
We have used the Predictive analytics program with some elementary concepts of machine
learning on the data set that is given below. This model can be generalized to all the industries
which use the variables that are described below, in their inventory management.

The data set: A hypothetical manufacturer has a data set and the challenge is to accurately predict
future backorder risk using predictive analytics and machine learning and then to identify the
optimal strategy for inventorying products with high backorder risk.

To model and predict the target, we’ll use the other features, which include:
 sku - Random ID for the product
 national_inv - Current inventory level for the part
 lead_time - Transit time for product (if available)
 in_transit_qty - Amount of product in transit from source
 forecast_3_month - Forecast sales for the next 3 months
 forecast_6_month - Forecast sales for the next 6 months
 forecast_9_month - Forecast sales for the next 9 months
 sales_1_month - Sales quantity for the prior 1 month time period
 sales_3_month - Sales quantity for the prior 3 month time period
 sales_6_month - Sales quantity for the prior 6 month time period
 sales_9_month - Sales quantity for the prior 9 month time period
 min_bank - Minimum recommend amount to stock
 potential_issue - Source issue for part identified
 pieces_past_due - Parts overdue from source
 perf_6_month_avg - Source performance for prior 6 month period
 perf_12_month_avg - Source performance for prior 12 month period
 local_bo_qty - Amount of stock orders overdue
 deck_risk - Part risk flag
 oe_constraint - Part risk flag
 ppap_risk - Part risk flag
 stop_auto_buy - Part risk flag
 rev_stop - Part risk flag
 went_on_backorder - Product actually went on backorder. This is the target value.

12
 TOOLWARE DESIGN:
With the help of the above variables and a dataset regarding a hypothetical manufacturer we have
built a tool ware with elementary concepts of data analytics and machine learning. Please refer to
the PowerPoint for the working and results of the tool ware.

Analysis of the Data

1. The data are a mix of string and floating point values.

2. The SKU has a unique value for each row, so it is the index column and should be dropped.
3. The features with string values, apart from SKU, are categorical features that only contain
'yes' and 'no'. They can be changed to represent the same information in numerical values.
4. The numerical features have different scales, which may be a problem for some machine
learning algorithms. The features should be rescaled to have similar scale.
5. There are missing values in lead_time, perf_6_month_avg and perf_12_month_avg. These
missing values need to be replaced or the samples with missing values need to be
removed.
6. lead_time has (1,687,860 - 1,586,967 = 100,893) missing values, which is 5.9% of total rows
7. perf_6_month_avg has (1,687,860 - 1,558,382 = 129,478) missing values or 7.68% of total
rows
8. perf_12_month_avg has (1687860 - 1565810 = 122,050) or 7.23% of missing values
9. The last row is not a valid sample in both the training and testing datasets, and thus should
be dropped.
10. There are 1,676,567 samples when the product did not go on backorder. There are 11,293
(0.67%) samples when the product did go on backorder.

Histogram for performance

over 6 months skewed
towards right

13
 Histogram for lead time
skewed towards left

Histogram for performance

over 12 months skewed
towards right

Here to make the data consistent we replace the missing or NaN values by taking a median of the
values of the above 3 variables and then we fill those missing values with the median value.

Correlation plot to find the variables affecting the backorders

 The correlation matrix shows that the quantity in transit, the forecast sales over 3,6, 9
months, the actual sales over the previous 1,3,6, 9 months, and minimum recommended
stock level are highly correlated. It is evident from the fact that if an item had high real
sales over the last 1,3,6, 9 months, then it is reasonable for the forecast sales over the next
3,6, 9 months to also be high. If forecast sales are high, then it would be useful to have
more of the stock in hand and to have more shipped in.
 Besides that, the average performance over the last 6 months strongly correlates with
that over the last 12 months.
 Basically the correlation matrix points out that the number of features used for predicting
whether an item goes on back order can be lower than the number of features in the
dataset.

14
Relation of Backorders with Sales and forecast variables

Scatter Plots for

forecast variables
15
 Scatter Plots for
sales variables

The deductions from the above plots are as follows:

 The forecast values over each time frame have very close linear correlation with each
other, as expected from the correlation matrix. The forecast values cover a wide range
from 0 to over 1 million. Backorders only occur when the forecast value is low.
 The sales over each time frame have good linear correlations with each other, as expected
from the correlation matrix. There are some instances when the sales at different time
frames fall away from the linear correlation. The sales range from 0 to over 1 million.
Backorders only occur when sales are low.

16
CONCLUSION AND RECOMMENDATIONS

In today’s hypercompetitive world, where the number of players in the industry are huge; that is
the competition is very high and brand loyalty is diminishing, a backorder is more of a threat and
less of a potential opportunity for any firm. Thus, it is very crucial to make use of Data Analytics
and Machine Learning to accurately predict backorders.

 The toolware can be extended to determine future backorder risk using predictive analytics
and machine learning in any firm that deploys an inventory of raw materials, work in
progress and finished goods and which employs the variables that have been considered in
the tool ware. This will help in formulating the optimal strategy for inventory products with
high backorder risk.

 At the same time, some of the crucial variables considered in the toolware deal with values
that are futuristic in nature (e.g. forecast_3_month, the forecast sales for the next 3
months; forecast_6_month, the forecast sales for the next 6 months and
forecast_9_month, the forecast sales for the next 9 months). Since future is uncertain and
subject to variability, it is not advisable to completely rely on the toolware.

 Hence, it is recommended that firms exploit the latest technological advancements (e.g. 3D
printing) in their manufacturing and production processes. 3D printing ensures a faster
delivery time through on-demand and decentralized production strategies. Moreover, the
time saved in the prototyping stages when using 3D printing, reduces considerably
turnaround time across all subsequent stages of manufacturing.

 Many successful companies around the world, say Volvo have successfully tested to use
3DP to solve BOs. Both tools and actual spare parts have been printed using this
technology in Volvo. Other renowned companies that have resorted to this technology are
Airbus, BMW and Volkswagen Autoeuropa.

17
 REFERENCES

1. Website accessed: http://fortune.com/2017/08/02/tesla-model-3-orders/

Accessed on: 16th February 2018

2. Website accessed: http://fortune.com/2018/01/03/why-tesla-model-3-deliveries-missed-

the-mark/
Accessed on: 16th February 2018

3. Website accessed: https://www.businessinsider.in/tens-of-thousands-of-people-have-

canceled-their-tesla-model-3-orders/articleshow/59893865.cms
Accessed on: 16th February 2018

4. Website accessed: https://www.recode.net/2017/8/2/16087432/tesla-model-3-electric-

car-manufacture-preorder-cancellations-elon-musk
Accessed on: 18th February 2018

5. Website accessed: https://www.bloomberg.com/news/articles/2017-11-13/tesla-model-3-

depositors-staying-put-as-wait-in-line-lengthens
Accessed on: 18th February 2018

6. Website accessed: https://waterwelljournal.com/the-backorder-analysis/

Accessed on: 18th February 2018

7. Hull, D. (2017). Number of Reservations of Tesla’s Model 3.

Retrieved from: https://www.bloomberg.com/news/articles/2017-11-13/tesla-model-3-
depositors-staying-put-as-wait-in-line-lengthens

8. Website accessed:
https://www.engineering.com/3DPrinting/3DPrintingArticles/ArticleID/15943/Saving-Big-
Bucks-in-the-Automotive-Industry-with-3D-Printing.aspx
Accessed on: 18th February 2018

9. Website accessed: http://www.divbyz.com/industries/automotive

Accessed on: 18th February 2018

10. Master Thesis by Yuqi Li(2017), Backorder Prediction using Machine Learning Danish Craft
Beer Industry

11. Master Thesis by Almqvist Elise & Carolina Corin(2016), Using 3D printing for improving
Backorder Recovery: A study performed at Volvo group,

18