D/2004/6482/09

Vlerick Leuven Gent Working Paper Series 2004/08

CONSTRUCTING A TOTAL COST OF OWNERSHIP

SUPPLIER SELECTION METHODOLOGY BASED

ON ACTIVITY BASED COSTING

AND MATHEMATICAL PROGRAMMING

Z. DEGRAEVE

EVA LABRO

FILIP ROODHOOFT

Filip.Roodhooft@vlerick.be

2

CONSTRUCTING A TOTAL COST OF OWNERSHIP

SUPPLIER SELECTION METHODOLOGY BASED

ON ACTIVITY BASED COSTING

AND MATHEMATICAL PROGRAMMING

Z. DEGRAEVE

London Business School

EVA LABRO

1

London School of Economics

FILIP ROODHOOFT

Vlerick Leuven Gent Management School,

KU Leuven

running head: Total Cost of Ownership for supplier selection

Contact:

Filip Roodhooft

Vlerick Leuven Gent Management School

Tel: +32 16 32 36 36

Fax: +32 16 32 85 81

Email: Filip.Roodhooft@vlerick.be

1

Corresponding author

3

ABSTRACT

In this paper we elaborate on a Total Cost of Ownership supplier selection methodology that

we have constructed using real life case studies of three different industrial components

groups in a firm. These case studies are presented in this article. Analysing the value chain of

the firm, data on the costs generated by the purchasing policy and on supplier performance are

collected using Activity Based Costing (ABC). Since a spreadsheet cannot encompass all

these costs, let alone optimise the supplier selection and inventory management policy, a

mathematical programming model is used. For a specific component group the combination of

suppliers is selected that minimises the Total Cost of Ownership. TCO takes into account all

costs that the purchase and the subsequent use of a component entail in the entire value chain

of the company. The TCO approach goes beyond minimising purchase price and studies all

costs that occur during the entire life cycle of the item in the organisation. Possible savings of

between 6 and 14% of the total cost of ownership of the current purchasing policy are

obtained for the three cases.

Keywords: Activity Based Costing, mathematical programming, supplier selection,

purchasing

4

INTRODUCTION

Purchasing determines an important part of the competitive position of most firms. It

accounts for 60% to 70% of total expenditures in manufacturing (Herberling, 1993), leads to

long term relationships and influences the activities in the complete value chain of the firm.

However, in both the operations management and operations research literature a lot more

effort has been put into obtaining cost reductions in further stages of the value chain,

especially in increasing production efficiency. Although purchasing probably does not receive

the attention it deserves in Western academic literature, it is a field where large cost

reductions can be obtained, as is shown by the Japanese who have traditionally paid more

attention to this field. The cases reported on here, however, are in a traditional Western firm.

The management accounting literature has recently picked up this inter-firm setting as an

interesting area to study. Seal et al. (1997) present evidence on the role of accounting in

developing a strategic supply partnership in an action research study. Ittner et al. (1999) look

at the effect supplier selection has on company performance and the intervening variables in

this relationship. Cooper and Slagmulder (1999) present a book with case studies of cost

management in the supply chain. Van der Meer-Kooistra and Vosselman (2000) discuss

management control issues in interfirm relationships. Baiman and Rajan (2002) provide an

overview of the incentive issues in inter-firm relationships identified by the incomplete

contracting literature. Dekker (2003) looks at the provision of information to coordinate and

optimise the supply chain in a case study.

Within the purchasing framework, decisions that have to be taken include supplier

selection and determination of order quantities to be placed with these selected suppliers

through time. Supplier selection decisions have a multiple objective character. At least 23

criteria for this selection problem have been identified in the literature (Dickson, 1966;

Weber, Current and Benton, 1991). These include amongst others: net price, quality, delivery,

supplier performance history, capacity, communication systems, service, geographical

location. The problem is how to select suppliers that perform satisfactorily on the desired

dimensions.

Published vendor selection 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 (e.g. Gregory,

1986; Nydick and Hill, 1992; Willis et al., 1993). These rating models are very subjective and

often very sensitive to different rating scales, weights and/or ratings by different people. Total

5

cost approaches ( e.g. Monckza and Trecha, 1988; Smytka and Clemens, 1993) attempt to

quantify all costs related to the selection of a vendor in monetary units but often fail to include

the more qualitative criteria. Mathematical programming models (e.g. Chaudhry et al., 1993;

Current and Weber, 1994; Sadrian and Yoon, 1994) often consider only the more quantitative

criteria.

In this paper, we combine a total cost approach with mathematical programming

because the dimensions of the problem cannot be handled using a simple spreadsheet. We

propose a Management Information System (MIS) that simultaneously treats the supplier

selection and the inventory management decision for multiple components over several time

periods in a mathematical programming model. This MIS is based on Total Cost of

Ownership (TCO) and Activity Based Costing (ABC) information (Degraeve and Roodhooft,

2000) and programmed in LINGO (Schrage, 1998). For a specific component group the

combination of suppliers is selected that minimises the Total Cost of Ownership. TCO takes

into account all costs that the purchase and the subsequent use of a component entail in the

entire value chain of the company (Shank and Govindarajan, 1992). The TCO approach goes

beyond minimising purchase price and studies all costs that occur during the entire life cycle

of the item in the organisation. These include amongst others costs related to service, quality,

delivery, administration, inventory holding, communication and defects. ABC makes the

quantification of the cost criteria possible.

Several authors have identified TCO analysis as a way to improve purchasing (e.g.

Ellram 1995a, Smytka and Clemens, 1993). Ellram (1995b) writes about the link between

TCO and ABC, but in our opinion has a fairly limited view on ABC. She asserts that, in

purchasing, ABC focuses on the internal administrative costs of the purchasing department

and assigns costs to the product, customer or distribution channel. In our opinion, however,

ABC can also study costs in other departments that can be influenced by the purchasing policy

and “the supplier selection policy” can be used as a cost object instead of the more traditional

cost objects mentioned by Ellram.

Mathematical programming (MP) techniques have been applied to purchasing issues

frequently, mainly in the domain of determining order quantities, specifically in environments

where complex discounts are offered by suppliers (e.g. Benton, 1991; Benton and Park, 1995;

Chaudhry et al., 1993; Sadrian and Yoon, 1993; Rosenthal et al., 1995), but also in supplier

selection (e.g. Akinc, 1993; Current and Weber, 1994).

Shapiro (1999) argues that mathematical programming models can serve as a template

for cost and resource data to be extracted by ABC methods. Our use of MP is classified in

6

what he calls a type 4 model where the objective is to minimise cost over supply chain

resources in a multiple period, deterministic environment. To our knowledge, Degraeve and

Roodhooft (2000) were the first to make the link between TCO, ABC and mathematical

programming for supplier selection. They illustrate this link on an imaginary case. Degraeve

and Roodhooft (1999, 1998 resp.) illustrate the use of this method on rather small case studies

of buying heating electrodes and ball bearings respectively, at Cockerill Sambre, a Belgian

firm in the steel industry.

The incremental contribution of this paper is threefold. Firstly, this is the first time that

the management accounting aspect of the methodology is elaborated on and worked out

extensively. Previous papers do not describe the value chain and ABC analysis linked to the

specificity of the different product groups, nor do these papers discuss how the data are

collected and what the problems related to this aspect are. We show how the mathematical

programming model serves to define and structure the decision problem at hand. We obtain

very good results compared to the current purchasing policy because of our thorough data

collection and ABC analysis within this structure.

Secondly, this paper situates the work specifically within the context of the

constructive case study research methodology in management accounting (Kasanen, Lukka,

Siitonen, 1993), thereby indeed focusing more on the process aspects of the case study. It

builds on the previous papers and proves that the financial results of the previous cases can be

transferred to other component groups in another firm thereby generalizing previous results.

Thirdly, the model in this paper is far more complex than those in the previous papers

on several dimensions. Because of the extended value chain and ABC analysis, the number of

criteria and different costs considered are substantially larger. Also the monetary amounts

involved are larger (16,011,000 euro vs. 200,000 (Degraeve and Roodhooft, 1999) and

1,303,000 euro (Degraeve and Roodhooft, 1998)). The complexity increase is also indicated

by the size of the component groups (1,052 different component types vs. 1 and 33

respectively) and the supplier base from which to select (88 suppliers vs. 3 and 6

respectively). This results in a substantially increased number of variables and constraints in

the mathematical programming model. Also, the overlap between the suppliers and

component types is substantial, preventing us from using a decomposition method to solve the

problem.

Using our theoretical ABC framework for supplier selection we developed a MIS for a

division in Europe of a world-wide telecommunications firm that is one of the leaders in the

high speed access and transmission market. The firm has 116,000 employees world-wide and

7

is represented in 130 countries. Global sales amount to 23 billion Euro. The division studied

offers a complete portfolio of the world-wide firm from micro-electronics and

telecommunications to cables and components and employs 4600 people in the country of its

location. The purchasing department’s mission statement includes “cost of ownership” as an

expectation of their (internal) customers:

To explore the market and to purchase and deliver […] conform to the

requirements, with a maximum of flexibility and reliability, with a competitive

“Cost of Ownership”, continuously.

(stress in original documentation)

To achieve this goal the firm uses a vendor rating system that takes price, technology,

quality, flexibility and delivery reliability into account. An effort is made to buy as much as

possible from suppliers with a preferred status. An Economic Order Quantity (EOQ) model

calculates the order points, but does not link this decision to the supplier selection decision.

The MIS is developed for three major bought-in product groups: resistors,

transformers and printed circuit boards (PCB). These component groups are selected because

the relevant criteria and costs differ substantially between them. In this way, the external

validity of the study is increased by constructing a toolbox that is widely applicable to similar

decision problems in different business contexts. We assume that the component groups are

independent from each other and study them separately. Although an occasional supplier

delivers items in more than one of the component groups, these links are negligible and the

total dimensions of the cases prevent us from looking at the three component groups at the

same time. Together, the three component groups account for about 14,000,000 euro in total

costs. The dimensions of the cases studied are vast and involve a considerable amount of

money, as shown in Table 1. The first column gives the number of different types of

components used in the firm. The second column states for how many of these types there was

a demand in 1999, the year of study. The third column indicates the number of possible

suppliers. The fourth column gives the current monetary purchasing price in euro.

Insert Table 1 About Here

Prices for different types of electrical components may vary substantially. The 1,729

resistor types are classified in two basic types, thickfilm chips and minimelfs with thin film

8

technology. Minimelfs have a lower temperature coefficient, better current-noise

characteristics and a better stability with respect to overheating, but are more expensive.

Prices quoted for transformers are a function of their core type, the number of windings, the

quantity ordered and insulation requirements. The production cost of the Printed Circuit Board

(PCB) suppliers depends, amongst others, on type of material, number of layers, drill size,

finishes, density, thickness and board area. Asian PCB suppliers are cheaper but have a longer

lead time, provide less service and do not have special technologies available.

The remainder of the paper elaborates on the supplier selection methodology

developed. Section 2 explores the activities performed in the value chain of the purchasing

firm. Section 3 explains how Activity Based Costing data were gathered to cost out these

activities and which types of information are collected about the performance of the suppliers

on the different supplier selection criteria that generate costs in the value chain of the firm.

Section 4 shows how the data are translated into the objective function and constraints of the

mathematical programming models. The next section interprets the results and discusses

strategic insights for the purchasing policy. The last section concludes.

THE VALUE CHAIN AND ACTIVITIES

We study the activities in the value chain of the firm that relate to the purchasing

policy in the first step of the vendor selection methodology. These can either be activities of

the purchasing department itself or activities further down the value chain that are influenced

by policy decisions made by the purchasing department. Figure 1 shows the activities, where

they are situated in the value chain and how they relate back to the purchasing policy in the

case study firm. It is important to perform this value chain analysis, as these activities and

their costs will later be modelled in the mathematical programming model. The rest of this

section describes the activities in the value chain of the firm.

Insert Figure 1 About Here

The purchasing engineer responsible for a component group negotiates with the

suppliers on amongst others price, discounts, quality, lead time and follows up the relationship

to sustain the supplier in the supply base. When the supplier is new to the firm or when quality

9

problems arise too frequently, the quality team sometimes performs a quality audit on the

supplier’s site.

Ordering can start once the relationship with the supplier is set up and the supplier is

selected for a certain component type. Depending on the supplier/component combination

orders can be placed through electronic data interchange (EDI) via a private network,

automatic call-off (ACO) on a frame agreement or the manuals by sending a fax.

2

The first

time an order is placed for a specific PCB with a supplier, a tooling cost might be charged by

the supplier to cover the supplier’s one/off costs on films, drilling information and electrical

testing. Suppliers with a shorter lead-time are more flexible in that they can accommodate to a

sudden change in demand on a shorter notice period and thus agree a delivery date that is

nearer in the future than other suppliers can.

3

A supplier’s delivery reliability depends on the

history of early and late deliveries around the agreed delivery date. Importing documents have

to be filled out and import duty has to be paid when the component is ordered with a supplier

outside the European Community.

Then the receiving department receives the delivery and inspects it together with the

inspecting department.

4

When no irregularities are discovered during the inspection, the

supplier accounting for the delivery is done and the invoice is paid. For transformers and

PCBs some suppliers offer component specific discounts on prices for larger orders and this

discount may rise with the quantity ordered. Some transformers and PCB suppliers add a lot

charge to the invoice. Payment delays typically range from cash payment to 60 days delay,

with 0 to 3% payment discounts. However, when a defect is discovered in inspection,

components are either sent back to the vendor who will send a credit note or will replace them

at his expense, or they are thrown away at the firm’s own expense. When the supplier replaces

2

The minimum order quantity and the lot size have to be adhered to when ordering. Orders for a component thus

have to exceed the minimum order quantity for that component with that supplier and be a multiple of the lot

size. As a rule, the lot size is always lower than or equal to the minimum order quantity.

3

Asian suppliers generally have a longer lead-time than European and American suppliers.

4

Different sorts of inspections are used, depending on the inspection class in which the supplier/component

combination is allocated and resulting in more or less time consuming inspection activities. For purchases with

certified suppliers, the receiving department may release the components without any quality verification. The

trust in these suppliers’ quality systems makes extra inspection superfluous, as the details on the specifications,

the level of quality, the criteria for acceptation of the delivery, the supplier’s auditable quality plan and the

markings on the packaging are agreed on in writing in the quality agreement. Other components are inspected

visually. A skip lot inspection may be performed for components that are delivered frequently. In this case the

first five deliveries and afterwards every fifth batch are each checked taking a sample, whereas the other four are

only checked visually. In a few exceptional cases, the reception department releases transformers and PCBs

delivered by uncertified suppliers without further inspection because their impact on business processes is

considered small. For the odd resistor delivery only the labels on the packaging are compared with the ones on

the travel documents without opening the packaging. Occasionally, every PCB lot is checked using a sample

from each lot. Some special PCBs are sent for verification to the engineer that ordered the component.

10

the components, they go through the whole cycle of importing, receiving and inspecting

again. After a satisfactory inspection the components are transported to the warehouse where

they are held in inventory until the production planning triggers a demand for the component

on the production floor. The components are used to manufacture more complex electronic

components or end-user products that are sold by the marketing department. However, some

defective components that have slipped through incoming inspection turn up during

production. This causes a lot of extra work to troubleshoot the problem, complain to the

supplier, repair and re-test the component. For PCBs the cost of discovering a problem in this

phase in the value chain is the highest because usually the entire expensive PCB has to be

thrown away and other components already fixed on it cannot easily be salvaged. Sold

products are delivered to the customer who, upon discovering a defect in this final phase of

the value chain, files a complaint that results in the after sales department investigating the

problem and writing an outgoing credit note.

5

DATA COLLECTION

The developments in ABC and the integration of these costing systems with company

wide information systems could enable us to collect all necessary data on activities and

supplier performance. However, in the case study firm, three data collection problems had to

be overcome.

Firstly, cost accountants in the firm put effort into defining activities, several non-unit

cost drivers such as throughput time and orders are used, and the company’s head of cost

accounting gives presentations about ABC and how it is applied in the company. The cost

accounting system is mainly used for variance analysis between the budgeted and the actual

figures as well as for calculation of the tariff for services that are provided internally, such as

information technology, training, and accounting. The company clearly expresses the wish to

be on the forefront of developments in this area by applying ABC.

But, in our opinion, the company is merely on the way to developing an ABC costing

system. The basic features of the accounting system remain of the standard costing type,

although a high level of detail is visible. The firm defines about 1500 different resource

5

The external customers of this firm assess only 1.7% of all defects, while the other complaints come from

internal customers in the production department. The analysis of these external customer complaints over the

year studied shows that none of them relate back to problems with the original component bought. Instead, they

are due to faults in the production process or wrong deliveries.

11

categories and 400 cost pools that are related to geographical locations, departments or

products. Strangely, purchasing department overhead and the cost pool related to reception

and inspection are allocated using a normal distribution over groups of suppliers, rated

excellent, mediocre and bad on a combination of indices on delivery, quality and flexibility, as

a percentage of the purchasing volume with these suppliers. Thus it is assumed that there will

always be a specific percentage of each category of suppliers and that purchasing volume is

the overall cost driver. Therefore, especially for this purchasing application of ABC, the

company required lots of assistance from the authors.

Secondly, the company has several databases that are not always integrated. The first

is the accounting database, discussed in the previous paragraph. The second is the Purchasing

Management System (PMS), which contains data on all component types and suppliers. Next,

the Material Requirements Planning (MRP) provides forecasted demand figures. An economic

order quantity (EOQ) calculation system that is directly fed by the MRP was put in place five

years ago. The cost updated figures are, however, not automatically plugged into this EOQ

model and have not been changed in five years. When the accountant participating in this

study was made aware of this during the process of the development of the vendor selection

model, he undertook the necessary steps to update the cost figures in the existing EOQ model.

Besides these four systems, some purchasing engineers make use of their own spreadsheets.

On some occasions, we discovered that the data in these spreadsheets did not correspond to

the data in the overall PMS system.

Thirdly, there was a big turnover of personnel involved in the study. Several people

left for other firms, including the original champion of our study in the company. Other

personnel took up completely different functions within the same company.

The process of collecting the data on costs and supplier performance and refining the

costing system to reflect ABC principles was consequently time consuming as we had to

consult several databases, always via a variety of company personnel, in order to set up our

own ABC system and sort out discrepancies between data in the different files along the way.

The most problematic discrepancy concerned the vendor lead time data, where purchasing

managers’ own spreadsheets were much more up to date than the EOQ system and mostly

showed shorter possible vendor lead times. The remainder of this section describes how we

proceeded with these difficult tasks.

First the resources available to perform all the activities discussed in the previous

section are examined. An example is the gross wage of the inspectors. Resource drivers

establish a relationship between these resources and the activities. We checked, for example,

12

how much time the inspectors spend on a skip lot inspection or on a full inspection. Personnel

in the firm was asked to participate in several time estimate studies. Some resources are linked

directly to the activity and need not be assigned through a resource driver. For example, the

yearly EDI service fee is exclusively related to the EDI ordering activity. Columns 1, 2 and 3

of Table 2 indicate the direct or indirect link between resources and activities and shows

which resource drivers are used in the latter case.

Insert Table 2 About Here

Once the cost of performing an activity is calculated, activity drivers that determine

the total cost of the purchasing policy are searched, using a cost hierarchy with several

activity levels: supplier-, component-, order-, batch- and unit-level. The first hierarchical level

describes costs incurred and conditions imposed whenever the purchasing company actually

uses the supplier over the decision horizon. Costs on the supplier level include a quality audit

cost incurred by the buyer for the evaluation of a supplier and the cost of a dedicated

purchasing manager. This purchasing manager is responsible for both setting up the

relationship with the supplier (e.g. writing up an overall quality agreement) and following up

the relationship. The component level indicates costs incurred whenever the firm needs to buy

this component. Tooling costs for the PCBs are incurred on this level as they are only charged

the first time that the component is ordered with the supplier. Tooling costs vary with the

supplier/PCB combination and might even be non existent for some combinations. 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 ordering and

invoicing. At batch level the firm incurs costs each time a batch is delivered e.g. costs for

reception, inspection, material handling, internal failure (components fail during production)

and late delivery of the batch. At the unit level we find costs incurred and conditions imposed

related to the units of the components for which the procurement decision has to be made, for

example, price, external failure (a component fails when used by the customer) and inventory

holding due to early delivery. The three cases studied illustrate that the ABC hierarchy is case

dependent, as is suggested in the literature (Ittner, Larcker and Randall, 1997). For the resistor

case, a hierarchy with only three levels, i.e. supplier, batch and unit, is used. Since an order

for transformers or PCBs can include more than one type of component and the bought-in

products are delivered per batch of the same component, we add an order level in these cases.

13

We include a component level in the hierarchy for the PCB case, as for some suppliers tooling

costs are incurred the first time a specific PCB is ordered to cover their costs on films, drilling

information and electrical testing. Table 3 shows how the hierarchy differs from case to case

and on what levels the costs are incurred.

Insert Table 3 About Here

It is important to make this classification of activities into separate levels since the

overall process driver for each level of activity, (1) number of suppliers, (2) number of

components, (3) number of orders, (4) number of batches and (5) number of units procured, is

assumed independent of the activities in other hierarchical levels. Column 4 of Table 2 shows

the more detailed cost pools at which level the information was gathered. From the variables

in the mathematical programming model, the level of variability of these details becomes

clear: per supplier (e.g. import duty dependent on location of supplier insider or outside the

European Community), per supplier-component combination (e.g. for some components an

automatic order through the automatic call off system is possible, while the same supplier may

only accept a fax order for other components) or only dependent on the purchasing firm (e.g.

material handling). Remark, however, that the purchasing firm can still work on the efficiency

and effectiveness improvement of the latter activities, or try to eliminate them when they are

non-value-adding activities such as inventory holding.

In this way, all costs caused by the selection of suppliers and the placement of orders

with them can be determined. Column 5 of Table 2 shows the process drivers that drive the

usage of activities by the supplier selection policy. These process drivers determine the level

in the ABC hierarchy where the costs are incurred and will become the decision variables in

the mathematical programming models.

In the next step, information is gathered on supplier performance at the detailed level

of the cost pools and also data on prices, quantity discounts, supplier’s lead time, tooling

costs, minimum order quantity, lot size as well as probabilities of detecting default in

inspection, production or by the external customer, are collected.

Before we proceed, an important caveat is in order. Applying Activity Based Costing

assumes that the costs are linear (or step-linear) with the cost drivers. Research (e.g. Noreen

(1991), Bromwich (1997), Maher and Marais (1998)), however, has shown that the conditions

under which ABC provides accurate costs are rather stringent and in some cases hard to meet,

14

especially when resources are provided on a joint and indivisible basis. Noreen and

Soderstrom (1994) empirically test this linearity assumption for different categories of

overheads and conclude that it is often not met. Datar and Gupta (1994) look at other possible

errors in costing systems in general (not just ABC) and show that there exists a trade off

between reducing specification and aggregation error (which is often used as a justification for

implementation of ABC systems) and increasing measurement error. We do acknowledge all

these possible problems with any application of ABC in general and ours in particular. We

would, however, argue that the use of ABC is already a leap forwards as it approximates the

linearity of the cost functions much better than the traditional volume related approaches, by

using a cost hierarchy where costs become variable at different levels. Costs that were

previously considered fixed or falsely considered variable at the unit level, can now become

variable at one of the other levels in the hierarchy. For our case in particular we have three

further reasons why the use of ABC may not be that problematic. Firstly, we have made a

deliberate attempt to reduce measurement error in units of allocation bases (Datar and Gupta,

1994) by not asking for too detailed information of personnel in their time allocation

estimates. Secondly, as you will learn from our result section, important parts of the possible

savings (between 3 and 11% of TCO of the current purchasing policy) are immediate cash

savings on price for which there cannot exist any measurement or accounting error. We

acknowledge, however, that the other part of the savings may require a thorough re-

engineering exercise to actually get rid of the freed up capacity or put it to an alternative use.

Thirdly, and most importantly, no joint resources are (or needed to be) included in our ABC

exercise since only those resources for which the resource consumption is different if different

suppliers are used are included in the model, as it is our objective to select those suppliers that

minimise TCO. Because of the focused scale of our ABC exercise, we did not have to deal

with joint resources such as investments in marketing and the brand name of the firm, as they

do not vary with the supplier selection policy, which is our cost object.

THE MATHEMATICAL PROGRAMMING MODEL

It is impossible to optimise the supplier selection and inventory management decision

taking all the relevant costs throughout the entire value chain of the firm into account in a

simple spreadsheet. Therefore, we develop mathematical programming models to determine

an optimum sourcing strategy for the different component groups. The models generate a

purchasing policy that minimises the Total Cost of Ownership taking into account constraints

15

relevant to the problem. As a result, the quantification of the vendor selection criteria and the

trade-off between them is no longer a problem because the objective function is defined as the

TCO related to the purchasing decision, and the supplier selection criteria are weighted by

their respective ABC costs.

Typically, a mathematical programming model consists of two main building blocks:

the objective and the constraints. The objective of the model is to minimise the Total Cost of

Ownership of the supplier selection and ordering policy for the decision period of the year. As

discussed in the previous section, the structure of the models is based on case specific ABC

hierarchies. This is shown in the objective function as it reflects net prices and resources

consumed by the activities in the three to five hierarchical levels distinguished: supplier,

component, order, batch and unit level. Subsequently, the mathematical programming model

defines the costs incurred on each of these levels and establishes cost drivers as decision

variables on all of these levels. The most important constraint for this procurement model is

that demand for each component in each time period has to be satisfied.

A more detailed explanation of the mathematical programming models, the exact

mathematical notation and information on the solving procedure can be found in the appendix

to the paper.

RESULTS

We have made an extensive comparison of our suggested purchasing policy with the

actual purchasing policy used. As we are not allowed to make the actual data available due to

confidentiality reasons, we present the results in Table 6 as percentages. The first row

indicates the possible savings as a percentage of the TCO of the current policy. The second

row gives the approximate TCO figures for the different component groups in euro. The next

eight rows show the hierarchy of costs for the optimal purchasing policy, as percentages of the

optimal TCO. The final seven rows indicate how the cost hierarchy is built up for the current

policy, as percentages of the optimal TCO.

Insert Table 6 About Here-

The purchasing policy proposed is able to save 14%, 6% and 11% on TCO on the

component groups resistors, transformers and PCBs respectively, compared to the current

16

purchasing policy

6

. As the benefits of applying the vendor selection model can be measured in

terms of the possible savings compared to the current purchasing policy, this methodology

overcomes the concern raised by Foster and Gupta (1997) about the difficulty of quantifying

benefits if new management accounting systems have to stand the cost/benefit test.

Several strategic insights can be gained form the analysis of the data and the solving of

the mathematical programming model.

As is to be expected in any purchasing application of ABC, the cost structure is unit

level dominated, since the whole turnover is taken into account on this level. Price, a unit

level cost, remains an important component of the TCO, in the optimal case making up

between 92 and 98% of TCO. As most of the purchasing entailed activities (as defined in

Figure 1) that have cost drivers at a higher level in the hierarchy are non value adding

activities, a cost structure dominated by price is good business practice in this case. Most of

the costly value adding activities such as the different steps in the production process were not

included in the analysis as supplier performance does not make a difference to the use of their

activity drivers. It would, however, be this type of activities that might shift the cost structure

from unit level dominated to batch level dominated. The dominance of unit level illustrates

the importance of getting the unit level costs right, also in an ABC environment where several

other levels are studied. You can also see in table 6 that the optimal policy cuts down on these

non-value added activities compared to the current policy.

Most of the possible savings also lie at unit level. Immediate cash savings could

amount to savings of 11.5%, 3% and 9% respectively by selecting a supplier with a lower

price and making optimal use of component specific discounts for transformers and especially

PCBs. We acknowledge that this is not an effect directly related to our ABC implementation,

but more a consequence of the operations research aspect of our methodology that brings in

more structure and objectivity. Using this TCO model, the selection of these lower price

suppliers can now be made with the assurance that quality and other costs are taken into

account and that the overall effect on TCO is positive. Almost all components have a single

6

These possible savings percentages are calculated for the last year for which full data were available. The TCO

model is used to retrospectively calculate the cost of the purchasing policy for that year by fixing all the decision

variables (when what was bought from which supplier) in the model to the values they took in that particular

year’s purchasing policy. In this way, we can calculate the TCO of the purchasing policy that was executed that

year (termed “current policy”). We then run the mathematical programming model again, this time to solve for

the optimal policy and compare the total cost of both (savings indicated as a percentage of the TCO of the

current policy). These savings could have been obtained had the company used the new method to determine

purchasing policy in that year instead of using the purchasing policy they actually implemented. Because the

firm faces a quite stable demand environment for these component groups, we can predict a level of savings of a

similar size for the next year.

17

source that is clearly better than the other possible suppliers. The current policy increases

TCO by splitting orders of the same component over several suppliers and therefore does not

make optimal use of the available quantity discounts.

In the rest of this section we report on non-cash savings that amount to 2.7%, 3% and

1.7% respectively. These are possibilities to save on resources, which would require a re-

engineering exercise to turn them into cash savings. Alternative allocations and selling off of

resources would need to be considered.

For transformers and PCBs, substantial savings on inventory holding costs (also on

unit level) are possible by ordering with suppliers who do not have a record of early

deliveries, and by placing orders just in time for the suppliers lead-time to be sufficient to

deliver the component exactly when needed. This saves a lot on warehousing costs that do not

add any value to the component. The savings we find here are partially due to the main

purchasing information system (PMS) not being updated regularly with respect to vendor lead

times, that had a tendency to decrease over time, and were only correctly written down in the

respective purchasing engineers own spreadsheets that were not linked to PMS. As a result,

the automatically placed orders were placed too early. Part of these inventory holding savings

could also be considered cash savings as there is an one-off freeing up of working capital due

to a lower inventory level. For components with a low unit price such as resistors, inventory

holding costs already make up a smaller percentage of the cost structure.

Savings can also be made on the batch level, by reducing quality problems for

transformers and PCBs. In our opinion, the savings created by a smaller percentage of

expensive defects often outweigh the cost of a quality audit. The batch level cost savings for

resistors, for which quality problems are not common, are a result of a policy of less frequent

ordering.

The firm can only make minor savings at order level costs for transformers and PCBs.

Rather surprisingly, the possible way of ordering through EDI, ACO or fax, do not save much

on ordering costs for the time being. The reason for this is that the cost differences between

these ordering techniques are small since the EDI system in place still requires checking every

order confirmation line by line, as the supplier can change quantities and prices without the

purchaser immediately noticing this. This is an example of an area where the ABC results had

an important policy impact as previously management had assumed that EDI was the most

cost-efficient way of ordering and pushed EDI systems take-up with their suppliers. They

decided to first sort out the technical aspects of the EDI system and automate order

18

confirmation checking, before stressing its importance in negotiations with suppliers any

further.

The component level for the PCBs turns out to be insignificant since the tooling costs

that are charged whenever an PCB is ordered for the first time with a particular supplier, are

small relative to the other costs, and often even non existent. Only minor savings are possible.

Because the purchasing engineers spend most of their time on the specifications for the

components, which are independent of the suppliers selected, the supplier level costs do not

make up a very substantial amount in the cost hierarchy. Narrowing down the supplier base

can result in savings at supplier level. The proposed supplier selection policy narrows down

supplier bases from 21 to 17, from 37 to 35 and from 16 to 13 for resistors, transformers and

PCBs respectively.

Since non-price costs make up between 3 and 9 percent of the cost structure, and still

between 2 and 8 percent in the proposed purchasing policy, it would be interesting to

investigate a broader use of vendor managed inventories (VMI), also called consignment

stocks, as this cuts down costs of activities performed in the value chain of the purchasing

firm and eliminates some of these activities. As for now, the firm is working on a pilot project

for VMI with one supplier of a component group - not studied here -. The consignment

inventory is kept at or near the purchasing firm’s site, but the inventory holding responsibility

rests with the supplier as the components remain property of the supplier until the purchasing

firm takes them out in agreed lot sizes. The supplier is responsible for keeping the

components in stock in sufficient quantities to keep production going. His inputs for the

replenishing of the inventory are forecasts directly from the purchasing firm’s MRP planning,

an agreed minimum and maximum stock level and component consumption data given by the

production department on a weekly basis. The value chain of activities related to the

purchasing process can thus be drastically shortened. Ordering is eliminated as the supplier

draws his information directly from the company’s MRP planning. Reception is also

eliminated and incoming inspection is replaced by outgoing inspection. The supplier is

responsible for material handling costs that includes transport to warehouse, removal of the

packaging and shelving the duly labelled components on the assigned locations. The

purchasing firm usually supplies the warehouse, but fire and water hazard insurance and

warehousing personnel costs, also part of the inventory holding cost are for the supplier’s

account. The supplier finds compensation in cost cuts in his own production, a larger share of

the business and increased partnership.

19

In summary, our recommendations to management to reduce the TCO of their supplier

selection policy were the following: narrow down the supplier base; select some lower price

suppliers with the assurance that quality and other costs are taken into account in the model;

make better use of discounts for transformers and PCBs; rely on a single sourcing policy on

the level of each component (not for the whole component group!), on top of penalizing late

deliveries (which is already purchasing policy), also penalize early deliveries as inventory

holding costs are high; reduce quality problems for transformers and PCBs, order less

frequently for resistors, look into the possibility of extending the use of VMI and search for

improvement in the EDI system.

Apart from providing purchasing management with a better supplier selection and

inventory management policy, the model can be used in two other ways. Firstly, the model

can give decision support using scenario analyses dealing with both strategic decision making

and cost management issues. The TCO of alternative procurement strategies can be

calculated, e.g. imposing a minimum or a maximum number of suppliers, excluding a supplier

etc. Management then can decide whether they are willing to pay the increase in TCO

compared to the optimal supplier selection policy to pursue these strategies. Areas can be

identified where internal improvements such as reducing cost driver rates of performing

value-added activities and/or eliminating non-value added activities, such as moving

materials, can generate the highest reduction in TCO.

Secondly, also areas where external improvements by suppliers are able to generate

decreases in TCO can be pinpointed. The model can then be used as a negotiation tool with

suppliers since proposals of discounts, quality improvements, lead-time reduction etc. made

by suppliers can be easily assessed. This clear communication on what drives costs in the

purchasing firm 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 that might lead to buyer-supplier partnerships.

CONCLUSION

In this paper we develop a Total Cost of Ownership supplier selection methodology

using Activity Based Costing data. In a first step, the activities in the value chain that relate to

the purchasing policy are analysed. Next, resources available to perform all these activities are

examined and resource drivers linking them are established. Once the costs of performing the

activities are calculated, activity drivers that determine the total cost of the purchasing policy

20

are defined, using case dependent cost hierarchies with three to five levels. Then, information

is gathered on supplier performance on these activity drivers. Since a spreadsheet cannot

encompass all these costs, let alone optimise the supplier selection and inventory management

policy, mathematical programming models minimising the TCO of the purchasing decision

are programmed and solved with a stepwise procedure. The ABC hierarchy forms the

backbone of the mathematical programming model, with decision variables on each level of

the hierarchy. Possible savings of between 6 and 14% are obtained for the three cases.

Along the way, several other lessons for accounting were learned. Firstly, the problem

of ABC not being applicable to joint costs (Noreen, 1991) does not need to be an issue in

every case study. Here, the focus of the case was such that we did not need to include any

joint costs since they do not vary with our cost object, the supplier selection policy. Secondly,

we illustrate that the ABC hierarchy is case dependent, even for several component groups

within the same firm. Thirdly, people may hold subjective beliefs on rankings of costs of

activities that may actually not be true. In this particular case, the belief that EDI ordering was

(a lot) cheaper than ordering via the traditional fax way turned out to be wrong when the

correct figures where put into the equation. Fourthly, from an information systems point of

view, we can re-iterate the need for regular updating of figures in all the decision models (as

e.g. in this case in the EOQ model) to ensure correct decision-making and integration of all

the databases used to avoid discrepancies between data. Fifthly, this case study illustrates how

survey evidence on the use of modern cost accounting concepts such as ABC may be

positively biased. We think that, would management accountants within this firm have been

asked to fill out such a survey, they would have ticked the ABC box. However, having dug

into the details of their costing system, we would argue that a full-blown ABC system was not

yet put in place.

There are several avenues for future research. Firstly, the research reported on here

studies off the shelve items. The biggest part of the purchasing entailed activities in the value

chain that are dependent on performance of the suppliers are non value adding and thus the

cost structure is dominated by unit level, although the savings generated by the ABC exercise

on the higher levels are still important. It would, however, be interesting to build a similar

model for a case study where more involvement of the suppliers is required in for example

research and design. More value adding activities that are influenced by the suppliers would

be included in the purchasing entailed value chain and higher ABC levels would gain in

importance. Secondly, research into the quantification of opportunity costs of lost sales due to

quality problems or late deliveries is very difficult but would be very useful in decision-

21

making practice. Ittner (1999) reports on a few companies that make an effort on calculating

quality related opportunity costs and quantifying the magnitude of lost sales.

22

REFERENCES

Akinc, U. (1993). ‘Selecting a set of vendors in a manufacturing environment’. Journal of

Operations Management, 11, 107-122.

Baiman, S., Rajan, M.V. (2002). ‘Incentive issues in inter-firm relationships’. Accounting,

Organizations and Society, 27, 213-238.

Benton, W.C. (1991). ‘Quantity discounts decisions under conditions of multiple items,

multiple suppliers and resource limitations’. International Journal of Production Research,

29(10), 1953-1961.

Benton, W.C., Park S. (1996). ‘A classification of literature on determining the lot size under

quantity discounts’. European Journal of Operational Research, 92, 219-238.

Bromwich, M. (1997). ‘Accounting for overheads. Critique and reforms.’ Acta Universitatis

Upsaliensis. Studia Oeconomiae Negotiorum, 41.

Chaudhry, S.S., Forst, F.G., Zydiak, J.L. (1993). ‘Vendor selection with price breaks’.

European Journal of Operational Research, 70, 52-66.

Cooper,R., Kaplan, R.S. (1998). The design of cost management systems, 2

nd

edition, Prentice

Hall, New Jersey.

Cooper, S., Slagmulder, R. (1999). Supply chain development for the lean enterprise –

interorganizational cost management, Productivity Press, Portland.

Current, J., Weber, C. (1994). ‘Application of facility location modeling constructs to vendor

selection problems’. European Journal of Operational Research, 76, 387-392.

Datar, S.M., Gupta, M. (1994). ‘Aggregation, specification and measurement errors in product

costing’. The Accounting Review, 69, 567-591.

Degraeve, Z., Roodhooft, F. (1998). ‘Determining sourcing strategies: a decision model based

on activity and cost driver information’. Journal of the Operational Research Society, 48(8),

pp.781-789.

23

Degraeve, Z., Roodhooft, F. (1999). ‘Improving the efficiency 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), 42-53.

Degraeve, Z., Roodhooft, F. (2000). ‘A mathematical programming approach for procurement

using Activity Based Costing’. Journal of Business Finance and Accounting, 27(1-2), 69-98.

Dekker, H.C. (2003). ‘Value chain analysis in interfirm relationships: a field study.’

Management Accounting Research, 14(1), 1-23.

Dickson, G.W. (1966). ‘An analysis of vendor selection systems and decisions’. Journal of

Purchasing, 2(1), 5-17.

Ellram, L.M. (1995a). ‘Total Cost of Ownership: an analysis approach for purchasing’.

International Journal of Physical Distribution & Logistics Management, 25(8), 4-23.

Ellram, L.M. (1995b). ‘Activity Based Costing and Total Cost of Ownership: a critical

linkage’. Journal of Cost Management, winter, 22-30.

Foster, G.; Gupta, S.M. (1997). ‘Frontiers of Management Accounting Research’. Journal of

Management Accounting Research, 9, 63-77.

Gregory, R.E. (1986). ‘Source selection: a matrix approach’. Journal of Purchasing and

Materials Management, summer, 24-29.

Herberling, M.E. (1993). ‘The rediscovery of modern purchasing’. International Journal of

Purchasing and Materials Management, 29(4), 48-53.

Ittner, C.D. (1994). ‘Activity-based Costing concepts for quality improvement’. European

Management Journal, 17(5), 492-500.

Ittner, C.D., Larcker, D.F., Nagar, V., Rajan, M.V. (1999). ‘Supplier selection, monitoring

practices and firm performance’. Journal of Accounting and Public Policy, 18(3), 253-281.

Ittner, C.D., Larcker, D.F., Randall, T. (1997). ‘The Activity-Based Cost hierarchy,

production policies and firm profitability’. Journal of Management Accounting Research, 9,

143-162.

24

Kasanen, E., Lukka, K., Siitonen, A. (1993). ‘The constructive approach in management

accounting research’. Journal of Management Accounting Research, 5, 243-264.

Maher, M.W., Marais, M.L. (1998). ‘A field study on the limitations of Activity-Based

Costing when resources are provided on a joint and indivisible basis’. Journal of Accounting

Research, 36(1), 129-142.

Monczka, R.M., Trecha, S.J. (1988). ‘Cost-based supplier performance evaluation’. Journal of

Purchasing and Materials Management, spring, 2-7.

Noreen, E. (1991). ‘Conditions under which Activity-Based Costing systems provide relevant

costs. Journal of Management Accounting Research, 3, 159-168.

Noreen, E., Soderstrom, N. (1994). ‘Are overhead costs strictly proportional to activity?’

Journal of Accounting and Economics, 17(1/2), 255-278.

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, 31-36.

Rosenthal, E.C., Zydiak, J.L., Chaudhry, S.S. (1995). ‘Vendor selection with bundling’.

Decision Sciences, 26(1), 35-48.

Sadrian, A.A., Yoon, Y.S. (1994). ‘A procurement decision support system in business

volume discount environments’. Operations Research, 42(1), 14-23.

Schrage, L. (1998). Optimization Modeling with LINGO. Lindo Systems inc., Chicago, USA.

Seal, W., Cullen, J., Dunlop, A., Berry, T., Ahmed, M. (1997). ‘Enacting a European supply

chain: A case study on the role of management accounting’. Management Accounting

Research, 10(3), 303-322.

Shank, J.K., Govindarajan, V. (1992). ‘Strategic cost management: the value chain

perspective’. Journal of Management Accounting Research, 4, 179-197.

Shapiro, J.F. (1999). ‘On the connections among activity-based costing, mathematical

programming models for analyzing strategic decisions, and the resource-based view of the

firm’. European Journal of Operational Research, 118, 295-314.

25

Smytka, D.L., Clemens, M.W. (1993). ‘Total cost supplier selection model: a case study’.

International Journal of Purchasing and Materials Management, winter, 42-49.

van der Meer-Kooistra, J., Vosselman, E. (2000). ‘Management control of interfirm

transactional relationships: The case of industrial renovation and maintenance.’, Accounting,

Organizations and Society, 25(1), 51-77.

Weber, C.A., Current, J., Benton, W.C. (1991). ‘Vendor selection criteria and methods’.

European Journal of Operational Research, 50, 2-18.

Willis, T.H., Huston, C.R., Pohlkamp, F. (1993). ‘Evaluation measures of just-in-time

supplier performance’. Production and Inventory Management Journal, 2

nd

quarter, 1-5.

Winston, W.L. (1994). Operations Research: applications and algorithms, 3

rd

edition Duxbury

Press, California.

26

APPENDIX

The details of the mathematical programming model.

Before stating the model, we provide a summary of the notation for later reference.

r : symbol referring to the resistors,

t : symbol referring to the transformers,

p : symbol referring to the printed circuit boards (PCBs),

n : index denoting resistors, n=r, transformers, n=t, or PCBs, n=p,

N(n) : set of resistors, n=r, transformers, n=t, or PCBs, n=p, index j,

K : set of monthly time periods, index k,

S(n) : set of suppliers for resistors, n=r, or transformers, n=t, or PCBs, n=p, index i,

M(n)

ij

: set of discount intervals given by supplier i for component j,

"

i

Œ

S(n),

"

j

Œ

N(n),

n=t,p, index m.

The parameters indicate the data required and all are expressed on an annual basis. As discussed in

section 3, the structure of the models is based on the case specific ABC hierarchy. At the first

hierarchical level, the supplier level, the parameters describe costs incurred and conditions imposed

whenever the purchasing company actually uses the supplier over the decision horizon. Unless

otherwise stated, the parameters and expressions are valid for the three models, i.e. for n=r,t,p. The

structure of the models differs from case to case as indicated in those paragraphs that are only valid for

some of the component groups. We consider at supplier level:

qc

i

: quality audit cost incurred by the buyer for the evaluation of supplier i,

"

i

Œ

S(n),

mh

i

: annual hours of a dedicated purchasing manager for supplier i for the time devoted to

managing and negotiating,

"

i

Œ

S(n),

wg : gross hourly wages of the purchasing manager who manages and negotiates with the

suppliers,

mis : minimum number of suppliers to be used,

mas : maximum number of suppliers to be used,

27

slc : total supplier level costs.

For the PCB case we introduce a component level in the hierarchy. The parameters describe the costs

incurred and the conditions imposed whenever the purchasing company actually buys the PCB. For

n=p we consider:

tlc

ij

: tooling cost,

"

i

Œ

S(p),

"

j

Œ

N(p),

clc : total component level costs.

Each type of resistor is ordered separately and also deliveries are per type. However, orders

for transformers and PCBs can include several types of components and are delivered in

batches of the same type. Costs related to both the delivery and ordering of resistors are thus

incurred at batch level, whereas for the latter component groups an order level is introduced to

take ordering costs into account.

For transformers and PCBs we introduce an order level where the parameters indicate cost

incurred and conditions imposed each time an order is placed with a particular supplier. For

n=t,p we consider:

vc : invoice cost per order,

oco : order cost per order for opening order line,

olcs

i

: order level cost for supplier i,

"

i

Œ

S(n),

olc : total order level costs.

The batch level parameters indicate cost incurred and conditions imposed each time a batch is

delivered by a particular supplier. For n=r,t,p we consider:

28

tl

i

: import duty per order from supplier i,

"

i

Œ

S(n) ); tl

i

= 0 for European Union

suppliers,

rc : reception cost per order,

ac : supplier accounting cost per order,

is

ij

: inspection cost per order with supplier i of component j,

"

i

Œ

S(n),

"

j

Œ

N(n),

wr : material handling cost per order in transportation to warehouse and shelving,

rb

i

: cost of returning an order to supplier i,

"

i

Œ

S(n),

ri : cost of re-inspecting a new delivery after a refusal,

pv

i

: probability of refusal at incoming inspection at vendor expense per order with

supplier i,

"

i

Œ

S(n),

ic : cost of incoming credit note,

pi

i

: probability of refusal at incoming inspection with incoming credit note per batch with

supplier i,

"

i

Œ

S(n),

po

i

: probability of refusal at incoming inspection and throwing away of component per

batch of supplier i,

"

i

Œ

S(n),

ts : cost of troubleshooting, repairing and re-testing when defect of component is

discovered during production,

ch : cost of complaint handling,

pif

i

: probability of defect discovered during production per batch of component from

supplier i,

"

i

Œ

S(n),

rp : cost of re-planning the production process,

pl1

i

: probability of a late delivery by supplier i of less than 1 month late per order with

supplier i,

"

i

Œ

S(n),

cl : cost of customer dissatisfaction due to late delivery,

29

pl2

i

: probability of a late delivery by supplier i of more than 1 month late per order with

supplier i,

"

i

Œ

S(n),

blcs

ij

: batch level costs for supplier i and component j,

"

i

Œ

S(n),

"

j

Œ

N(n),

blc : total batch level costs.

For n=t,p we consider:

ocl

ij

: order cost per order line placed with supplier i for component j,

"

i

Œ

S(n),

"

j

Œ

N(n),

lc

ij

: lot charge per batch with supplier i for component j,

"

i

Œ

S(n),

"

j

Œ

N(n).

For n=r we consider:

oc

ij

: order cost per order with supplier i of component j,

"

i

Œ

S(r),

"

j

Œ

N(r).

On the final hierarchical level, the unit level parameters specify costs incurred and conditions

imposed related to the units of the components for which a procurement decision has to be

made. For n=r,t,p we consider:

p

ij

: price of component j with supplier i,

"

i

Œ

S(n),

"

j

Œ

N(n),

dp

i

: price discount as a percentage due to payment delay and conditions,

"

i

Œ

S(n),

purc : total monetary purchasing costs,

ocn : cost of outgoing credit note,

cq : cost of customer dissatisfaction due to quality problems,

pef

i

: probability of defect discovered by external customer per unit of component from

supplier i,

"

i

Œ

S(n),

eqc : total costs related to quality problems discovered by external customers,

h : inventory holding costs per period k as a percentage of the component’s price,

ap

j

: average price of component j,

"

j

Œ

N(n),

ls

ij

: lot size for component j when bought with supplier i,

"

i

Œ

S(n),

"

j

Œ

N(n),

30

ame

i

: average number of months, i.e. the number of time periods k, supplier i is early when

he delivers early,

"

i

Œ

S(n),

pe

i

: probability of early delivery for supplier i,

"

i

Œ

S(n),

invc : inventory costs,

bc : backlog costs, as explained infra.,

mo

ij

: minimum order quantity in number of batches for component j when bought with

supplier i,

"

i

Œ

S(n),

"

j

Œ

N(n),

b

j

: beginning inventory of component j,

"

j

Œ

N(n),

d

jk

: demand for component j in time period k,

"

j

Œ

N(n),

"

k

Œ

K.

Transformers and resistor suppliers often offer component specific discounts when

components are ordered in larger quantities. For n=t,p we consider:

lb

ijm

: minimum quantity to buy in discount interval m set by supplier i for component j,

"

i

Œ

S(n),

"

j

Œ

N(n),

"

m

Œ

M(n)

ij

,

ub

ijm

: maximum quantity to buy in discount interval m set by supplier i for component j,

"

i

Œ

S(n),

"

j

Œ

N(n),

"

m

Œ

M(n)

ij

,

dc

ijm

: price discount as a percentage given by supplier i for component j in discount

interval m,

"

i

Œ

S(n),

"

j

Œ

N(n),

"

m

Œ

M(n)

ij

.

The decision variables can also be subdivided in the same hierarchical levels. The supplier decision

variable models whether or not the purchasing company will use the supplier over the planning

horizon. It is as follows, for n=r,t,p:

z

i

= 1, if we buy from supplier i, 0 otherwise,

"

i

Œ

S(n).

The component level decision variable only exists in the PCB case, for n=p:

31

yk

ij

= 1, if PCB j is ordered at least once with supplier i, 0 otherwise,

"

i

Œ

S(p),

"

j

Œ

N(p).

The order level decision variable only exists in the transformer and PCB cases. For n=t,p:

yj

ik

= 1, if any component is ordered with supplier i in time period k, 0 otherwise,

"

i

Œ

S(n),

"

k

Œ

K.

The batch level decision variable is, for n=r,t,p:

y

ijk

= 1 if component j is ordered by supplier i in time period k, 0 otherwise,

"

i

Œ

S(n),

"

j

Œ

N(n),

"

k

Œ

K.

The unit level decision variables pertain to the units of the components for which a

procurement decision has to be made and are defined as follows, for n=r,t,p:

x

ijk

= number of lot sizes of component j ordered with supplier i in time period k,

"

i

Œ

S(n),

"

j

Œ

N(n),

"

k

Œ

K,

vi

jk

= inventory of component j at the end of time period k,

"

j

Œ

N(n),

"

k

Œ

K.

For the transformer and PCB cases two extra decision variables are introduced to model the

component specific discounts, for n=t,p:

w

ijkm

= 1 if component j is bought with supplier i in discount interval m in time period k, 0

otherwise,

"

i

Œ

S(n),

"

j

Œ

N(n),

"

k

Œ

K,

"

m

Œ

M(n)

ij

,

xw

ijkm

= number of batches of component j ordered with supplier i in discount interval m in

time period k,

"

i

Œ

S(n),

"

j

Œ

N(n),

"

k

Œ

K,

"

m

Œ

M(n)

ij

.

Table 4 summarises how the main decision variables are associated with the hierarchical levels in the

three cases.

32

Insert Table 4 About Here

With the notation given above, the mathematical decision model is described below.

Objective: minimise the Total Cost of Ownership of the supplier selection policy over the time

horizon;

ulc

blc

olc

clc

slc

Min

+

+

+

+

(1)

The objective function (1) reflects net prices and resources consumed by the activities in the

five hierarchical levels distinguished.

Define the supplier level costs, for n=r,t,p;

(

)

( )

?

?

+

=

n

S

i

i

i

i

z

wg

mh

qc

slc

(2)

The supplier level costs are incurred whenever the purchasing company actually uses supplier

i over the planning horizon, i.e. z

i

=1. The time spent by a dedicated purchasing manager on

negotiating, managing and following up the relationship with supplier i can be put to some

alternative use if supplier i is not chosen, i.e. z

i

=0 nor does a quality audit need to be

performed as the supplier is no longer kept in the supply base. What are called supplier level

costs in this paper is often referred to as vendor-sustaining costs in the ABC literature (Cooper

and Kaplan, 1998).

Define the component level costs for the PCB case, n=p;

? ?

? ?

=

)

( )

(

p

S

i

p

N

j

ij

ij

yk

tlc

clc

(3)

33

The component level costs are incurred the first time a specific PCB is ordered with a particular

supplier and consist of tooling costs.

Define the order level costs for the transformers and PCB cases, n=t,p;

? ?

? ?

+

=

)

(

)

(

n

S

i K

k

ik

yj

vc

oco

olc

(4)

The order level costs are incurred in those time periods k an order is placed with supplier i and

are made up of ordering costs for the first order line and invoicing costs.

Define the batch level costs, n=r,t,p;

? ? ?

? ? ?

=

)

(

)

(

n

S

i

n

N

j

K

k

ijk

ij

y

blcs

blc

(5a)

ij

i

i

i

i

i

ij

i

ij

ij

ij

p

po

ci

pi

cv

pv

is

wr

ac

rc

tl

lc

ocl

blcs

+

+

+

+

+

+

+

+

+

=

cl

pl

rp

pl

ifc

pif

i

i

ij

i

2

1

+

+

+

p

t

n

n

N

j

n

S

i

,

),

(

),

(

=

?

?

?

?

(5b1)

ij

i

i

i

i

i

ij

i

ij

ij

p

po

ci

pi

cv

pv

is

wr

vc

ac

rc

tl

oc

blcs

+

+

+

+

+

+

+

+

+

=

cl

pl

rp

pl

ifc

pif

i

i

ij

i

2

1

+

+

+

)

(

),

(

r

N

j

r

S

i

?

?

?

?

(5b2)

wr

ri

rc

tl

rb

cv

i

i

i

+

+

+

+

=

)

(n

S

i

?

?

(5c)

ic

rb

ci

i

i

+

=

)

(n

S

i

?

?

(5d)

ij

ij

ij

ls

p

ch

ts

ifc

+

+

=

)

(

),

(

n

N

j

n

S

i

?

?

?

?

(5e)

The batch level costs are incurred only in those time periods k a batch of component j is

ordered with supplier i resulting in a delivery, i.e. y

ijk

=1. As is indicated in (5b1), (5c), (5d)

and (5e), the batch level costs for the transformers and PCBs are made up of ordering costs

ocl, lot charges lc, import duty tl, receiving costs rc, supplier accounting ac, material handling

34

and shelving wr, inspecting costs is, the cost of discovering a default during incoming

inspection

p

po

ci

pi

cv

pv

+

+

, the cost of a quality problem discovered during production

(internal failure) ifc

pif and re-planning and customer dissatisfaction costs when a delivery is

late

cl

pl

rp

pl

2

1

+

. The cost of discovering a default during incoming inspection consists of

the costs related to refusing a delivery and sending it back at vendor expense cv

pv in which

case the supplier replaces the resistors that will have to be re-inspected, the costs related to

refusing a delivery and receiving a credit note from the supplier ci

pi and the price of

throwing a defect resistor delivery away p

po . Internal failure costs consist of

troubleshooting, repair and retest costs ts, complaint costs ch and the price of the component

ls

p . We assume that the probabilities of discovering a default during incoming inspection

and internal default apply equally over all batches of a component bought-in from a specific

supplier. When a delivery is less than a month late only re-planning costs are incurred, but

when there is more delay, the purchasing firm will have problems in delivering its products to

its own customers. The batch level costs for resistors are very similar to those in the

transformer and PCB cases, except that the order cost per line is now replaced by the full

order cost oc, invoicing costs vc are added and lot charges lc are deleted, as indicated in (5b2),

(5c), (5d) and (5e). Order costs for resistors, order costs per line for transformers and PCBs

and inspecting costs are different for different components j with the same supplier i since

they are dependent upon the type of agreement with the supplier for this specific component.

Note that, in contradiction to EOQ models, part of the inventory related cost is recognised at

batch level as material handling cost per batch in transportation to warehouse and shelving

costs are included on this level.

Define the unit level costs, n=r,t,p;

35

bc

invc

eqc

purc

ulc

+

+

+

=

(6)

Specifically, the unit level costs consist of the monetary purchase cost, the quality costs of

defects discovered by external customers, inventory holding costs and backlog inventory

costs.

Define the annual purchasing costs;

(

)

(

)

? ? ? ?

? ? ? ?

-

-

=

)

(

)

(

)

(

1

1

n

S

i

n

N

j

K

k

n

M

m

ij

i

ijm

ijkm

ij

p

dp

dc

xw

purc

(7)

The annual purchasing costs are equal to the sum of all purchases made from all suppliers,

taking the component specific discounts and the payment delay and discount offered into

account.

Define the external failure costs;

? ? ?

? ? ?

=

)

(

)

(

n

S

i

n

N

j

K

k

ijk

ij

i

x

ls

pef

efc

eqc

(8a)

cq

ocn

ch

efc

+

+

=

(8b)

External failure costs are incurred when external customers of the firm discover a quality

problem. They consist of complaint handling, making an outgoing credit note and cost of

customer dissatisfaction due to quality problems. We assume the probability of external

default applies equally to all units of a specific component bought-in from a specific supplier.

For the cases considered here, however, external complaint records showed that none of the

customer complaints about the final product related back to defect components delivered by

the supplier. These problems were always discovered in earlier stages in the value chain,

36

either in incoming inspection or during production. Thus,

0

=

i

pef

,

).

(n

S

i

?

?

. Therefore

there was no need for us to go into detailed calculation of the opportunity costs of lost sales

due to customer dissatisfaction. However, this would prove a fruitful area for future research.

Define the inventory holding costs;

? ? ?

? ?

? ? ?

? ?

+

=

)

( )

(

)

(

n

S

i

n

n

j K

k

i

i

ijk

ij

j

n

N

j

K

k

jk

j

ame

pe

x

ls

ap

h

vi

ap

h

invc

(9)

The inventory holding cost applies to the total amount of component j held in inventory

during each time period k, denoted by vi

jk

, and to the components that are delivered early and

thus have to be kept in inventory longer than necessary.

Define the backlog inventory costs;

(

)

? ?

? ?

+

=

)

(n

N

j

K

k

jk

bl

cl

rp

bc

(10)

Backlog inventory is a term used in Operations Research (e.g. Winston, 1994, p.870). A

backlog inventory is used whenever the demand for a component is not met in the time period

the demand exists, but only in a later time period. It could be seen as a negative inventory. In

the meantime, costs of re-planning the production process and a very high cost of customer

dissatisfaction due to late delivery are incurred. There is no cost of production standstill as the

case study firm only starts producing when all necessary components are available. A backlog

inventory could for example be used when a supplier who scores excellent on all other TCO

issues doesn’t have a short enough lead time. In this paper, however, we did not attempt to

quantify the opportunity costs related to late delivery and customer satisfaction as this is an

extremely difficult task. As demand in this case study is quite stable and predictable, we

37

plugged in a prohibitively high figure for this cost so that when running the model well in

advance of the time period considered, the optimal solution would never make use of the

backlog inventory and order well in time with a supplier whose lead time is low enough to

cover demand in time. The inclusion of a backlog inventory in the objective function,

however, gives the decision maker the possibility to use the mathematical programming

model also under circumstances of an uncertain demand, where flexibility and the possibility

to deliver on a short lead time become extremely important. In this case, management should

make several estimates of the opportunity cost of late deliveries and customer dissatisfaction

and run the model again under these different scenarios. When there is a sudden change in the

demand that was originally derived from the MRP system, the demand figures d

jk

can be

adapted from that time period k on and all earlier placed orders (before time period k) can be

fixed in the model. When running the optimisation model again, it will choose these suppliers

with a short enough vendor lead time to adapt to the new demand constraints or make use of

backlog inventories that generate an opportunity cost to be included in the TCO.

This concludes the derivation of the objective function. The constraints relevant to the

procurement problem are as follows.

Satisfy demand;

( )

jf

jf

jf

n

S

i

vlt

f

ij

ij

j

d

bl

vi

x

ls

b

ij

=

+

-

+

?

?

-

)

(

)

(n

N

j

?

?

(11a)

( )

jk

jk

jk

jk

n

S

i

vlt

k

ij

ij

jk

d

bl

bl

vi

x

ls

vi

ij

=

-

+

-

+

-

?

-

-

?

1

)

(

1

K

k

n

N

j

?

?

?

?

),

(

\{f} (11b)

The demand for each component in the first time period f, d

jf

, modelled by constraint (11a),

can be satisfied from either beginning inventory b

j

, and/or from purchases from the potential

suppliers,

( )

ij

vlt

f

ij

x

-

, and/or be put in a backlog inventory bl

jf

which is only satisfied in a later

38

time period. The amount that remains is the end-of-period inventory vi

jk

. When purchasing

management wants a component to be delivered in this time period k by supplier i it will have

to place this order in time period k-vlt

ij

. f is equal to the maximum vendor lead time offered by

a supplier of the component group studied. f determines how many time periods in advance

the model has to be solved in a deterministic environment in order not to mount up the

considerable cost of using a backlog inventory. In a stochastic environment where demand is

uncertain, f can be used to fix the current time period and the orders placed before f, and re-

run the model with the new demand data. Constraints (11b) model the demand for each

component j in later time periods, d

jk

and the backlog demand from the previous period, bl

jk-1

,

which adds to the normal demand in time period k. This demand is satisfied either from begin-

of-period inventory, which equals the ending inventory of the previous period vi

jk-1

, and/or

from purchases from potential suppliers,

( )

ij

vlt

k

ij

x

-

, and/or is put in a backlog inventory bl

kj

.

Again, the amount that remains is the end-of-period inventory, vi

jk-1

.

Enforce the bounds on the number of suppliers used;

?

?

=

)

(n

S

i

i

mis

z

(12a)

?

?

=

)

(n

S

i

i

mas

z

(12b)

? ? ?

? ? ?

=

)

(

)

(

n

S

i

n

N

j

K

k

ijk

i

y

z

)

(n

S

i

?

?

(12c)

i

ijk

z

y

=

K

k

n

N

j

n

S

i

?

?

?

?

?

?

),

(

),

(

(12d)

Conditions (12a) and (12b) force the purchasing plan to have at least the minimum number,

mis, and at most the maximum number, mas, of suppliers over the complete time horizon for

each component group. Using constraint (12c), the decision variable z

i

will be equal to 0 if the

39

model suggests not to buy from supplier i, while constraint (12d) forces z

i

to be equal to 1 if

during some time period k an order has been placed with supplier i.

Enforce the proper relationships between x

ijk

and y

ijk

and impose the minimum order quantity;

ijk

ijk

y

M

x

=

K

k

n

N

j

n

S

i

?

?

?

?

?

?

),

(

),

(

(13a)

ijk

ijk

ij

x

y

mio

=

K

k

n

N

j

n

S

i

?

?

?

?

?

?

),

(

),

(

(13b)

If an order is not placed with supplier i in period k, condition (13a) with M a large number

will ensure that the amounts of each component that can be bought from the supplier will

indeed be zero. Since the minimum order quantity mio

ij

is expressed as a number of batches,

condition (13b) forces the batch size to be at least this amount if an order is placed.

Enforce the proper relationships between y

ijk

and yj

ik

for the transformer and PCB case, n=t,p;

?

?

=

)

(n

N

j

ijk

ik

y

yj

K

k

n

S

i

?

?

?

?

),

(

(14a)

?

?

=

)

(n

N

j

ijk

ik

y

yj

nn

K

k

n

S

i

?

?

?

?

),

(

(14b)

with nn the number of components, transformers or PCBs, to be bought. Condition (14a)

ensures that if no component is bought in time period k with supplier i yj

ik

is 0. Condition

(14b) ensures that yj

ik

takes the value of 1 when a component is bought with supplier i in time

period k.

Enforce the proper relationships between y

ijk

and yk

ij

for the PCB case, n=p;

?

?

=

K

k

ijk

ij

y

yk

)

(

),

(

n

N

j

n

S

i

?

?

?

?

(15a)

40

?

?

=

K

k

ijk

ij

y

yk

nk

)

(

),

(

n

N

j

n

S

i

?

?

?

?

(15b)

with nk the number of time periods over the time horizon. Condition (15a) ensures that if,

over the time horizon, component j is not bought with supplier i, yk

ij

is 0. Condition (15b)

ensures that yk

ij

takes the value of 1 when component j is bought with supplier i over the time

horizon.

Model the component specific quantity discounts, for n=t,p;

? ? ? ?

? ? ?

? ? ? ?

? ? ?

=

)

(

)

(

)

(

)

(

)

(

n

S

i

n

N

j

K

k

n

M

m

n

S

i

n

N

j

K

k

ijk

ijkm

ij

x

xw

K

k

n

N

j

n

S

i

?

?

?

?

?

?

),

(

),

(

(16a)

ijkm

ijm

ijkm

ij

w

lb

xw

ls

=

ij

n

M

m

K

k

n

N

j

n

S

i

)

(

,

),

(

),

(

?

?

?

?

?

?

?

?

(16b)

ijkm

ijm

ijkm

ij

w

ub

xw

ls

=

ij

n

M

m

K

k

n

N

j

n

S

i

)

(

,

),

(

),

(

?

?

?

?

?

?

?

?

(16c)

i

n

S

i

n

N

j

K

k

n

M

m

ijkm

z

w

ij

=

? ? ? ?

? ? ? ?

)

(

)

(

)

(

K

k

n

N

j

n

S

i

?

?

?

?

?

?

),

(

),

(

(16d)

Expression (16a) computes the amount bought over all discount intervals. The lower bound on

the amount of component j to buy in the discount intervals is set by constraint (16b), while

constraint (16c) imposes the same condition for the upper bound. Condition (16d) ensures that

we cannot obtain discounts on a component if we do not buy anything from the supplier. The

discounting percentage is then applied in equation (7).

Integrality and nonnegativity;

{ }

1

,

0

?

i

z

)

(n

S

i

?

?

(17a)

41

{ }

1

,

0

?

ij

yk

p

t

n

n

N

j

n

S

i

,

),

(

),

(

=

?

?

?

?

(17b)

{ }

1

,

0

?

ik

yj

K

k

p

S

i

?

?

?

?

),

(

(17c)

{ }

0

,

1

,

0

=

?

ijk

ijk

x

y

K

k

n

N

j

n

S

i

?

?

?

?

?

?

),

(

),

(

(17d)

{ }

0

,

1

,

0

=

?

ijkm

ijkm

xw

w

ij

n

M

m

K

k

n

N

j

n

S

i

)

(

,

),

(

),

(

?

?

?

?

?

?

?

?

(17e)

To conclude the model specification, constraints (17a) through (17e) impose the proper

integrality and nonnegativity conditions that apply to the decision variables.

Model (1) through (17e) is a mixed integer linear programming model.

Table 5 indicates the dimensions of the mathematical programming model, reporting on the

number of variables and constraints, respectively, the mixed integer program uses when the

input is read in.

Insert Table 5 About Here

Problems of a smaller size can be solved straightforwardly with optimising software such as

LINGO (Schrage, 1998) on any IBM-compatible 486 or higher PC in times going from a few

minutes to a few hours. The existing computer technology and software, however, does not

allow to solve the case studies reported in this paper in a straightforward way because of the

very large dimensions.

Therefore we have developed a stepwise procedure to achieve a good approximation to the

optimal supplier selection and inventory management policy while analysing the data. In a

first step, all components for which only a single source exists and for which therefore no

42

better supplier selection is possible, are solved in separate models per supplier to optimise the

ordering policy. Then a cluster matrix is drawn for all components that can be delivered by

more than one supplier, which we called multiple sourcing components in the remainder of

the paper. The cluster matrix indicates how many components the suppliers have in common

and how they are clustered together around component groups. Mathematical programming

models for small clusters of multiple sourcing components without links to other components

or suppliers, can then be solved. The sequence in which the remaining big cluster of suppliers

and multiple sourcing components is solved, starts with the suppliers with the least

components and with the least links to other suppliers, for which the mathematical

programming models can usually be solved for all components of a supplier in one go. Then

we solve the optimising models for part of the components of the bigger suppliers. For each

group of components we include all the possible suppliers for these components, and proceed

until we have solved models for all components. Each time the supplier level costs slcs

i

in the

input of a subsequent mathematical programming model are set to zero if this supplier is

already chosen in an earlier solved model to avoid double counting. For the transformer and

PCB cases also the order level costs olcs

ik

are set to zero when in earlier solved models an

order is already placed with supplier i in time period k, i.e. if

1

=

ik

yj

. All the mathematical

programming models are solved with an optimality tolerance between 0 and 3%. The

optimality tolerance indicates to the branch-and-bound solver in LINGO that it should only

search for integer solutions with objective values at least x % better than the best integer

solution found so far. The results of modifying the search procedure in this way are twofold.

First, on the positive side, solution times can be improved enormously. Second, on the

negative side, the final global solution obtained by LINGO may not be the optimal solution.

However, a solution within x % of the TCO optimum is guaranteed. On larger mixed integer

models like these, the alternative of getting a solution within a few percentage points of the

43

optimum after several minutes of runtime, as opposed to the optimum after several days,

makes the use of an optimality tolerance quite attractive. Using this procedure, the TCO

reached might be slightly higher than the optimal TCO that could be reached if it were

possible to solve the mathematical programming model for the whole component group at

once, because the sequence of the components and the optimality tolerance percentage used

influence the solution obtained. However, possible savings of between 6 and 14% compared

to the current policy of the firm, discussed in the next section, prove this procedure definitely

obtains good results. Going through this solution procedure on a yearly basis is sufficient.

Selected suppliers can then be fixed and smaller models can be solved to generate the order

and inventory policy only. The firm in this case study could decide to continue to use their

existing EOQ model to determine the ordering policy on an operational basis and use the TCO

model for strategic supplier selection decision making.

A simpler ABC hierarchy as the three level resistor hierarchy compared to the five level PCB

hierarchy, does not help to solve the models in these cases more easily. Less hierarchical levels may

lead to less decision variables, but in the cases studied here the number of possible suppliers that can

deliver a component increases when the number of hierarchical levels decreases, and thus the number

of variables is increased in that way again.

44

FIGURE 1

The value chain and activities influenced by the purchasing policy

negotiating & managing

supplier relationship

quality audit

ordering

importing

receiving

production planning

arrangements

defect

supplier accounting

inspecting

material handling

producing

selling

delivery to

complaint handling

defect

defect

customer

paying invoice

(c)

sending back

refusing

inventory holding

troubleshooting

outgoing

credit

(a)

note

(b)

incoming credit note

throwing away

repairing

complaining to

re-testing

supplier

PURCHASING DEP.

RECEIVING

WAREHOUSING

PRODUCTION

MARKETING DISTRIBUTION AFTER SALES

& INSPECTING

& WARRANTIES

45

TABLE 1

Dimensions of the case studies

total number

of different

component types

number of

different

component types

needed in

1999

number of

suppliers

current monetary

purchasing price

in euro

Resistors

1729

660

25

1,225,000

Transformers 543

260

39

8,044,000

PCBs

336

132

24

5,263,000

46

TABLE 2

The ABC details for the supplier selection cases

Resources

Resource drivers

Activities

Detailed activity

cost pools

Activity driver as

decision variable in

mathematical

programming model

gross wages of purchasing

manager

% of time spent

negotiating &

managing relationship

hours negotiating &

managing

# suppliers

gross wages of auditor +

quality engineer

% of time spent

quality audit

0/1 quality audit

# suppliers

% of time spent on EDI orders

% of time spent on ACO orders

EDI orders

gross wages of secretaries,

buyers and purchasing

engineers

% of time spent on

manual orders

ACO orders

yearly EDI service fee +

gateway fee

direct with EDI ordering

computer used for EDI

direct EDI ordering

fax machine

direct with manual ordering

ordering

manual orders

# orders

gross wages of secretaries,

buyers and purchasing

engineers

% of time spent on invoices

invoicing

# orders

import duty

direct with importing

gross wages of secretaries

% of time spent on import declarations

importing

batches from outside

E.U.

# batches

gross wages of accountant % of time spent

supplier accounting

# batches

gross wages of receiving

personnel

% of time spent

receiving

# batches

% of time spent on immediate releases

immediate releases

% of time spent on visual inspections

visual inspections

% of time spent on comparisons of

labels on packaging

comparisons of labels on

packaging

gross wages of inspecting

personnel

% of time spent on skip lot inspections

inspecting

skip lot inspections

# batches

P

U

R

C

H

A

S

I

N

G

P

O

L

I

C

Y

47

% of time spent on engineer

verifications

engineer verifications

% of time spent on sample inspections of every

lot

inspecting equipment

direct with inspecting

sample inspections of

every lot

gross wages of inspecting

personnel

% of time spent on refusing incoming

orders or batches

gross wages of receiving

personnel

% of time spent on refusing incoming

orders or batches

gross wages of purchasing

personnel

% of time spent on refusing incoming

orders or batches

refusing a delivery

# batches

price

direct with throwing away

throwing away

# batches

gross wages of secretaries

% of time spent on sending

back & administration

gross wages of accountant % of time spent on accounting for

sending back

sending back to E.U.

supplier

postage

direct with sending back

export duty to outside E.U.

suppliers

direct with sending back outside E.U.

sending back

sending back to

outside E.U. supplier

# batches

gross wages of accountant % of time spent on incoming credit

notes

incoming credit note

# batches

gross wages of warehousing personnel

% of time spent on transporting orders

or batches to warehouse and shelving

material handling

# batches

gross wages of production

personnel

% of time spent on troubleshooting

after discovery of defect component

troubleshooting

# batches

gross wages of production

personnel

% of time spent on repairing after

discovery of defect component

repairing equipment

direct with repairing

repairing

# batches

gross wages of production

personnel

% of time spent on re-testing after

discovery of defect component

testing equipment

direct

re-testing

# batches

48

gross wages of warehousing personnel

% of time spent on maintaining

inventory

heating costs

m²

warehouse maintenance

m²

fire insurance

m²

opportunity cost

interest % to be gained on risk free investment

obsolescence cost

% of unit price

inventory holding

# units of component

gross wages of personnel in complaint

handling department

% of time spent on complaint handling complaint handling

# units of component

gross wages of personnel in complaint

handling department

% of time spent on making up

outgoing credit note

outgoing credit note

# units of component

inventory holding

# : number of

49

TABLE 3

Activity Based Costs and Hierarchy for the three case studies

resistors

transformers

printed circuit boards

supplier level costs quality audit

negotiating and managing

quality audit

negotiating and managing

quality audit

negotiating and managing

component level costs

N/A.

N/A.

tooling cost

order level costs N/A.

ordering opening line

invoicing

ordering opening line

invoicing

batch level costs ordering

invoicing

importing

supplier accounting

receiving

inspecting

refusing

throwing away

sending back

incoming credit note

material handling to warehouse

late delivery

troubleshooting

repairing

re-testing

ordering subsequent lines

importing

supplier accounting

receiving

inspecting

refusing

throwing away

sending back

incoming credit note

material handling to warehouse

late delivery

troubleshooting

repairing

re-testing

ordering subsequent lines

importing

supplier accounting

receiving

inspecting

refusing

throwing away

sending back

incoming credit note

material handling to warehouse

late delivery

troubleshooting

repairing

re-testing

unit level costs inventory holding (normal & early delivery)

price

complaint handling

outgoing credit note

inventory holding (normal & early delivery)

price

complaint handling

outgoing credit note

inventory holding (normal & early delivery)

price

complaint handling

outgoing credit note

50

TABLE 4

The decision variables

Resistors

Transformers PCBs

Supplier level z

i

z

i

z

i

Component levelN/A.

N/A.

yk

ij

Order level N/A.

yj

ik

yj

ik

Batch level y

ijk

y

ijk

y

ijk

Unit level

x

ijk

x

ijk

x

ijk

51

TABLE 5

Dimensions of the mathematical programming model

number of

variables

number of

constraints

Resistors

117,125

95,231

Transformers 59,268

62,497

PCBs

31,456

28,904

52

TABLE 6

Results

Resistors

Transformers PCBs

Total

possible

savings

as a percentage

of TCO of

current policy

14.26%

5.97%

11.00%

8.49%

optimal TCO in EURO

1,151,000

7,974,000

4,827,000 13,952,000

optimal TCO 100%

100%

100%

100%

SLC

1.76%

0.42%

0.33%

0.50%

CLC

N/A.

N/A.

0.06%

0.02%

OLC

N/A.

0.03%

0.03%

0.03%

BLC

3.96%

1.21%

0.51%

1.19%

ULC

94.82%

98.34%

99.06%

98.25%

PURC

92.82%

97.71%

98.52%

97.58%

optimal policy

as a % of

optimal

TCO

INV

1.46%

0.63%

0.54%

0.67%

SLC

2.28%

0.43%

0.37%

0.56%

CLC

N/A.

N/A.

0.08%

0.03%

OLC

N/A.

0.03%

0.03%

0.03%

BLC

6.61%

2.03%

0.79%

1.98%

ULC

107.75%

103.86%

111.09%

106.68%

PURC

106.33%

100.87%

109.02%

104.14%

current policy

as a % of

optimal

TCO

INV

1.42%

2.99%

2.07%

2.54%

D/2004/6482/09

Vlerick Leuven Gent Working Paper Series 2004/08

CONSTRUCTING A TOTAL COST OF OWNERSHIP

SUPPLIER SELECTION METHODOLOGY BASED

ON ACTIVITY BASED COSTING

AND MATHEMATICAL PROGRAMMING

Z. DEGRAEVE

EVA LABRO

FILIP ROODHOOFT

Filip.Roodhooft@vlerick.be

2

CONSTRUCTING A TOTAL COST OF OWNERSHIP

SUPPLIER SELECTION METHODOLOGY BASED

ON ACTIVITY BASED COSTING

AND MATHEMATICAL PROGRAMMING

Z. DEGRAEVE

London Business School

EVA LABRO

1

London School of Economics

FILIP ROODHOOFT

Vlerick Leuven Gent Management School,

KU Leuven

running head: Total Cost of Ownership for supplier selection

Contact:

Filip Roodhooft

Vlerick Leuven Gent Management School

Tel: +32 16 32 36 36

Fax: +32 16 32 85 81

Email: Filip.Roodhooft@vlerick.be

1

Corresponding author

3

ABSTRACT

In this paper we elaborate on a Total Cost of Ownership supplier selection methodology that

we have constructed using real life case studies of three different industrial components

groups in a firm. These case studies are presented in this article. Analysing the value chain of

the firm, data on the costs generated by the purchasing policy and on supplier performance are

collected using Activity Based Costing (ABC). Since a spreadsheet cannot encompass all

these costs, let alone optimise the supplier selection and inventory management policy, a

mathematical programming model is used. For a specific component group the combination of

suppliers is selected that minimises the Total Cost of Ownership. TCO takes into account all

costs that the purchase and the subsequent use of a component entail in the entire value chain

of the company. The TCO approach goes beyond minimising purchase price and studies all

costs that occur during the entire life cycle of the item in the organisation. Possible savings of

between 6 and 14% of the total cost of ownership of the current purchasing policy are

obtained for the three cases.

Keywords: Activity Based Costing, mathematical programming, supplier selection,

purchasing

4

INTRODUCTION

Purchasing determines an important part of the competitive position of most firms. It

accounts for 60% to 70% of total expenditures in manufacturing (Herberling, 1993), leads to

long term relationships and influences the activities in the complete value chain of the firm.

However, in both the operations management and operations research literature a lot more

effort has been put into obtaining cost reductions in further stages of the value chain,

especially in increasing production efficiency. Although purchasing probably does not receive

the attention it deserves in Western academic literature, it is a field where large cost

reductions can be obtained, as is shown by the Japanese who have traditionally paid more

attention to this field. The cases reported on here, however, are in a traditional Western firm.

The management accounting literature has recently picked up this inter-firm setting as an

interesting area to study. Seal et al. (1997) present evidence on the role of accounting in

developing a strategic supply partnership in an action research study. Ittner et al. (1999) look

at the effect supplier selection has on company performance and the intervening variables in

this relationship. Cooper and Slagmulder (1999) present a book with case studies of cost

management in the supply chain. Van der Meer-Kooistra and Vosselman (2000) discuss

management control issues in interfirm relationships. Baiman and Rajan (2002) provide an

overview of the incentive issues in inter-firm relationships identified by the incomplete

contracting literature. Dekker (2003) looks at the provision of information to coordinate and

optimise the supply chain in a case study.

Within the purchasing framework, decisions that have to be taken include supplier

selection and determination of order quantities to be placed with these selected suppliers

through time. Supplier selection decisions have a multiple objective character. At least 23

criteria for this selection problem have been identified in the literature (Dickson, 1966;

Weber, Current and Benton, 1991). These include amongst others: net price, quality, delivery,

supplier performance history, capacity, communication systems, service, geographical

location. The problem is how to select suppliers that perform satisfactorily on the desired

dimensions.

Published vendor selection 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 (e.g. Gregory,

1986; Nydick and Hill, 1992; Willis et al., 1993). These rating models are very subjective and

often very sensitive to different rating scales, weights and/or ratings by different people. Total

5

cost approaches ( e.g. Monckza and Trecha, 1988; Smytka and Clemens, 1993) attempt to

quantify all costs related to the selection of a vendor in monetary units but often fail to include

the more qualitative criteria. Mathematical programming models (e.g. Chaudhry et al., 1993;

Current and Weber, 1994; Sadrian and Yoon, 1994) often consider only the more quantitative

criteria.

In this paper, we combine a total cost approach with mathematical programming

because the dimensions of the problem cannot be handled using a simple spreadsheet. We

propose a Management Information System (MIS) that simultaneously treats the supplier

selection and the inventory management decision for multiple components over several time

periods in a mathematical programming model. This MIS is based on Total Cost of

Ownership (TCO) and Activity Based Costing (ABC) information (Degraeve and Roodhooft,

2000) and programmed in LINGO (Schrage, 1998). For a specific component group the

combination of suppliers is selected that minimises the Total Cost of Ownership. TCO takes

into account all costs that the purchase and the subsequent use of a component entail in the

entire value chain of the company (Shank and Govindarajan, 1992). The TCO approach goes

beyond minimising purchase price and studies all costs that occur during the entire life cycle

of the item in the organisation. These include amongst others costs related to service, quality,

delivery, administration, inventory holding, communication and defects. ABC makes the

quantification of the cost criteria possible.

Several authors have identified TCO analysis as a way to improve purchasing (e.g.

Ellram 1995a, Smytka and Clemens, 1993). Ellram (1995b) writes about the link between

TCO and ABC, but in our opinion has a fairly limited view on ABC. She asserts that, in

purchasing, ABC focuses on the internal administrative costs of the purchasing department

and assigns costs to the product, customer or distribution channel. In our opinion, however,

ABC can also study costs in other departments that can be influenced by the purchasing policy

and “the supplier selection policy” can be used as a cost object instead of the more traditional

cost objects mentioned by Ellram.

Mathematical programming (MP) techniques have been applied to purchasing issues

frequently, mainly in the domain of determining order quantities, specifically in environments

where complex discounts are offered by suppliers (e.g. Benton, 1991; Benton and Park, 1995;

Chaudhry et al., 1993; Sadrian and Yoon, 1993; Rosenthal et al., 1995), but also in supplier

selection (e.g. Akinc, 1993; Current and Weber, 1994).

Shapiro (1999) argues that mathematical programming models can serve as a template

for cost and resource data to be extracted by ABC methods. Our use of MP is classified in

6

what he calls a type 4 model where the objective is to minimise cost over supply chain

resources in a multiple period, deterministic environment. To our knowledge, Degraeve and

Roodhooft (2000) were the first to make the link between TCO, ABC and mathematical

programming for supplier selection. They illustrate this link on an imaginary case. Degraeve

and Roodhooft (1999, 1998 resp.) illustrate the use of this method on rather small case studies

of buying heating electrodes and ball bearings respectively, at Cockerill Sambre, a Belgian

firm in the steel industry.

The incremental contribution of this paper is threefold. Firstly, this is the first time that

the management accounting aspect of the methodology is elaborated on and worked out

extensively. Previous papers do not describe the value chain and ABC analysis linked to the

specificity of the different product groups, nor do these papers discuss how the data are

collected and what the problems related to this aspect are. We show how the mathematical

programming model serves to define and structure the decision problem at hand. We obtain

very good results compared to the current purchasing policy because of our thorough data

collection and ABC analysis within this structure.

Secondly, this paper situates the work specifically within the context of the

constructive case study research methodology in management accounting (Kasanen, Lukka,

Siitonen, 1993), thereby indeed focusing more on the process aspects of the case study. It

builds on the previous papers and proves that the financial results of the previous cases can be

transferred to other component groups in another firm thereby generalizing previous results.

Thirdly, the model in this paper is far more complex than those in the previous papers

on several dimensions. Because of the extended value chain and ABC analysis, the number of

criteria and different costs considered are substantially larger. Also the monetary amounts

involved are larger (16,011,000 euro vs. 200,000 (Degraeve and Roodhooft, 1999) and

1,303,000 euro (Degraeve and Roodhooft, 1998)). The complexity increase is also indicated

by the size of the component groups (1,052 different component types vs. 1 and 33

respectively) and the supplier base from which to select (88 suppliers vs. 3 and 6

respectively). This results in a substantially increased number of variables and constraints in

the mathematical programming model. Also, the overlap between the suppliers and

component types is substantial, preventing us from using a decomposition method to solve the

problem.

Using our theoretical ABC framework for supplier selection we developed a MIS for a

division in Europe of a world-wide telecommunications firm that is one of the leaders in the

high speed access and transmission market. The firm has 116,000 employees world-wide and

7

is represented in 130 countries. Global sales amount to 23 billion Euro. The division studied

offers a complete portfolio of the world-wide firm from micro-electronics and

telecommunications to cables and components and employs 4600 people in the country of its

location. The purchasing department’s mission statement includes “cost of ownership” as an

expectation of their (internal) customers:

To explore the market and to purchase and deliver […] conform to the

requirements, with a maximum of flexibility and reliability, with a competitive

“Cost of Ownership”, continuously.

(stress in original documentation)

To achieve this goal the firm uses a vendor rating system that takes price, technology,

quality, flexibility and delivery reliability into account. An effort is made to buy as much as

possible from suppliers with a preferred status. An Economic Order Quantity (EOQ) model

calculates the order points, but does not link this decision to the supplier selection decision.

The MIS is developed for three major bought-in product groups: resistors,

transformers and printed circuit boards (PCB). These component groups are selected because

the relevant criteria and costs differ substantially between them. In this way, the external

validity of the study is increased by constructing a toolbox that is widely applicable to similar

decision problems in different business contexts. We assume that the component groups are

independent from each other and study them separately. Although an occasional supplier

delivers items in more than one of the component groups, these links are negligible and the

total dimensions of the cases prevent us from looking at the three component groups at the

same time. Together, the three component groups account for about 14,000,000 euro in total

costs. The dimensions of the cases studied are vast and involve a considerable amount of

money, as shown in Table 1. The first column gives the number of different types of

components used in the firm. The second column states for how many of these types there was

a demand in 1999, the year of study. The third column indicates the number of possible

suppliers. The fourth column gives the current monetary purchasing price in euro.

Insert Table 1 About Here

Prices for different types of electrical components may vary substantially. The 1,729

resistor types are classified in two basic types, thickfilm chips and minimelfs with thin film

8

technology. Minimelfs have a lower temperature coefficient, better current-noise

characteristics and a better stability with respect to overheating, but are more expensive.

Prices quoted for transformers are a function of their core type, the number of windings, the

quantity ordered and insulation requirements. The production cost of the Printed Circuit Board

(PCB) suppliers depends, amongst others, on type of material, number of layers, drill size,

finishes, density, thickness and board area. Asian PCB suppliers are cheaper but have a longer

lead time, provide less service and do not have special technologies available.

The remainder of the paper elaborates on the supplier selection methodology

developed. Section 2 explores the activities performed in the value chain of the purchasing

firm. Section 3 explains how Activity Based Costing data were gathered to cost out these

activities and which types of information are collected about the performance of the suppliers

on the different supplier selection criteria that generate costs in the value chain of the firm.

Section 4 shows how the data are translated into the objective function and constraints of the

mathematical programming models. The next section interprets the results and discusses

strategic insights for the purchasing policy. The last section concludes.

THE VALUE CHAIN AND ACTIVITIES

We study the activities in the value chain of the firm that relate to the purchasing

policy in the first step of the vendor selection methodology. These can either be activities of

the purchasing department itself or activities further down the value chain that are influenced

by policy decisions made by the purchasing department. Figure 1 shows the activities, where

they are situated in the value chain and how they relate back to the purchasing policy in the

case study firm. It is important to perform this value chain analysis, as these activities and

their costs will later be modelled in the mathematical programming model. The rest of this

section describes the activities in the value chain of the firm.

Insert Figure 1 About Here

The purchasing engineer responsible for a component group negotiates with the

suppliers on amongst others price, discounts, quality, lead time and follows up the relationship

to sustain the supplier in the supply base. When the supplier is new to the firm or when quality

9

problems arise too frequently, the quality team sometimes performs a quality audit on the

supplier’s site.

Ordering can start once the relationship with the supplier is set up and the supplier is

selected for a certain component type. Depending on the supplier/component combination

orders can be placed through electronic data interchange (EDI) via a private network,

automatic call-off (ACO) on a frame agreement or the manuals by sending a fax.

2

The first

time an order is placed for a specific PCB with a supplier, a tooling cost might be charged by

the supplier to cover the supplier’s one/off costs on films, drilling information and electrical

testing. Suppliers with a shorter lead-time are more flexible in that they can accommodate to a

sudden change in demand on a shorter notice period and thus agree a delivery date that is

nearer in the future than other suppliers can.

3

A supplier’s delivery reliability depends on the

history of early and late deliveries around the agreed delivery date. Importing documents have

to be filled out and import duty has to be paid when the component is ordered with a supplier

outside the European Community.

Then the receiving department receives the delivery and inspects it together with the

inspecting department.

4

When no irregularities are discovered during the inspection, the

supplier accounting for the delivery is done and the invoice is paid. For transformers and

PCBs some suppliers offer component specific discounts on prices for larger orders and this

discount may rise with the quantity ordered. Some transformers and PCB suppliers add a lot

charge to the invoice. Payment delays typically range from cash payment to 60 days delay,

with 0 to 3% payment discounts. However, when a defect is discovered in inspection,

components are either sent back to the vendor who will send a credit note or will replace them

at his expense, or they are thrown away at the firm’s own expense. When the supplier replaces

2

The minimum order quantity and the lot size have to be adhered to when ordering. Orders for a component thus

have to exceed the minimum order quantity for that component with that supplier and be a multiple of the lot

size. As a rule, the lot size is always lower than or equal to the minimum order quantity.

3

Asian suppliers generally have a longer lead-time than European and American suppliers.

4

Different sorts of inspections are used, depending on the inspection class in which the supplier/component

combination is allocated and resulting in more or less time consuming inspection activities. For purchases with

certified suppliers, the receiving department may release the components without any quality verification. The

trust in these suppliers’ quality systems makes extra inspection superfluous, as the details on the specifications,

the level of quality, the criteria for acceptation of the delivery, the supplier’s auditable quality plan and the

markings on the packaging are agreed on in writing in the quality agreement. Other components are inspected

visually. A skip lot inspection may be performed for components that are delivered frequently. In this case the

first five deliveries and afterwards every fifth batch are each checked taking a sample, whereas the other four are

only checked visually. In a few exceptional cases, the reception department releases transformers and PCBs

delivered by uncertified suppliers without further inspection because their impact on business processes is

considered small. For the odd resistor delivery only the labels on the packaging are compared with the ones on

the travel documents without opening the packaging. Occasionally, every PCB lot is checked using a sample

from each lot. Some special PCBs are sent for verification to the engineer that ordered the component.