Sie sind auf Seite 1von 25

European Journal of Operational Research 125 (2000) 3458

www.elsevier.com/locate/dsw

Theory and Methodology

An evaluation of vendor selection models from a total cost of


ownership perspective
Zeger Degraeve, Eva Labro *, Filip Roodhooft
Department of Applied Economics, Katholieke Universiteit Leuven, Naamsestraat 69, 3000 Leuven, Belgium
Received 1 March 1998; accepted 1 March 1999

Abstract
Many dierent vendor selection models have been published in the purchasing literature. However there has been no
systematic approach to compare the relative eciency of the systems. In this paper we propose to use the concept of
Total Cost of Ownership as a basis for comparing vendor selection models. We illustrate the comparison with a real life
data set of the purchasing problem of ball bearings at Cockerill Sambre, a Belgian multinational company in the steel
industry. From a Total Cost of Ownership perspective mathematical programming models outperform rating models
and multiple item models generate better results than single item models for this specic case study. 2000 Elsevier
Science B.V. All rights reserved.
Keywords: Purchasing; Management accounting/Operations research

1. Introduction
In the literature (Dickson, 1966; Weber et al., 1991) several dimensions are mentioned that are important
for the multiple objective vendor selection decision. These include net price, quality, delivery, performance
history, capacity, communication system, service, geographical location, etc. The problem is how to select
suppliers that perform optimally on the desired dimensions. The published vendor selection decision models
formulate answers to this multiple objective problem. Some authors propose linear weighting models in
which suppliers are rated on several criteria and in which these ratings are combined into a single score.
Others propose mathematical programming formulations in which quantiable criteria are taken into
account. Some approach the problem on an item-by-item basis, others consider it a multiple item decision.
Only a few authors incorporate the multi-period inventory management issue into the supplier selection
decision (Bender et al., 1985; Degraeve and Roodhooft, 1999b; Ronen and Trietsch, 1988).

Corresponding author. Tel.: +32-16-32-6937; fax: +32-16-32-6732.


E-mail address: eva.labro@econ.kuleuven.ac.be (E. Labro).

0377-2217/00/$ - see front matter 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 7 - 2 2 1 7 ( 9 9 ) 0 0 1 9 9 - X

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

35

No research has been done on how to compare these dierent approaches to vendor selection and to nd
out the ``best'' way to handle the decision. The problem is to nd a basis for comparison that is theoretically
sound. In this paper we propose the concept of Total Cost of Ownership (TCO) to compare the relative
eciency of dierent vendor selection decision models. The Total Cost of Ownership quanties all costs
associated with the purchasing process throughout the entire value chain of the rm. We apply each of the
vendor selection models proposed in the literature to a real life data set describing the purchasing problem
of ball bearings at Cockerill Sambre. Subsequently, we calculate the TCO of the resulting solutions, i.e.
choices of what to buy from whom and when. In this way, we are able to evaluate and compare the existing
vendor selection models from a TCO perspective.
The contribution of this paper is fourfold. First, we present a literature review and classication of
published vendor selection models. Second, these models are evaluated from a TCO perspective. Third,
the application and evaluation of the dierent vendor selection models is done using a real life data set
describing the purchasing problem of ball bearings at Cockerill Sambre. The data consist of information
on various criteria other than the traditional quality, time and quantity discount parameters. Fourth,
several conclusions will be drawn regarding the eciency of the dierent kinds of vendor selection
models.
The rest of the paper is organised as follows. In Section 2, we explain the TCO perspective from which
the evaluation will be made and all parameters associated with it. We discuss the classication of the vendor
selection models in Section 3. In Section 4 we describe the real life case used to evaluate the published
vendor selection models. In Section 5 the results of the comparison will be discussed. Finally, we will draw
conclusions and make suggestions for future research in Section 6.
2. The total cost of ownership approach
The TCO quanties all costs associated with the purchasing process throughout the entire value chain of
the rm. The cost of the acquisition and subsequent use of an item or service that is to be purchased is
determined. The approach goes beyond price to consider all costs over the items entire life such as those
related to service, quality, delivery, administration, communication, failure, maintenance, etc. (Ellram,
1994, 1995b). The analysis of costs throughout the extended value chain of a company is an important topic
in todays management accounting literature (Shank and Govindarajan, 1992). Activity Based Costing
(ABC) permits us to analyse activities and determine cost drivers for the dierent activities dened. While
suppliers are an important part of the total value chain, the application of ABC ideas to the vendor selection problem has received little attention. Roehm et al. (1992) discuss the use of the system in a purchasing department. They assign additional purchasing costs to products, but not to suppliers. Ellram
(1995a) and Roodhooft and Konings (1997) develop the link between the selection of suppliers and ABC.
Also Carr and Ittner (1992), Cavinato (1992) and Ellram and Sierd (1993) elaborate on the use of the TCO
concept in purchasing.
Degraeve and Roodhooft (1999b) recognise a hierarchical structure in activities with respect to the
purchasing problem: (1) the supplier level activities, (2) the order level activities and (3) the unit level
activities. The rst hierarchical level describes costs incurred and conditions imposed whenever the
purchasing company actually uses the supplier over the decision horizon. Examples of costs on the
supplier level include a quality audit cost incurred by the buyer for the evaluation of a supplier, the cost
of a dedicated purchasing manager and additional research and development costs due to using a particular supplier. The order level parameters indicate costs incurred and conditions imposed each time an
order is placed with a particular supplier and include, amongst others, costs associated with reception,
invoicing, transportation, ordering and receiving credit notes. At the unit level we nd costs incurred and
conditions imposed related to the units of the products for which the procurement decision has to be

36

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

made, for example, price, internal failure, external failure and inventory holding. It is important to make
this classication of activities into separate levels since the overall cost driver (number of suppliers,
number of orders, number of units procured) for each level of activity is independent of the activities in
other levels.
The use of ABC in supplier selection models has several advantages. First, it is important to note that
the quantication of the criteria and the trade-o between them is no longer a problem, because the objective function is dened as the TCO with respect to the purchasing decision caused by the suppliers.
Second, an important advantage of this approach over other methodologies exists in arriving at objective
cost measures in a systematic way. Third, the system will enable companies to develop interorganisational
activity based management opportunities given the importance of close relationships between the purchaser
and a limited number of reliable suppliers. Fourth, the model allows us to answer all sorts of ``what if''
questions dealing with cost management and strategic decision making such as (1) the cost impact of
making dierent/alternative supplier selections, (2) the consequences of performance improvement by
suppliers with respect to dierent important criteria and the reduction or elimination by the purchasing
company of some of the costs or activities caused by the purchasing decision and (3) the evaluation of
alternative company policies with respect to the number of suppliers, order quantities and minimum and/or
maximum quantities to buy. A disadvantage of the approach is that determining the TCO of selecting a
supplier for the delivery of a certain item based on ABC information requires an extensive management
accounting system that captures the relevant costs of the activities by supplier and item purchased. We
recognise the importance of the incremental costs of developing, installing and maintaining such an ABC
system. Experience shows however (Turney, 1990) that most companies overestimate these costs and that
several rms have implemented ABC systems at a cost that was acceptable for management. Recent
management accounting literature (Kaplan and Cooper, 1997) states that rms start to and should integrate comprehensive ABC systems with enterprise-wide information systems from SAP, Oracle, PeopleSoft,
Baan, and others, which makes it easy to collect the necessary data. However, it is not necessary to implement a company-wide ABC system to provide decision support for vendor selection. Only information
on activities relevant to the purchasing function has to be gathered. The benets of adopting this approach
outweigh the cost. For example, Degraeve and Roodhooft (1998,1999a) succeed in saving more than 11%
on TCO for two dierent product groups at Cockerill Sambre. This resulted in the development of a
company-wide purchasing system based on ABC and TCO information.
In the following sections we discuss the dierent vendor selection models more thoroughly and evaluate
them from a TCO viewpoint whenever the Cockerill Sambre ball bearings data set allows us to do so. First,
we have applied the dierent vendor selection models to the data set using the advocated methodology
while trying to stay as close as possible to the philosophy of the authors. Then the TCO resulting from the
vendor selection and inventory management models, i.e. combinations of what to buy, from whom and
when is computed. These are compared to the solution that minimises TCO (Degraeve and Roodhooft,
1999b) and to the other vendor selection models. When a model only gives a solution to the vendor selection problem, i.e. what to buy from whom, the TCO is calculated under two dierent assumptions. First
it is assumed that all orders for the full time horizon are placed in the rst period. Subsequently we assume a
TCO optimal inventory policy, xing what to buy from whom using the original vendor selection model.
This second assumption gives the maximum possible credit to the vendor selection model that is compared
to the TCO approach. When a single item model is to be compared, we have applied it item by item to the
ball bearings case in order to nd a solution for the multiple item problem. A few of the compared models
were build around a specic case. For these approaches either the inventory carrying costs were of minor
importance or there was only a single item to be purchased. When applying them to the Cockerill Sambre
ball bearings case, we tried to stay within the philosophy of the original paper as much as possible and by
making the TCO optimal inventory assumption, we gave these vendor selection models the most credit
possible.

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

37

3. Classication of vendor selection decision models


As reported in Table 1, a distinction can be made between single item (Timmerman, 1986; Gregory,
1986; Nydick and Hill, 1992; Barbarosoglu and Yazgac, 1997; Willis et al., 1993; Li et al., 1997; Soukup,
1987; Thompson, 1990; Monczka and Trecha, 1988; Smytka and Clemens, 1993; Chaudhry et al., 1993;
Weber and Current, 1993; Pan, 1989) and multiple item models (Grando and Sianesi, 1996; Turner, 1988;
Current and Weber, 1994; Akinc, 1993; Sadrian and Yoon, 1994; Rosenthal et al., 1995; Benton, 1991;
Bender et al., 1985; Degraeve and Roodhooft, 1999b; Ronen and Trietsch, 1988). Single item models select
vendors for one product, but fail to take into account various interdependencies that could exist among the
dierent products. For example, a supplier can be oering a larger discount based on total sales volume,
irrespective of the product mix. Order level costs could be minimised by combining orders for several
products into one single order form. Single item models also underestimate the supplier level costs that arise
because of working with a supplier (e.g., plant visit, purchasing manager's time to negotiate). Moreover,
those costs are often completely disregarded. More than half of the vendor selection models handle the
problem on an item-by-item basis and have to be applied iteratively to select suppliers for multiple items.
Most of the existing literature treats vendor selection without multi-period inventory management. It
could be argued that purchasing managers should incorporate the decision to schedule orders over time
with the vendor selection decision. For example, at the order level, costs can dier substantially between the
dierent possible suppliers due to the possibility of ordering via Electronic Data Interchange (EDI). If, due
to inventory management reasons frequent ordering is necessary, a supplier with a low unit price but a high
order cost e.g., no EDI, can generate a higher TCO than a supplier with a higher unit price and an EDI
system. Another example is the trade o between receiving a quantity discount and the inventory holding
costs when buying larger lotsizes. It should be noted that, to our knowledge, no single item models with

Table 1
Classication of vendor selection models
Single item

Multiple item

Without inventory management over time

Without inventory management


over time

With inventory management over


time

Rating/linear
weighting

Total cost
approaches

Mathematical
programming

Rating/linear
weighting

Mathematical
programming

Mathematical
programming

Statistical

Timmerman
(1986) (categorical
method, linear
averaging)
Gregory (1986)

Timmerman
(1986) (cost
ratio method)

Chaudhry
et al. (1993)

Grando and
Sianesi (1996)

Turner (1988)

Bender et al.
(1985)

Ronen and
Trietsch (1988)

Monczka and
Trecha (1988)

Weber and
Current (1993)

Current and
Weber (1994)

Smytka and
Clemens
(1993)

Pan (1989)

Akinc (1993)

Degraeve and
Roodhooft
(1999b)

Nydick and Hill


(1992)
Barbarosoglu and
Yazgac (1997)
Willis et al. (1993)
Li et al. (1997)
Soukup (1987)
Thompson (1990)

Sadrian and
Yoon (1994)
Rosenthal
et al. (1995)
Benton (1991)

38

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

inventory management over time exist, except for Degraeve and Roodhooft (1998), but this is in fact the
multiple item model of Degraeve and Roodhooft (1999b) applied to the single item purchasing case of
heating electrodes at Cockerill Sambre.
To our knowledge, Degraeve and Roodhooft (1999b) is the rst model that makes the widely accepted
theoretical construct (e.g. Ellram, 1995a,b) of TCO operational in a purchasing context and uses ABC and
TCO information in an objective mathematical programming model to simultaneously select vendors and
determine order quantities for multiple items over a multiple period time horizon. The model is programmed in LINGO (Schrage, 1998) and can be solved on a Pentium with 32 Mb RAM in about 5 minutes
for the case of the ball bearings (Degraeve and Roodhooft, 1999a). Apart from Degraeve and Roodhooft
(1999b), only Bender et al. (1985) and Ronen and Trietsch (1988) deal with inventory management over
time and vendor selection in one model. Bender et al. do not include the mathematical programming model
in their paper. Ronen and Trietsch propose a decision support system that selects suppliers and schedules
order placements over time, but that is focusing on the lead time management of large projects. In this
specic situation there is a demand for a particular item at only one moment in time, xed via the PERT
environment in which the DSS is embedded. The inventory management problem here is essentially answering the question for every item ``how long before the due date will the order have to be placed?''
A third distinction exists between rating and linear weighting models, total cost approaches, mathematical programming models and statistical models. Rating models (Timmerman, 1986; Gregory, 1986;
Nydick and Hill, 1992; Barbarosoglu and Yazgac, 1997; Willis et al., 1993; Li et al., 1997; Soukup, 1987;
Thompson, 1990; Grando and Sianesi, 1996) are very subjective and often very sensitive to dierent rating
scales, weights and/or ratings by dierent people. Most of the linear weighting models are compensatory,
though some are non-compensatory (Grando and Sianesi, 1996). In a compensatory model a low rating on
one criterion can be compensated by a high rating on another criterion, whereas in non-compensatory
models dierent minimum levels for each criterion are required. Soukup (1987) and Thompson (1990)
include uncertainty with respect to certain features of the problem in their rating models. Total cost approaches (Timmerman, 1986 (cost ratio method); Monczka and Trecha, 1988; Smytka and Clemens, 1993)
attempt to quantify all costs related to the selection of a vendor in monetary units. Mathematical programming models (Chaudhry et al., 1993; Weber and Current, 1993; Pan, 1989; Turner, 1988; Current and
Weber, 1994; Akinc, 1993; Sadrian and Yoon, 1994; Rosenthal et al., 1995; Benton, 1991; Bender et al.,
1985; Degraeve and Roodhooft, 1999b) often consider only the more quantitative criteria. They can be
subdivided in linear, (mixed) integer or goal programming models. Statistical models (Ronen and Trietsch,
1988) incorporate uncertainty into the vendor selection decision.
Only few of the vendor selection models could not be applied to the problem of the ball bearings
procurement at Cockerill Sambre because of data availability or applicability reasons. Nydick and Hill
(1992) and Barbarosoglu and Yazgac (1997) use the analytic hierarchy process (AHP) to structure the
vendor selection problem following Narasimhan (1983). They formalise the trade os between the conicting selection criteria that are weighted relative to the importance attached to them by several specialists
from dierent subelds in the company. To evaluate this paper this process should have been carried out by
dierent managers at Cockerill Sambre, which was impossible.
Soukup (1987) introduces uncertainty with respect to the requirements patterns in a single item rating
model without inventory management. The forecasting of the market probabilities remains a subjective
process. Since in the Cockerill Sambre case the demand is stable and certain, we do not apply this approach
to the ball bearings problem.
Grando and Sianesi (1996) develop a multiple item rating model to help visualise possible vendor selection strategies. The dierent criteria are assessed on the basis of historical longitudinal data that are not
available for the Cockerill Sambre ball bearings case either. Although the model could be used noncompensatory the authors propose to give weights to each of the indices to get a single rating for each
supplier.

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

39

To apply the Ronen and Trietsch (1988) decision support system to the vendor selection and inventory
management problem the distribution of the lead time has to be known. The authors assume an exponential
distribution for simplicity but contend that in practice the distribution can be deduced from historical data.
It seems to us that in the context of large one-o projects this could pose severe problems.
Bender et al. (1985) describe a mixed integer programming model used at IBM that simultaneously
selects vendors and determines order quantities over a multiple time frame horizon with the objective to
minimise purchasing, inventory management and transportation costs. We cannot evaluate this model,
since the specic mathematical formulation is not included in their paper.
4. The problem of purchasing ball bearings at Cockerill Sambre
We study the procurement of ball bearings at Cockerill Sambre S.A., a Belgian multinational
company in the steel industry with external purchases approaching 0.6 billion annually accounting for
more than 70% of total costs. Management wants to improve the eciency of the purchasing process
and to reconsider the sourcing strategies for dierent product groups. Our case study refers to the ball
bearings, a product selected for study by the purchasing managers of the rm and a business of about
833,000 per year. There are 33 ball bearings types in the problem for which purchasing decisions must
be made.
The ball bearings are mainly used for transportation of the hot steel slabs after steel has been produced
in the converters and cast to form the slabs. The transportation lines consist of several rows of vertical steel
cylinders as depicted in Fig. 1.
The steel cylinders and the ball bearings are used in very arduous conditions under extremely high
temperatures. This causes the surface of the steel cylinders to deteriorate quickly such that they have to be
replaced frequently and brought to a maintenance department for reproling. At the time of replacement of
the cylinders, the ball bearings are also replaced in anticipation of potential problems and thus before they
have been used for their full lives. There is a salvage value associated with used ball bearings based on their
weight, amounting to 100 per ton. There are six possible suppliers, two of which are currently used by the
company. Important price dierences exist at the level of the individual ball bearing type. No quantity
discounts apply to this product group.
Whereas the literature (Dickson, 1966; Benton, 1991) mentions a variety of dimensions which could be
considered in vendor selection, the presented evaluation of vendor selection models is based on the case of

Fig. 1. Usage of the Ball Bearings.

40

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

ball bearings and does encompass only the criteria relevant to this purchasing problem. The dimensions on
which the suppliers dier from each other are price, possibility to deliver a certain item, service level, cost of
the purchasing manager to establish a relationship with a certain supplier, the possibility of EDI, the reliability of delivery visualised by the need to maintain a safety stock and the terms of payment. The more
traditionally used dimensions of quality, lead time management and quantity discounts are not relevant to
this case. The operational life span of the steel cylinders is shorter than that of the ball bearings, but the ball
bearings are replaced every time a cylinder is changed, before they are worn out. Therefore, the quality is
not a crucial issue in the case of ball bearings. Lead time is the same for all possible suppliers, but delivery
reliability diers from supplier to supplier, as visualised in the need to maintain safety stock. There are no
quantity discounts available.
For the ball bearings problem specically we consider at the supplier level, the cost of a dedicated
purchasing manager and a discount resulting from service provided by the supplier. Service includes
technical assistance, training, taking back of scrap and exibility. In this specic case study there is a return
instead of a cost on supplier level, because the service provided by the supplier exceeds the cost of the
purchasing manager. The order level costs include invoice cost per order, order cost per order and reception
cost per order. At the unit level we nd the salvage value of the used ball bearings, the price of the ball
bearings, the price discount as a percentage per time bucket due to payment delay given by a particular
supplier and the inventory holding cost.
5. Results of comparison from a total cost of ownership perspective
In Table 2, we summarise the resulting TCO for the solutions of the dierent vendor selection models.
The second and third column, respectively, state the inventory management and order splitting assumption, if applicable. The fourth column gives the TCO in percentages of the results of the TCO
minimising Degraeve and Roodhooft (1999a) model. The fth, sixth and seventh column give the three
components of the TCO, namely supplier level cost, order level cost and unit level cost in percentages.
Observe that there is a return (or negative cost) on the supplier level in this case. In Appendix A to the
paper we give details of the application of the dierent supplier selection models to the Cockerill Sambre
ball bearings case.
Almost half of the evaluated vendor selection models are single item rating and linear weighting models.
Timmerman (1986) proposes three dierent methods in order to accommodate the dierent situational
requirements of the procuring organisation. In the case of medium-sized manufacturers he proposes ``linear
averaging'' with a moderate ease of implementation and the highest cost/benet ratio. For the application
to the ball bearings case we have used the price and the payment delay as cost items. Service includes
technical assistance, education, buy back and recycling of scrap, competence and swiftness of intervention,
delivery reliability and the administrative costs of ordering and paying, as rated by the purchasing managers of Cockerill Sambre. No product specic items are taken into account, since for the ball bearings case
there are no dierences in the quality of the products for dierent suppliers. To obtain a rating on the price
component we used the method of Zenz (1994) who uses lowest price divided by actual price multiplied by
maximum rating. A problem not treated in Timmerman (1986) is what to do when two or more suppliers
receive the same rating for a specic item. To evaluate the model from a TCO perspective we have used
three assumptions to solve the problem: (1) the orders are split so that each supplier gets 1/x of the order
with x the number of suppliers receiving the same maximum rating for the item, (2) the whole order is given
to the supplier with the lowest price of those who attain the maximum rating and (3) we have only constrained the choice of the supplier(s) to those receiving the maximum rating and let the TCO model optimise the decision. Again, this assumption gives the most credit to the supplier selection model studied. As
can be seen in Table 2, the TCO of Timmerman's (1986) linear averaging method, varies, depending on the

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

41

Table 2
Evaluation of vendor selection models from a total cost of ownership perspective results

Degraeve and Roodhooft


Timmerman linear averaging
Timmerman linear averaging
Timmerman linear averaging
Timmerman linear averaging
Timmerman linear averaging
Timmerman linear averaging
Timmerman categorical method
Timmerman categorical method
Timmerman categorical method
Timmerman categorical method
Timmerman categorical method
Timmerman categorical method
Gregory
Gregory
Thompson
Thompson
Willis, Huston and Pohlkamp
Willis, Huston and Pohlkamp
Li, Fun and Hung on example
Willis et al.
Li, Fun and Hung on example
Willis et al.
Li, Fun and Hung
Li, Fun and Hung
Timmerman cost ratio method
Timmerman cost ratio method
Monczka and Trecha
Monczka and Trecha
Smytka and Clemens
Smytka and Clemens
Pan
Pan
Weber and Current
Weber and Current
Models that simplify to minimising
price
Models that simplify to minimising
price
Benton
Benton
Current and Weber SPLP
Current and Weber SPLP
Current and Weber PMLP 6 suppl.
Current and Weber PMLP 5 suppl.
Current and Weber PMLP 4 suppl.
Current and Weber PMLP 3 suppl.
Current and Weber PMLP 2 suppl.
Current and Weber PMLP 1 suppl.
Current and Weber PMLP 6 suppl.
Current and Weber PMLP 5 suppl.

Inventory
management
assumption

Order splitting
assumption

Total cost
of ownership (%)

Supplier
level cost
(%)

Order
level cost
(%)

Unit level
cost
(%)

N.A.
Period 1
Period 1
Period 1
TCO optimal
TCO optimal
TCO optimal
Period 1
Period 1
Period 1
TCO optimal
TCO optimal
TCO optimal
Period 1
TCO optimal
Period 1
TCO optimal
Period 1
TCO optimal
Period 1

N.A.
Split orders
Lowest price
TCO optimal
Split orders
Lowest price
TCO optimal
Split orders
Lowest price
TCO optimal
Split orders
Lowest price
TCO optimal
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.

100
115,53
114,72
114,72
108,33
107,53
107,53
119,01
114,62
114,62
110,06
107,40
107,40
122,58
115,33
122,58
115,33
121,29
114,04
120,95

100
179,40
175,75
175,75
179,40
175,75
175,75
166,85
149,83
149,83
154,01
149,83
149,83
217,61
217,61
217,61
217,61
211,13
211,13
209,01

100
16,67
16,67
16,67
87,5
100
100
8,33
8,33
8,33
50
50
50
2,08
12,5
2,08
12,5
4,17
18,75
4,17

100
116,85
115,98
115,98
109,73
108,86
108,86
120,03
115,39
115,39
110,96
108,27
108,27
124,52
117,40
124,52
117,40
123,12
116,00
122,75

TCO optimal

N.A.

113,71

209,01

Period 1
TCO optimal
Period 1
TCO optimal
Period 1
TCO optimal
Period 1
TCO optimal
Period 1
TCO optimal
Period 1
TCO optimal
Period 1

N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.

120,39
113,15
109,89
102,72
107,59
100,40
107,59
100,40
110,76
103,57
119,02
111,79
107,30

201,30
201,30
129,11
129,11
80,32
80,32
80,32
80,32
139,68
139,68
192,09
192,09
91,31

TCO optimal

N.A.

100,13

91,31

Period 1
TCO optimal
Period 1
TCO optimal
Period 1
Period 1
Period 1
Period 1
Period 1
Period 1
TCO optimal
TCO optimal

N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.

107,30
100,13
107,32
100,13
107,29
107,30
107,31
107,30
107,06
106,28
100,12
100,12

91,31
91,31
91,12
91,12
91,29
91,29
91,28
91,26
95,73
100
91,29
91,29

25
4,17
25
20,83
125
16,67
100
16,67
100
20,83
95,83
16,67
52,08
20,83
125
20,83
125
14,58
87,5
16,67
16,67
16,67
16,67
16,67
16,67
100
100

115,63
122,05
114,93
110,34
103,22
107,14
100,02
107,14
100,02
111,40
104,28
120,52
113,40
107,06
99,94
107,06
99,94
107,09
99,97
107,07
107,07
107,07
107,04
106,85
106,07
99,95
99,95

42

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

Table 2 (Continued)

Current and Weber PMLP 4 suppl.


Current and Weber PMLP 3 suppl.
Current and Weber PMLP 2 suppl.
Current and Weber PMLP 1 suppl.

Inventory
management
assumption

Order splitting
assumption

Total cost
of ownership (%)

Supplier
level cost
(%)

Order
level cost
(%)

Unit level
cost
(%)

TCO
TCO
TCO
TCO

N.A.
N.A.
N.A.
N.A.

100,12
100,12
100,07
100

91,28
91,26
95,73
100

100
100
100
100

99,95
99,95
99,98
100

optimal
optimal
optimal
optimal

assumptions, between 107.5% and 115.5 % of the TCO minimising model. This illustrates the advantage in
TCO of including inventory management in the supplier selection decision. In the sensitivity analyses we
have counted how many item-supplier combinations changed in comparison to those of the original model
when changing weights and rating limits. Although Timmerman's (1986) linear averaging method is not
sensitive to changes of weights within a certain category such as service, we have observed that it is very
sensitive to changes in weights across categories and to changes in rating limits. In investigating the impact
of such changes, we observed in our computational experiments that up to half of the items were ordered
from a dierent supplier.
For small organisations, Timmerman (1986) proposes the categorical method which is the easiest to
implement and has the lowest cost, but also the least clear results. For the application to the ball bearings
case cost, service and speed of delivery are rated good, neutral or unsatisfactory and combined into a total
rating. Again, the product quality is irrelevant to the case at hand. Making the same assumptions w.r.t.
inventory management and order splitting as with the comparison of the linear averaging method, we
obtain a TCO varying from 107.40% to 119.01%. The categorical method is very sensitive to changes in
ratings: 29 out of 33 combinations suppliers-item change. For large organisations with computerised cost
accounting systems Timmerman (1986) proposes the cost-ratio method. This method is treated below since
it is not a rating method.
Gregory (1986) describes the vendor rating system used by Texas Instruments, that again works with
weights on two levels. He recognises ve main categories, each divided into subcategories. For our application we use, in the ``proposal responsiveness'' category, ratings on terms and conditions and on
timeliness of deliveries. In the ``quality/reliability'' category we make two subcategories that are relevant for
the case: experience with the company and electronic data interchange (EDI). In the ``cost'' category we
rate the unit price, again using Zenz method. The ``general'' category encompasses ratings on past delivery
history and payment provisions. In our case there is no ``technical'' category since all products have similar
specicity and quality. In contrast to Timmerman, Gregory proposes two methods for order splitting.
When applying the method to the ball bearings case no two suppliers received the same maximum rating.
From Table 2 we observe that the TCO of Gregory (1986) varies between 115.3% and 122.6 % depending
on the inventory management assumption made. The model is not sensitive to changes in rating limits, i.e.
no combination supplier/item changed and only somewhat sensitive to changes in weights. The sensitivity
analyses resulted in 1 and 4 changes in combinations supplier-item on 33 possible combinations for changes
in weights within and between category, respectively.
Thompson (1990) introduces Monte Carlo simulation to reduce the uncertainty innate to the rating
mechanism. Ratings can be simulated using a uniform or a triangular distribution. The interpretation of the
resulting distribution of scores happens by judging modus, variance and overlap. Giving weights to the
various criteria remains a subjective process. Only uncertainty related to the rating process itself is dealt
with in this model. Soukup (1987) introduces uncertainty related to the requirements. Quite strange, in the
Thompson paper, only qualitative criteria are rated and in the presented example no price variable is included. To evaluate the TCO of this model we have made a simulation for the ratings from a uniform

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

43

distribution, using scores on compatibility of systems (EDI), service and delivery reliability. From Table 2,
we see that the TCO of Thompson (1990) varies between 115.3% and 122.6% depending on the inventory
management assumption made.
Willis et al. (1993) use ``dimensional analysis'' in a single item model where a series of pairwise comparisons are made among suppliers using a Vendor Performance Index, dened as follows:
s
w
n 
Y
Xi i
w
VPI
Yi
i1
with Xi and Yi criterion performance score for supplier X and Y, respectively, wi is weight assigned to
criterion i and w sum of all weights.
The advantage of this method is that each criterion can be measured in its own units, but the rating and
weighting system remains highly subjective. Also it is impossible to obtain a zero score on a criterion since
division by zero is not dened. For the application to the ball bearings case, in the philosophy of the paper,
we have used the criteria price, delivery performance and stocking policies, ease of ordering (EDI) and
service. The TCO of Willis et al. (1993) varies between 114% and 121.3 % depending on the inventory
management assumption. Their model proves to be insensitive to changes in weights and ratings, i.e. no
combination supplier-item changed.
Li et al. (1997) have published a paper in reaction to the Willis et al. (1993) paper in which they describe
the disadvantages of the VPI. Alternatively, they propose a fuzzy sets methodology by introducing the SUR
index that takes the inconsistency of the evaluator into account for each qualitative criterion. If we apply
this approach to the criteria proposed in Willis et al. (1993) the TCO varies between 113.7% and 121%
depending on the inventory management assumption, which is slightly better than the TCO of Willis et al.
being 99.7%. We have observed that the sensitivity to changes in weights and ratings is very high with 11
and 18 changes out of 33 possible combinations, respectively, surely compared to that of the Willis et al.
model. If we apply the Li et al. methodology and their proposed criteria and weights to the ball bearings
case, using ratings on price, delivery, exibility and response without the specic compensation for
judgement since there are no data available for this case, we nd TCO between 113.2% and 120.4 %. Our
sensitivity analyses indicate that the model remains very sensitive to changes in weights and ratings with 12
and 15 changes out of 33, respectively.
Observe from Table 2 that all these single item rating models without inventory management overestimate the return on supplier level and the cost on unit level and underestimate the cost on order level. The
cost on supplier level is underestimated because in most of the rating models the cost of the dedicated
purchasing manager is not taken into account. The cost on order level is underestimated because of the oneperiod time frame. Many of the costs that are in fact varying with the number of orders are regarded in
these single item rating models as driven by unit.
To overcome the subjectivity of the rating process, single item total cost approaches without inventory
management quantify the costs associated with working with a specic vendor as much as possible in
monetary units. For large organisations with computerised cost accounting systems Timmerman (1986)
proposes the ``cost-ratio method''. This method collects all costs related to quality, delivery and service
and expresses them as a percentage of unit price. Again, quality is not relevant to this specic case.
Summing the two percentages for the above categories leads to a total penalty or benet percentage that
is used to adjust the quoted price to obtain the net cost. Although this method is not a rating method and
goes far in collecting total cost information, the dierent categories are implicitly weighted equally since
the percentages are summed. From Table 2 we see that the TCO of Timmerman (1986), cost-ratio
method, varies between 109.89% and 102.72% depending upon the inventory management assumption
made.

44

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

Monczka and Trecha (1988) develop a supplier performance index that adjusts the net price for nonperformance costs associated with the supplier. Apart from this cost index, a separate rating on the service
level of the supplier is calculated, but it is not clear how this plays in the vendor selection decision. For the
application to the ball bearings case net price and the costs of ordering, invoicing and the purchasing
manager are included, in the philosophy of the paper. Depending on the inventory management assumption, TCO of 100.40% and 107.59% are obtained.
Smytka and Clemens (1993) also develop a more elaborated total cost approach in which they rst assess
`risk factors' on a go/no go basis. Then they develop rates on several ``business desirable factors'' such as
delivery performance, but it is not clear how these ratings are used in the vendor selection process. Finally,
they collect information on a very extensive list of ``measurable cost factors'' and calculate total cost. In the
ball bearings case all six vendors are considered `go' and price, payment delay, ordering, invoicing, purchasing manager and inventory holding are the measurable cost factors. Depending on the inventory
management assumption, TCO of 100.40% and 107.59% are obtained. All the single item total cost approaches without inventory management underestimate the return on supplier level and overestimate the
cost on unit level.
To the best of our knowledge three mathematical programming single item models without inventory
management exist. Pan (1989) proposes a single item linear programming model in order to allocate order
quantities among suppliers in a multiple sourcing context. He minimises aggregate price subject to predetermined constraints on quality and service level and lead time. For the application to the ball bearing
case quality is irrelevant and the delivery constraint is met by all six vendors, although their delivery
performance dier. Only a constraint on service is included. TCO varies from 110.76% to 103.57%, depending on the inventory management assumption made. Pan (1989) is very sensitive to changes in predetermined levels on the constraints. In the application to the ball bearings case, only 7 out of 33 items
where bought from the same suppliers if the predetermined levels on service were changed by one percent
up or down. The allocation of order quantities stays the same for only 4 out of these 7 items.
Weber and Current (1993) use multi-objective mixed integer programming to select vendors for a
single item. They propose price, delivery and quality objectives. Using more complex weighting and
constraint methods, they overcome the problem of sensitivity to predetermined levels inherent to Pan
(1989). Weber and Current (1993) provide decision support to purchasing managers by presenting the
trade o curves among the dierent objectives. When we apply this method to the ball bearings case only
price and delivery objectives are relevant. Instead of the quality constraint we included a service constraint since this is very important to the case at hand. We nd TCO of 119.02% and 117.79% depending
on the inventory management assumption. Weber and Current (1993) use a value path analysis to
graphically display the results. Their method can be easily extended to handle multiple items (Weber,
1991). Weber and Desai (1996) propose Data Envelopment Analysis (DEA) for evaluation of vendors
that were already selected. Weber et al. (1998) combine the multi-objective and the DEA method to
provide a negotiation tool with vendors that are not selected in the rst instance. The methodology can
be used by purchasing managers to negotiate with these vendors so that they can move to ecient points
on the DEA curve.
Chaudhry et al. (1993) develop linear and mixed binary integer programming models for a single item
and take special interest in the modelling of price breaks, i.e. quantity discounts vs. surcharges, all-units vs.
incremental discounts. The authors include constraints on quality level and delivery. Since for the ball
bearings case there is no dierence in quality, the delivery constraint is met by all vendors and there are no
discounts available, the model simplies to minimising net price. A TCO between 100.1% and 107.3% is
obtained depending on the inventory management assumption.
Also, many of the mathematical programming multiple item models without inventory management
simplify to minimising price for the ball bearings case (Rosenthal et al., 1995; Sadrian and Yoon, 1994;
Akinc, 1993; Turner, 1988). Rosenthal et al. (1995) develop a multiple item mixed integer programming

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

45

model for the special case in which suppliers oer discounted prices for bundled products. The same quality
and delivery constraints as in Chaudhry et al. (1993) are added for every item.
Sadrian and Yoon (1994) propose a multiple item mixed integer programming model that is focusing on
the modelling of business volume discounts. The so called ``reliability costs'' are not modelled. None of the
optional constraints the authors propose, nor the volume discounts are relevant to the Cockerill Sambre
case.
Akinc (1993) concentrates mainly on the number of suppliers. To start with, he proposes two models to
obtain the extreme number of vendors. One model chooses the cheapest supplier for each item, what leads
to the largest rational number of suppliers. The second model chooses the smallest number of suppliers that
can deliver all items, disregarding cost. For the ball bearings case this number of suppliers is 6 and 1 respectively. Then, model one and two support model three in which the trade-o between number of suppliers and cost is analysed. A number of heuristics are proposed to make the model workable for large data
sets. Since in the case of Cockerill Sambre there is no dierence on the quality criterion nor are dierent
minimum levels of delivery performance required for dierent items, also this model results into minimising
prices.
Turner (1988) describes the linear programming routine for the multiple item problem of British Coal.
Instead of solving the integer problem, the LP formulation is repeated in an interactive manner to allow the
purchasing managers to check on several possible solutions. Apart from price and discounts only capacity
constraints and region limits are modelled. Since there are no discounts nor region limits in the Cockerill
Sambre case, also Turner (1988) model simplies to minimising net prices.
Current and Weber (1994) and Benton (1991) are the only two papers of the multiple item mathematical
programming variety without inventory management that do not result into minimising prices.
Benton (1991) presents a heuristic procedure to solve the multiple item problem with a non-linear
objective function with discontinuities that minimises the sum of ordering costs, holding costs and net
price, giving special attention to the modelling of quantity discounts. He adds constraints on the total
inventory investment and the warehousing space occupied. Since in the ball bearings case there are no
quantity discounts, nor constraints on inventory investment or space, this single time period model
simplies to minimising the sum of price, ordering and holding costs. TCO of 107.3% and 100.1% are
obtained depending on the inventory management assumption. The return on supplier level is underestimated.
Current and Weber (1994) make an application of facility location modelling constructs to the vendor
selection problem. First they formulate the Single Plant Location Problem (SPLP) as a vendor selection
model minimising the sum of xed costs and ``actual purchasing'' costs. The decision variables are the
fraction of item is demand purchased from vendor j and a binary variable that denotes whether vendor j is
selected or not. Expediting costs and internal processing costs are classied with the xed costs, though they
are dependent on the number of orders or set ups in multiple time frame models. Because, the SPLP
formulation does not take inventory management into account these costs are underestimated. For the
application to the ball bearings case, xed costs are the sum of contract set up, internal processing, ordering
and invoicing costs. TCO of 107.3% and 100.1% are obtained depending on the inventory management
assumption. They also propose a SPLP formulation with the minimisation of late deliveries, but this is
irrelevant to our case. Also, this model sums two dierent units, namely number of units and cost in a
currency. Second, they propose a model based on the p-Median Location Problem (PMLP). The decision
variables are the same as in the SPLP formulation. After removing a typographical error in the objective
function (ai has to be removed or else the unit would be quantity squared) and adding a constraint that
forces at least one unit to be ordered from a selected supplier, the formulation is identical to the SPLP case
except for xing the desired number of suppliers. The constraint has to be added because by xing the
number of suppliers the program could add the xed cost of a selected supplier without ordering anything
because of the high variable cost. The evaluation of the PMLP model is made in comparison to the TCO

46

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

model with a xed number of suppliers. TCO varies between 100.1% and 107.3%. Finally, they introduce
the Set Covering Location Problem (SCLP) minimising the number of suppliers, assuming that total cost
and price are unimportant. A disadvantage of this model is that it can not select suppliers when more than
one supplier is able to deliver all the desired products. When applied to the Cockerill Sambre case one single
supplier is selected since only this supplier is able to deliver all the types of ball bearings. TCO of 122.6%
and 115.3% are obtained depending on the inventory management assumption. All the Current and Weber
(1994) models underestimate the return on supplier level and the cost on order level.
6. Conclusions and suggestions for future research
Given the TCO perspective, several conclusions can be drawn from Table 2. First, multiple item
mathematical programming models are always performing better than single item rating models. Single
item models fail to take into account the interdependencies that could exist among the dierent products. A
supplier can be oering a larger discount based on total sales volume, irrespective of the product mix. Order
level costs could be minimised by combining orders for several products into one single order form.
Mathematical programming models approach the problem in a more objective way than rating models by
optimising an explicitly stated objective function. Second, total cost approaches outperform rating models
by objectifying the supplier selection process. Third, it is a good strategy to incorporate inventory management into the vendor selection decision. For example, at the order level, costs can dier substantially
among the dierent suppliers due to the possibility of ordering via EDI. If, due to inventory management
reasons frequent ordering is necessary, a supplier with a low unit price but a high order cost (e.g., no EDI),
can generate a higher Total Cost of Ownership than a supplier with a higher unit price and an EDI system.
Another example is the trade-o between the receiving a quantity discount and the inventory holding costs
when buying larger lotsizes. Fourth, it is not rewarding to x in advance the number of suppliers to use.
There is no unequivocal relationship between the number of suppliers and the TCO. Fifth, dierent levels in
the ABC hierarchy are under- or overestimated in the dierent vendor selection models. All the single item
rating models without inventory management overestimate the return on supplier level and the cost on unit
level and underestimate the cost on order level. The multiple item mathematical programming models
underestimate the return on supplier level. All the models without inventory management underestimate
order level costs seriously and overestimate unit costs, under the assumption that everything is bought in
the rst period.
Future research should be conducted in developing multiple item mathematical programming vendor
selection models with inventory management, since the simultaneous decision to select vendors and to
determine order quantities seems to be saving on TCO. A second fruitful path for future research is to
introduce uncertainty with respect to requirements, deliveries, quality, prices etc. in decision models.
Thirdly, the same methodology of comparison can also be applied to other real life data sets to check
whether the conclusions remain. At the moment we are working on a very extensive data set of Alcatel
Bell where other criteria and costs are of importance. As one of the referees suggested, the models being
benchmarked will perform quite dierently under dierent procurement situations. In the repeat buying
situation of procuring ball bearings at Cockerill Sambre where all supplier characteristics are readily
known from procurement history, most costs will be incurred at the unit level. From Table 2, we see
that many of the benchmarked models perform well in this case. However, we speculate that the TCO
approach will even perform better in other procurement situations like such as rst time buys where
costs on supplier and order level are of considerable importance. The costs on these levels are often
neglected or underestimated in the benchmarked models. Future research could investigate which
approach performs best in dierent buying situations such as repeat buy vs rst time buy and JIT vs
non-JIT.

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

47

Appendix A. Details of the application of the dierent supplier selection models to the Cockerill Sambre ball
bearings case
Due to condentiality reasons we are not allowed to present the actual data of the case study used to
evaluate the dierent vendor selection models. Therefore, we choose to work with the following symbols,
representing the real data. While using these symbols to discuss all the supplier selection models studied, it
becomes clear how they were applied to the ball bearings case.
N
M
P
mcs
ses
vcs
ocs
rcs
rev
wti
psi
dcs
SSi
dit
di
h
nis
slc
mins
maxs
olc
bsi
api
aulc
arev
purc
invc
ulc
zs
ust
xtst
sdsit
xsit
ysit
vsit
sxit

set of ball bearing types, index i (1 to 33)


set of time periods, index t
set of suppliers, index s (1 to 6)
yearly cost of a dedicated purchasing manager for supplier s incurred for the time devoted to
managing the specic ball bearings in the problem, "s 2 P
cost savings resulting from extra service provided by the supplier, "s 2 P
cost of invoicing per order placed with supplier s, "s 2 P
cost of ordering per order placed with supplier s, "s 2 P
cost of receiving per order placed with supplier s, "s 2 P
salvage value of used ball bearings in BF per kilogram
weight in kilogram of ball bearing type i, "i 2 N
price for ball bearing type i oered by supplier s, "i 2 N, "s 2 P
price discount as a percentage per time period due to a credit period given by supplier s,
"s 2 P
safety stock of ball bearing type i, "i 2 N
demand for ball bearing type i in time period t, "i 2 N, "t 2 M
total demand for ball bearing type i, "i 2 N
inventory holding cost per period as a percentage of the products price
number of dierent items that can be delivered by supplier s, "s 2 P
total supplier level costs per year
minimum number of suppliers to use over the total time horizon
maximum number of suppliers to use over the total time horizon
total order level costs per year
beginning inventory of ball bearing type i bought from supplier s, "s 2 P, "i 2 N
average price of ball bearing type i in BF, "i 2 N
the ``additional'' unit level costs generated per year
revenue generated from selling o used ball bearings per year
total purchase costs per year
the yearly inventory holding cost
total unit level costs per year
1, if we buy from supplier s during the year, 0, otherwise, "s 2 P
1, if we buy from supplier s in time period t, 0, otherwise, "s 2 P, "t 2 M
total number of ball bearings bought from supplier s in period t, "s 2 P, "t 2 M
consumption of ball bearings type i bought from supplier s in period t, "s 2 P, "i 2 N, "t 2 M
amount of ball bearings type i bought from supplier s in period t, "s 2 P, "i 2 N, "t 2 M
1, If we buy ball bearing type i from supplier s in period t, "s 2 P, "i 2 N, "t 2 M
inventory of ball bearing type i bought from supplier s at the end of time period t, "s 2 P,
"i 2 N, "t 2 M
total amount of ball bearing type i bought in period t, "i 2 N, "t 2 M

48

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

syit
svit
rsi
VEN
Pi

1, If we buy ball bearing type i in period t, "i 2 N, "t 2 M


inventory of ball bearing type i at the end of time period t, "i 2 N, "t 2 M
fraction of ball bearing is demand purchased from vendor s, "s 2 P, "i 2 N
number of vendors to be employed
set of vendors that can supply item i, "i 2 N

A.1. Degraeve and Roodhooft (1996)


A more thorough explanation of how the Degraeve and Roodhooft (1996) model applies to the
Cockerill Sambre ball bearings case is provided in Degraeve and Roodhooft (1998b). The mathematical
formulation is repeated here:
Min

slc olc ulc;


X XX
X
mcs zs
ses
psi xsit ;

slc

s2P

olc

s2P

i2N t2M

XX
vcs ocs rcs ust ;
s2P t2M

ulc purc invc arev;


purc

XXX

psi 1 dcs xsit ;

s2P i2N t2M

XXX

invc

h psi vsit

s2P i2N t2M

arev

X
i2N

6 h api SSi ;

XX
rev wti dit ;
i2N t2M

XX

sdsit dit

8t 2 M; 8i 2 N ;

s2P i2N

bsi xsi1 vsi1 sdsi1


vsit1 xsit vsit sdsit
xsit 6

dil ysit

8s 2 P ; 8i 2 N ;
8s 2 P ; 8i 2 N ; 8t 2 M n f1g;

8s 2 P ; 8i 2 N ; 8t 2 M;

l2M; l P t

X
xsit xtst
i2N

8s 2 P ; 8t 2 M;

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

xtst 6

ust 6

i2N

l2M;l P t

ysit

dil

ust

8s 2 P ; 8t 2 M;

8s 2 P ; 8t 2 M;

i2N

ysit 6 ust

8s 2 P ; 8i 2 N ; 8t 2 M;

X
zs P mins;
s2P

X
zs 6 maxs;
s2P

zs 6

X
ust

8s 2 P ;

t2M

ust 6 zs

8s 2 P ; 8t 2 M;

zs 2 f0; 1g

8s 2 P ;

ust 2 f0; 1g; xtst P 0

8s 2 P ; 8t 2 M;

ysit 2 f0; 1g; xsit P 0; sdsit P 0; vsit P 0


X
xsit sxit

8s 2 P ; 8i 2 N ; 8t 2 M;

8i 2 N ; 8t 2 M;

s2P

X
vsit svit

8i 2 N ; 8t 2 M;

s2P

sxit 6

l2M; l P t

syit 6

ysit

dil syit

8i 2 N ; 8t 2 M;

s2P

ysit 6 syit

8s 2 P ; 8i 2 N ; 8t 2 M;

l
X X
X
sxit 6
dir
t2C

8i 2 N ; 8t 2 M;

t2C

rt

!
syit svil ;

1 6 l 6 j M j; L f1; 2; . . . ; lg; and C  L:

49

50

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

A.2. Timmerman (1986)


(a) Linear averaging. Ratings were given by management of Cockerill Sambre for every item i 2 N and
every vendor s 2 P using the following criteria and weights.
Item i

Weight

Cost criteria
BEF cost
Terms of sale
Quality costs

0.4
0.1
0a

Service criteria
Technical support, education, buy back and recycling
of scrap, competence, swiftness of intervention
Delivery reliability
Administrative costs of ordering and paying
Lead time

s1

s2

s3

s4

s5

s6

0.3
0.1
0.1
0a
0a

Product criteria
Weighted R
a

Some of the proposed criteria are given a weight of zero either because they are not relevant to the ball
bearings case (e.g. quality), or because they are the same for all suppliers (e.g. lead time, product
characteristics).
The actual purchase price for the item is converted into a rating on BEF cost using Zenz method (1994,
price
 maximumrating.
p. 136): lowest
actual price
(b) Categorical method. For every item, each vendor is rated on cost, service and delivery performance
``good (+)'', ``neutral (0)'' or ``unsatisfactory ())''. The combination of ratings result in, from high to low:
++,++,+, 0,),) ),) ) ) . The vendor with the highest score is selected.
(c) Cost-ratio method. For every item, the costs associated with quality, delivery and service are determined. Then, each is converted into a cost ratio, in which the cost is expressed as a percentage of the total
value of the purchase. Since quality is not relevant to the ball bearings case, only a delivery and a service
cost ratio are calculated. These are combined into a total cost adjustment percentage, which is applied to
the quoted price per unit, resulting in the net adjusted cost per unit in BEF. The supplier with the lowest net
adjusted cost per unit is selected
Delivery cost ratio (dcris ) for item i

s1

s2

s3

s4

s5

s6

s1

s2

s3

s4

s5

s6

Ordering (ocs )
Invoicing (vcs )
Delivery performance cost SSys  6h  psi  di
(-) Terms of payment (dcs )
R
R as a percentage of psi  di
Service cost ratio (scris ) for item i
Service as a percentage of psi  di ses  psi  di

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

51

Then, for every item i and every supplier s, dcris + scris is calculated and applied to psi  di . For every item i,
the supplier s with the lowest net adjusted cost is selected.

A.3. Gregory (1986)


Ratings were given by management of Cockerill Sambre for every item i 2 N and every vendor s 2 P
using the following criteria and weights.

Item i

Weight

Proposal responsiveness
Problem understanding
Terms and conditions
Timeliness of deliveries
Weighted total

0a
5
5
(10)
0a

Technical
Quality/reliability
Experience
Performance history
Quality of data (EDI)
Survey score
Weighted total
Cost
Unit price
Price curve
Weighted total
General
Past delivery history
Management organisation
Personnel qualications
Facilities
Payment provisions
Weighted total

4
0a
0a
0a
1
(5)

Summary
Proposal responsiveness

Technical
Quality/reliability
Cost
General
Weighted total

0a
5
4
1
(12)

s1

s2

s3

s4

s5

s6

x
y
x y=
10  100

5
0a
3
0a
(8)
7
0a
(7)

x y=
10  100

weighted
P

Some of the proposed criteria are given a weight of zero either because they are not relevant to the ball
bearings case (e.g. quality performance history), or because they are the same for all suppliers (e.g.
technical product characteristics).

52

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

Again, Zenz method was used to convert price into a cost rating.
A.4. Willis et al. (1993)
Item i

Weight

s1

s2

p1

p2

dp1

dp2

eo1
se1

eo2
se2

s3

s4

s5

s6

Quality
Price
Response to special orders and problems
Delivery performance and stocking policies
Financial stability
Ease of ordering
Service
P
|weight|

0
)5
0a
5
0a
2
4
16

a
Some of the proposed criteria are given a weight of zero either because they are not relevant to the ball
bearings case (e.g. quality), or because they are the same for all suppliers (e.g. nancial stability).

For every item, vendors are


compared pairwisely, using the VPI-index. For example, for the comparison
r

5  5  2  4
16
p1
1
1
1
 dp
 eo
 se
is calculated. If the result is bigger than 1,
of vendor 1 to vendor 2
p
dp
eo2
se2
2

supplier 1 is compared to supplier 3; otherwise supplier 2 is compared to supplier 3, and so on.


A.5. Li et al. (1997)
1. Applying the methodology to the example for the Willis et al. (1993) paper. For every item, vendors
are compared to all other vendors, using the SUR-index. For example,
p


p1 p
5
dp1 d
5
eo1 eo
2
se1 se
4




 :
pmax pmin 16 dpmax dpmin 16 eomax eomin 16 semax semin 16
2. Applying the methodology to the proposed criteria of the Li et al. (1997) paper
Item i
Quality
Cost (price)
Delivery performance
Flexibility (ease of ordering)
Response
(service)
P
|weight|

Weight

s1

s2

p1
dp1
eo1
se1

p2
dp2
eo2
se2

s3

s4

s5

s6

0
)0.25
0.2
0.1
0.1
0.65

For every item, vendors are compared to all other vendors, using the SUR-index. For example,
p


p1 p
0:25
dp1 d
0:2
eo1 eo
0:1
se1 se
0:1

:




pmax pmin 0:65
dpmax dpmin 0:65 eomax eomin 0:65 semax semin 0:65

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

53

A.6. Thompson (1990)


For every item i and every criterion, a highest (H) and a lowest (L) rating is given by management. A
thousand simulations are performed using a uniform distribution between these highest and lowest rating.
These are weighted and a distribution for the score of each supplier is presented visually.
Item i

Weight

Compatibility
(EDI)
Service
Delivery
reliability

s1
H

s2
L

s3
L

s4
L

s5
L

5
3

A.7. Monczka and Trecha (1988)


For every item i 2 N and every supplier s 2 P the following costs are summed:
psi

ocs vcs mcs

:
di
nis

The lowest total cost supplier is selected.


A.8. Smytka and Clemens (1993)
For every item i 2 N and every supplier s 2 P the following costs are summed:
psi 1 dcs

ocs vcs mcs

6hpsi :
di
nis

The lowest total cost supplier is selected.


A.9. Pan (1989)
For every item i 2 N the following LP is solved:
X
psi  xsi ;
min
s2P

X
xsi di ;
s2P

X
ses  xsi P prelevs  di ;
s2P

xsi 2 Z

8s 2 P ; 8i 2 N :

s6
L

54

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

A.10. Weber and Current (1994)


For every item i 2 N several non-inferior solutions are generated using the following mixed integer
programming models. Then the purchasing manager selects the option of his or her preference. The rst
MIP model minimises price under dierent constraints for delivery performance and service.
min

X
psi  xsi ;
s2P

X
xsi di ;
s2P

X
SSys  xsi 6 delivery performance level  di ;
s2P

X
ses  xsi P service level  di ;
s2P

xsi 2 Z:
The second MIP model maximises service under dierent constraints for price and delivery performance.
max

ses  xsi ;

s2P

X
xsi di ;
s2P

X
psi  xsi 6 price level  di ;
s2P

X
SSys  xsi 6 delivery performance level  di ;
s2P

xsi 2 Z:

A.11. Current and Weber (1994)


1. SPLP
Min
F

C F;
X
s2P

mcs ocs vcs zs ;

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

XX

psi  di  rsi ;

s2P i2N

XX

rsi 1

8i 2 N ;

s2P i2N

X
zs P min s;
s2P

X
zs 6 max s;
s2P

rsi 6 zs

8s 2 P ; 8i 2 N ;

zs 2 f0; 1g

8s 2 P ;

rsi 2 0; 1 8s 2 P ; 8i 2 N :
2. PMLP
Min
F

C F;
X

mcs ocs vcs zs ;

s2P

XX

psi  di  rsi ;

s2P i2N

XX

rsi 1

8i 2 N ;

s2P i2N

rsi 6 zs

8s 2 P ; 8i 2 N ;

X
zs VEN;
s2P

zs 2 f0; 1g

8s 2 P ;

rsi 2 0; 1 8s 2 P ; 8i 2 N :
3. SCLP
Min

VEN;

VEN

X
zs ;
s2P

55

56

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

X
zs P 1 8i 2 N ;
s2Pi

zs 2 f0; 1g

8s 2 P :

A.12. Models that simplify to minimising net prices for the ball bearings case
Akinc (1993); Chaudhry et al. (1993); Rosenthal et al. (1995); Sadrian and Yoon (1994); Turner (1988).
X
mipri  di ;
Min
i2N

mipri min psi


VEN

8i 2 N ; 8s 2 P ;

X
zs ;
s2P

X
zs P min s;
s2P

X
zs 6 max s;
s2P

ysi 1 if psi mipri

8i 2 N ; 8s 2 P ;

ysi 0 if psi 6 mipri

8i 2 N ; 8s 2 P ;

zs 2 f0; 1g
zs 6

XX

8s 2 P ;
ysi

8s 2 P ;

s2P i2N

ysi 6 zs :

A.13. Benton (1991)


The objective is to minimise the sum of price, ordering costs and holding costs.
X
XX
XX
psi  di  xsi
oc  zs
0:5  h  psi  xsi ;
Min
s2P i2N

X
s2P xsi 1

8i 2 N ;

s2P

s2P i2N

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

57

X
zs P min s;
s2P

X
zs 6 max s;
s2P

zs 2 f0; 1g
xsi 6 zs

8s 2 P ;

8s 2 P ; 8i 2 N :

References
Akinc, U., 1993. Selecting a set of vendors in a manufacturing environment. Journal of Operations Management 11, 107122.
Barbarosoglu, G., Yazgac, T., 1997. An application of the analytic hierarchy process to the supplier selection problem. Production and
Inventory Management Journal, 1st quarter, 1421.
Bender, P.S., Brown, R.W., Isaac, M.H., Shapiro, J.F., 1985. Improving purchasing productivity at IBM with a normative decision
support system. Interfaces 15 (3), 106115.
Benton, W.C., 1991. Quantity discount decisions under conditions of multiple items, multiple suppliers and resource limitations.
International Journal of Production Research 29 (10), 19531961.
Carr, L.P., Ittner, C.D., 1992. Measuring the cost of ownership. Journal of Cost Management 6 (3), 713.
Cavinato, J.L., 1992. A total cost/value model for supply chain competitiveness. Journal of Business Logistics 13 (2), 285301.
Chaudhry, S.S., Forst, F.G., Zydiak, J.L., 1993. Vendor selection with price breaks. European Journal of Operational Research 70, 52
66.
Current, J., Weber, C., 1994. Application of facility location modelling constructs to vendor selection problems. European Journal of
Operational Research 76, 387392.
Degraeve, Z., Roodhooft, F., 1998. Determining sourcing strategies: A decision model based on activity and cost driver information.
Journal of the Operational Research Society 49 (8), 781789.
Degraeve, Z., Roodhooft, F., 1999a. Improving the eciency of the purchasing process using total cost of ownership information: The
case of heating electrodes at Cockerill Sambre S.A. European Journal of Operational Research 112 (1), 4253.
Degraeve, Z., Roodhooft, F., 1999b. A mathematical programming approach for procurement using activity based costing.
forthcoming in Journal of Business Finance and Accounting.
Dickson, G.W., 1966. An analysis of vendor selection systems and decisions. Journal of Purchasing 2 (1), 517.
Ellram, L.M., 1994. Total Cost Modeling in Purchasing. Center for Advanced Purchasing Studies, Tempe, AZ.
Ellram, L.M., 1995a. Activity-based costing and total cost of ownership: A critical linkage. Journal of Cost Management, winter,
2230.
Ellram, L.M., 1995b. Total cost of ownership an analysis approach for purchasing. Journal of Physical Distribution and Logistics 25
(8), 423.
Ellram, L.M., Sierd, S.P., 1993. Purchasing: The cornerstone of the total cost of ownership concept. Journal of Business Logistics 14
(1), 163184.
Grando, A., Sianesi, A., 1996. Supply management: A vendor rating assessment. CEMS Business Review 1, 199212.
Gregory, R.E., 1986. Source selection: A matrix approach. Journal of Purchasing and Materials Management, summer, 2429.
Kaplan, R.S., Cooper, R., 1997. Cost & eect: Using integrated systems to drive protability and performance. Harvard Business
School Press, Boston, MA.
Li, C.C., Fun, Y.P., Hung, J.S., 1997. A new measure for supplier performance evaluation. IIE Transactions on Operations
Engineering 29, 753758.
Monczka, R.M., Trecha, S.J., 1988. Cost based supplier performance evaluation. Journal of Purchasing and Materials Management,
spring, 27.
Narasimhan, R., 1983. An analytic approach to supplier selection. Journal of Purchasing and Supply Management 1, 2732.
Nydick, R.L., Hill, R.P., 1992. Using the analytic hierarchy process to structure the supplier selection procedure. International Journal
of Purchasing and Materials Management, spring, 3136.
Pan, A.C., 1989. Allocation of order quantities among suppliers. Journal of Purchasing and Materials Management, fall, 3639.
Roehm, H.A., Critcheld, M.A., Castellano, J.F., 1992. Yes ABC works with purchasing too. Journal of Accountancy 11, 5862.

58

Z. Degraeve et al. / European Journal of Operational Research 125 (2000) 3458

Ronen, B., Trietsch, D., 1988. A decision support system for purchasing management of large projects. Operations Research 36 (6),
882890.
Roodhooft, F., Konings, J., 1997. Vendor selection and evaluation: an activity based costing approach. European Journal of
Operational Research 96, 97102.
Rosenthal, E.C., Zydiak, J.L., Chaudhry, S.S., 1995. Vendor selection with bundling. Decision Sciences 26 (1), 3548.
Sadrian, A.A., Yoon, Y.S., 1994. A procurement decision support system in business volume discount environments. Operations
Research 42 (1), 1423.
Shank, J.K., Govindarajan, V., 1992. Strategic cost management: The value chain perspective. Journal of Management Accounting
Research 4, 179197.
Schrage, L., 1998. Optimization Modeling with LINGO. Lindo Systems inc., Chicago, USA.
Smytka, D.L., Clemens, M.W., 1993. Total cost supplier selection model: A case study. International Journal of Purchasing and
Materials Management, winter, 4249.
Soukoup, W.R., 1987. Supplier selection strategies. Journal of Purchasing and Materials Management, summer, 712.
Thompson, K., 1990. Vendor prole analysis. Journal of Purchasing and Materials Management, winter, 1118.
Timmerman, E., 1986. An approach to vendor performance evaluation. Journal of Purchasing and Supply Management 1, 2732.
Turner, I., 1988. An independent system for the evaluation of contract tenders. Journal of the Operational Research Society 39 (6),
551561.
Turney, P.B.B., 1990. Ten myths that create barriers to the implementation of activity based cost systems. Journal of Cost
Management, Spring, 2432.
Weber, C.A., 1991. A decision support system using multicriteria techniques for vendor selection, Doctoral dissertation. Ohio State
University, Columbus, OH.
Weber, C.A., Current, J.R., 1993. A multiobjective approach to vendor selection. European Journal of Operational Research 68, 173
184.
Weber, C.A., Current, J.R., Benton, W.C., 1991. Vendor selection criteria and methods. European Journal of Operational Research
50, 218.
Weber, C.A., Current, J.R., Desai, A., 1998. Non cooperative negotiation strategies for vendor selection. European Journal of
Operational Research 108, 208223.
Weber, C.A., Desai, A., 1996. Determination of paths to vendor market eciency using parallel co-ordinates representation: A
negotiation tool for buyers. European Journal of Operational Research 90, 142155.
Willis, T.H., Huston, C.R., Pohlkamp, F., 1993. Evaluation measures of just-in-time supplier performance. Production and Inventory
Management Journal, 2nd quarter, 15.
Zenz, G.J., 1994. Purchasing and the Management of Materials. Wiley, Chichester.