The current issue and full text archive of this journal is available at

http://www.emerald-library.com

IJOPM
20,6 Stocking strategy for service
parts ± a case study
Rene Botter
656 Andersen Consulting, Amsterdam, The Netherlands
Leonard Fortuin
Eindhoven University of Technology, Eindhoven, The Netherlands
Keywords Inventory, Service, Parts, Costs
Abstract Service part inventories cannot be managed by standard inventory control methods,
as conditions for applying the underlying models are not satisfied. Nevertheless, the basic
questions of inventory control have to be answered: Which parts should be stocked? Where should
they be stocked? How many of them should be stocked? Presents a case study in which a pragmatic
but structured approach is followed ± a framework is developed and built into a spreadsheet. The
resulting tool has been tested in a real-life situation, indicating that considerable amounts of
money can be saved.

1. Introduction
For many companies faced with world-wide competition, customer satisfaction
has become crucial. An important means to keep customers satisfied is quick
repair of a product or system that has failed. To this end, enough parts, usually
called ``spare parts'' or ``service parts'', have to be stocked at appropriate points
in the supply chain to guarantee a high service level. As most of these parts are
expensive, this requires large amounts of money to be invested.
For service parts (the name used throughout this paper), three control
situations have to be distinguished (Fortuin and Martin, 1999):
(1) service parts to maintain the company's own (production) facilities and
systems;
(2) service parts to service (professional) systems installed at customer
sites; and
(3) service parts to repair consumer products, at service workshops.
As in any inventory control situation, some basic questions have to be
answered. The following apply here:
. Which service parts have to be stocked?
. Where are the service parts to be stocked?
. How many units have to be kept in stock for each of these service parts
(reorder level and reorder quantity)?
Service part management means finding feasible and efficient answers to these
International Journal of Operations &
basic questions. That is a difficult task as standard methods for inventory
Production Management,
Vol. 20 No. 6, 2000, pp. 656-674.
control fall short ± consumption may be so erratic and low that there is no
# MCB University Press, 0144-3577 demand process that allows forecasting of future demand, service parts are
expensive and customers very demanding. Criticality of service parts is a Stocking
useful concept to bring some order in this complex situation. It expresses the strategy for
importance of a part in case of system failure. A generally valid definition of service parts
``criticality'' is hard to find, as local circumstances play an important role.
Service parts can be divided into two categories:
(1) Repairables. Service parts that are technically and economically
repairable. In case of failure, such a part is swapped with a new one and 657
sent to a repair centre. In literature and in practice they are also called
repairable parts, recoverable parts, recoverables, rotation parts and
rotables.
(2) Consumables. Service parts that are technically and/or economically not
repairable. In case of failure the part is replaced by a new one and
scrapped. Other names are consumable parts, expendable parts,
expendables, disposable parts and disposables.
Following Sherbrooke (1992) we will use the term ``item'' to designate a specific
type of part and ``units'' for the quantity of the item.
This paper is based on a case study (Botter, 1997). It describes a solution for
the second of the three control situations mentioned at the beginning of this
section, namely the control of service parts for the repair of professional
electronic systems at customer sites. The parts involved are repairable. For
obvious reasons, sensitive information about the company, its position within
its industry, and its operations has been left out. The company is called here
EDIAP, which stands for electronic devices for industrial applications. The
outline of the paper is as follows. In Section 2, attention is paid to the related
literature. Following this, there are sections describing the company (Section 3),
the problems (Section 4), and the approach followed (Section 5). The application
of the model and some results are described in Section 6, and conclusions are
drawn in Section 7.

2. Related literature
Stock control has received much attention from the scientific community for
many decades now. The majority of literature is focused on (re)ordering
policies in order to supply production lines with material, and stock control for
distribution networks; see for instance Silver et al. (1998). Then demand and
delivery can be reliably predicted and theoretical models turn out to be useful
in practice. Our paper deals with an entirely different situation, the (re)ordering
policies in service environments. There managers have to face a complex
problem: service parts are expensive, demand is erratic and hence hard to
predict, delivery times are long and stochastic, and customers want their
service parts quickly. Literature on that situation is relatively scarce. Recent
papers deal with the final order problem (Hill et al., 1999), optimal review
policies (Shibuya et al., 1998), stocking policies based on criticality (Dekker et
al., 1998), and new concepts for the control of service parts (Fortuin and Martin,
1999). We are primarily interested in realistic and tested approaches such as
IJOPM Vereecke and Verstraeten (1994), rather than in approaches of a purely
20,6 hypothetical and theoretical nature. Zero Inventory, for instance, is not a
realistic proposition; managers are already happy if they can determine the
Minimum Reasonable Inventory (GruÈnwald and Fortuin, 1992).
From a literature survey (Botter, 1996) we derive the following observations:
(1) Most of the professional literature deals with one of four branches:
658 . electronic industry, including computers;
. automotive industry;
. aviation industry; and
. production companies keeping stock of parts for maintenance.
(2) The electronic industry, the automotive industry, and the airline
operators more and more tend to outsource service parts management.
(3) In these industries there is also a tendency towards co-operation, e.g.
joint exploitation of service parts inventories, for the sake of cost
reduction. Joint acquisition of parts and collective maintenance are also
examples.
(4) Investments in automation of administrative processes and warehouse
operations are high, in order to reduce the time needed for processing,
handling and transport. Stock reduction is the result.
(5) Classification of service parts by labelling them as ``fast mover'' or ``slow
mover'' is no longer sufficient. Other classification criteria emerge, such
as:
. criticality of the function to be performed by a system that has
become defective;
. price;
. delivery time; and
. life-cycle phase of the service part.
(6) Methods to improve productivity of field service are getting ample
attention, especially for the electronic industry.
Practical methods to deal with service parts inventory control, have been
studied at Eindhoven University of Technology for many companies in The
Netherlands. They all signal that existing mathematical models do not suffice.
Instead they design a pragmatic solution especially for the situation at hand.

3. The company
3.1. The logistics organisation
EDIAP is a large multinational company; it develops, manufactures, sells,
supports and services electronic devices to be used in industry. In order to
support the installed base throughout the product lifecycle, EDIAP has a multi-
echelon distribution and repair structure to supply customers with service
parts. It operates in a market characterised by growing importance of after Stocking
sales activities and severe competition. A European supply centre (ESC) is the strategy for
main distribution and repair centre for service parts in Europe, the Middle East service parts
and Africa. Its main responsibility is the procurement, warehousing and
distribution of service parts for and to the stocking points in the countries.
There are 16 countries to be served, with 80 national warehouses in total. The
ESC also serves the national warehouses with logistical advice in order to 659
achieve a better use of the European stock. Each national warehouse employs a
number of service engineers. These engineers visit customers in case of a
system failure. The term ``installed base'' is used to indicate all the systems in
use at customers' sites. Figure 1 gives an impression of EDIAP's supply chain
for service parts.
Besides this distribution function, the ESC has three other important
functions:
(1) Repair of returned defective service parts in the repair service centre
(RSC), so they can be used again for maintenance.
(2) Disassembly and scrap of defective systems and parts that are
technically or economically no longer repairable. Usable parts are
stocked or sold to third parties.
(3) Refurbishment and upgrading of used systems.

3.2. Products and parts

The product assortment of EDIAP is divided into six categories. Each category
consists of several families, each family includes several systems. A system can
be equipped with different options. These options are the smallest saleable

Suppliers

Factory

ESC

National
warehouses

Service
engineers
Figure 1.
An impression of
EDIAP's supply chain
Installed base for service parts
IJOPM items. Systems and built-in options have a modular structure. Modules, here
20,6 called field-replaceable units (FRU), contain several parts. The ESC has to
provide maintenance for the products sold to the market, the installed base.
This ``after-sales service'' is offered on the basis of service contracts. Hence
service engineers must have service parts at their disposal at the right time and
at the right place. Here, a service part can be a system, an option, or a FRU.
660 Within the supply chain of the ESC, two material flows can be observed:
(1) The flow of newly made or repaired service parts towards service
engineers. This flow is driven by replenishment and emergency orders
from the various stocking points. These stocking points supply service
engineers with service parts for repair at customer sites. When the
needed parts are in stock, they are allocated to the service engineer
requesting them and immediately a replenishment order is generated.
Service parts for replenishment are shipped to the lower echelons via a
consolidated transport network.
(2) The flow of defective repairable service parts towards the repair centres.
After replacing a defective part by a new one at the customer site, the
service engineer takes the part back home, where consumables are
destroyed. Depending on the available repair facilities, repairables are
repaired at country level or shipped to the ESC.
In each country, management is responsible for service performance and related
costs. Stocks for service parts are kept at a recommended (order-up-to) level,
called target stock level (TSL). When customer demand appears, the stock level
will drop below the TSL and automatically an order is generated and placed for
replenishment at the ESC. Prices of service parts are high, and customers demand
usually one piece at a time of a certain item. Hence the ordering procedure follows
the ``one-for-one replenishment'' rule, or (S-1,S) rule. This is supported by
fundamental research (Sherbrooke, 1992). At European level, at the ESC,
inventory control follows the (RsQ) policy ± stock levels are reviewed weekly (R),
and a quantity of service parts (Q) is ordered whenever the stock level has fallen
below the reorder point (s). In total, there are about 50,000 active part numbers.

4. The problem
The company operates in a market with severe global competition. Strategic
market research has revealed that customers want a wider range of services at
lower cost. This means, among other factors, that the level of service agreed
upon acquisition of a product depends on the application of the product, rather
than on the product itself. Customers choose the service level per application,
i.e. they desire a high level for critical applications, whereas they try to reduce
costs in the less critical areas. A plan to re-engineer the global supply chain was
formulated, with its two key goals being:
(1) better financial performance; and
(2) higher customer satisfaction.
To achieve these goals, the company defined four key issues to focus on: Stocking
(1) improvement of supplier performance; strategy for
(2) speeding up delivery by the logistics organisation; service parts
(3) improvement of the field stocking strategy; and
(4) better contract support.
661
This paper deals with improvement of the field stocking strategy, but it will be
clear that in practice none of these issues can be dealt with in isolation.
A Pareto analysis of service after sales revealed that 90 per cent of the
service level was achieved with only 10 per cent of the service parts assortment.
Therefore management wanted to concentrate on those 10 per cent of items. An
important parameter was the stock on hand (SOH), defined as available
inventory divided by consumption per week. The new stocking strategy should
lead to a low SOH for fast movers and higher SOHs for slow movers. The very
slow movers are a class apart; they will be ordered when needed, they will not
not be kept in stock (SOH = 0).
Now the problem to be solved can simply be stated as follows: create a tool
for stock control of service parts that guarantees a good balance between
customer satisfaction and costs. With this tool two questions have to be
answered:
(1) Which items have to be stocked?
(2) How many units have to be stocked for each item, and where?

5. The approach
5.1. Selection of items
For the sake of clarity we repeat the convention adopted here ± the term ``õÈtem''
designates a specific type of service part and ``units'' indicates the quantity of
the item (Sherbrooke, 1992).
There are two good reasons to reject the idea of developing one optimising
mathematical method for all service parts:
(1) Conventional models for inventory control are hardly suitable for
service parts, especially parts with a low frequency of demand.
(2) EDIAP wanted a tool for the entire supply chain of service parts, hence a
tool to cover the whole service parts assortment. But not all parts were
the same, and so a classification was required.
Consequently, instead of one optimising method, a practical method was
developed. It yielded a performance that satisfied both customers and
management. In this method the criterion of criticality was applied to
distinguish important parts from others. The point of departure thereby was
the VED approach (VED = vital, essential, desirable), but only two of the three
classes were used (see Sections 5.2 and 5.3).
IJOPM As there are about 50,000 active items, from the beginning it was clear that
20,6 the method had to be based on modern IT techniques, easy to implement and to
maintain. Given the limited amount of time for this job[1], some more
restrictions had to be made:
. No changes in the existing distribution network were allowed. This
meant that no attempts were made to define new locations for stocking
662 parts.
. Only two echelons of the distribution network were to be considered: the
ESC and the national warehouses. Stocks in branches, in technician's
cars, and at customer sites were outside the scope of this project.
. The tool to be developed had to be simple to implement and use. As will
be seen in Section 6, this resulted in a spreadsheet application.

5.2. From the analytic hierarchy process to a simplified VED approach

According to the regular VED method, there are three types of service parts:
(1) Vital parts. Items that cause high losses due to non-availability of
equipment, in case they are needed while not in stock.
(2) Essential parts. Items that cause moderate losses due to non-availability
of equipment, in case they are needed while not in stock.
(3) Desirable parts. Items that cause minor disruptions, in case they are
needed while not in stock.
Criticality is here the ruling criterion. From the point of view of its functional
necessity in production or service operations, criticality of a service part has
many facets. EDIAP distinguishes the factors listed in Table I. However,
evaluating the criticality of items is a difficult task that often can only be

Number Factor Comment

1 Response time Maximum time between a call for help and restoration
of the system's functionality, as agreed by contract
2 Functionality Effect of the failure of an item on the system's
availability: an item is functional if the system cannot
function without it, or merely cosmetic if the system can
continue to run without it, possibly with some minor
restrictions
3 Consumption Total demand for an item, per unit of time, expressed in
number of units or in money
4 Stage of the life cycle Newly developed, established and to be continued, or
soon to be phased out
5 Price An item can be (relatively) cheap or expensive
6 Purchase lead time Time between placing an order at the supplier of an
Table I. item and the moment it is available for use
The many facets of 7 Repairability The possibility to restore an item's functionality after
criticality failure
accomplished by using subjective judgements. A systematic procedure that Stocking
could be helpful here is the analytic hierarchy process (AHP). This multi- strategy for
criteria decision-making tool was developed by Saaty (1980) to find the relative service parts
priorities or weights to be assigned to different criteria and alternatives which
characterise a decision. The decision model is based on the idea of structuring
the problem into a hierarchy with three levels ± the overall objective is at the
apex, criteria characterising the objective are placed in the middle, whereas the 663
decision alternatives can be found at the bottom. Criteria and alternatives have
to be determined by management. In case of EDIAP the overall objective is
``Evaluation of the criticality of spare parts''. Criteria at the middle level and
their alternatives are those collected in Table II.
After a thorough explanation of this method, management of EDIAP
rejected it as a way to classify service parts. The main reason was that one of
the most important factors, response time agreed upon in service contracts
(Factor 1 in Table I), can vary for the same item from one customer to another.
Consequently, such a service part cannot uniquely be assigned to a class.
Moreover, management found the proposed method ``too theoretical''. It was
then decided to modify the VED method in such a way as to accommodate local
knowledge.

5.3. Which items have to be stocked?

The next factor influencing criticality is functionality (Factor 2 in Table I).
There are two possibilities: a service part is either functional or cosmetic.
According to the VED analysis, the notions vital, essential and desirable must
also be considered. However, a functional part can be vital as well as essential,
as can be seen from the definitions of these terms in Section 5.2. But the

Criteria Alternatives

Response time agreed upon in service contracts 2-4 hours

Next business day
Later than next business day
Life cycle Introduction
Maturity
Decline
Demand (AMC = average monthly consumption) AMC 5
5 < AMC 100
AMC > 100
Purchase lead time (LT in weeks) LT 1
1 < LT 3
LT > 3
Table II.
Price (P in $) P 100 Criteria and
100 < P 1,000 alternatives for service
P > 1,000 parts
IJOPM distinction between huge loss and moderate loss is hard to make. For practical
20,6 reasons, therefore, all functional parts are assigned to Class E (essential).
Because failure of a cosmetic part does not cause failure of the whole system,
cosmetic parts can be assigned to a less important class. In terms of the VED
analysis, this class is called Class D (desirable). Therefore, along the dimension
functionality there now are two possibilities:
664 (1) cosmetic (Class D); and
(2) functional (Class E).
Consumption of service parts is another important factor (Factor 3 in Table I).
Two versions can be met in practice: consumption expressed in money and
consumption in units. EDIAP holds the latter to be the most important, because
a cheap service part might cause a serious breakdown for a customer. Via the
well-known Pareto analysis of cumulative demand in units, three classes of
service parts are distinguished: fast moving parts (Class X), slow moving parts
(Class Z), and parts of Class Y in between. Combining these two dimensions, we
obtain Framework 1 for answering the first question: Which items have to be
stocked? It consists of six segments, as shown in Figure 2. Each segment
represents a particular group of service parts. The segment EX, for instance,
contains all fast moving essential parts; DZ contains all slow moving cosmetic
parts. The size of a group depends on the definition of fast moving, etc. It is
common practice to take 90 per cent and 99 per cent as boundaries.
Each segment requires a different stocking strategy. These strategies are
mainly based on decisions made by EDIAP management. As an example, it is
possible that management decides not to stock cosmetic parts. From a systems
point of view this is acceptable, because failures of cosmetic parts will not
immediately influence the operation of the system. Then only the parts in
segments EX, EY and EZ are left as candidates to be stocked. If next the
decision is taken to cover 90 per cent of cumulative consumption, only the parts
in segment EX have to be stocked.
The remaining factors in Table I played no part in the method developed.

5.4. How many units have to be stocked, and where?

For the items selected for stocking, we will now address the second question:
How much stock has to be maintained of those service parts, and where do they

Functionality

D
Desirable
(cosmetic) DX DY DZ
Figure 2. E
Essential
Framework 1 for (functional) EX EY EZ
answering the question:
X Y Z
Which items have to be fast moving slow moving
stocked?
Percentage consumption in units
have to be stocked? The answer strongly depends on the desired level of service Stocking
(LOS). Conventional inventory theory is not applicable here, as demand strategy for
frequency is far too low. Sophisticated mathematical models, with complex service parts
distribution functions to approximate low demand levels, have to be rejected
also. The reason for this rejection is simple ± management finds them too
difficult to implement and too hard to maintain.
In Section 5.3, it was stated that consumption in units is regarded as more 665
important than consumption expressed in money. However, when deciding
about how many and where to stock parts, price is important too. After all, it is
undesirable to stock very expensive service parts in every local warehouse. So
if possible such parts will be stored only centrally in the countries or even only
at the European supply centre, and the number of units to be stocked will be
kept to a minimum. On the other hand, logistics managers will have few
objections when cheap items such as nuts and bolts are stored locally, even in
large quantities.
Even after applying Framework 1, a majority of the parts to be stocked will
be characterised by low consumption. Therefore, usage in units will also be
relevant when deciding about where to stock parts and how many. It is likely
that fast moving service parts are stocked in local warehouses, as large(r)
quantities can be shipped by cheap(er) means of transport. On the other hand,
slow movers will be stocked only centrally, to decrease inventory costs. If
demand occurs, such parts will be shipped to the customer, if necessary by
courier or taxi.
Although both factors, i.e. consumption in units and price, are very
important, again the most important factor is the time that is allowed to
provide customers with service parts. After all, one of the functions of stock is
to decouple successive processes in such a way that these processes can be
executed independently. In our case, the successive processes are
replenishment and repair. If the repair process allows service parts to arrive
within a day after a call, the stocking strategy will differ from the strategy
when parts have to be available within a few hours. Obviously, this time
depends on the service contracts in which service response times are defined.
The three dimensions described above can be depicted in another framework
(see Figure 3). Along each axis two general classes are defined. This choice is
arbitrary: it is up to management to decide upon the number of classes for each
of the dimensions. We suggest taking no more than two. Then the number of
different segments becomes eight, which is manageable. Each segment
represents a particular group of service parts, each with its own approach. The
segments are shown in Figure 3:
(1) Low price, short response time, high usage. These cheap, fast moving
items have to be stocked in large quantities in local warehouses, i.e. close
to the market.
(2) Low price, short response time, low usage. These cheap, slow moving
items also have to be stocked close to the market, but in lower quantities.
IJOPM (3) Low price, long response time, high usage. For these items inventory
20,6 costs and transport costs should be investigated, in order to determine
whether or not local stocking is better than central stocking. Local
stocking of fast moving parts could decrease transport costs, as larger
quantities can be shipped by cheaper means of transport.
666 (4) Low price, long response time, low usage. These service parts are only to
be stocked centrally, at the ESC.
(5) High price, short response time, high usage. These parts require firm
management, as stocking is expensive. Owing to the short response
time, parts primarily have to be stocked in local warehouses. The
quantities should be as low as possible, and depend on the desired
customer service level.
(6) High price, short response time, low usage. Again firm management is
needed. It may be worthwhile to consider a fast means of transport, even
if it is expensive (e.g. taxi). In this case stocking centrally in the countries
becomes possible, thus reducing inventory costs.
(7) High price, long response time, high usage. For these parts a trade-off has
to be made to choose between central stocking in the countries and at the
ESC.
(8) High price, long response time, low usage. Owing to the long response
time these items can be stocked centrally at the ESC and shipped by
regular means of transport when needed.
In this way an integrated stocking strategy is obtained for all segments. A
complication, however, may arise if a service part belongs to two different
systems, each with its own response time. In fact, this happens often at EDIAP.
Nevertheless, in this paper this possibility will not be considered.
Consumption (pieces)

1
high

Figure 3. low high

Example of framework
depiction for low
ice

consumption, response
Pr

short long
time, and price variables
Response time
5.5. Quantitative analysis Stocking
In the previous section, Framework 2 dealt with the question of how many strategy for
units should be stocked and where in a qualitative sense. When stock levels service parts
have to be quantified, the desired LOS plays an important role. For this study,
LOS is defined as the percentage of demand delivered directly from the shelf, as
experienced by the customers. Because of the last words, ``as experienced by
the customers'', the LOS may be regarded as a measure for the final customer
667
service.
Suppose that for a certain segment of Framework 2, N different service parts
(items) have to be stocked. Consumption per item and total consumption are
given, both in units. For each item p (p = 1, 2, . . ., N) two quantities are
determined:
(1) The fractional consumption rate, Cp (0 Cp 1), i.e. consumption for
item p divided by total consumption.
(2) The fill-rate, Fp (0 Fp 1), i.e. the fraction of demand that is met from
stock on the shelf.
Then the LOS for the segment can be computed from:
X
LOS ˆ … Cp Fp† 100%: …1†
p

Formula (1) indicates that the pre-specified overall LOS as required by EDIAP,
depends on the fill-rate of the individual service parts multiplied by their share
in the total consumption in pieces. We call this the ``weighted fill-rate''. The next
step is to determine for each service part the stock level needed to realise a
sufficient fill-rate for that item, in such a way that the sum of all weighted fill-
rates is equal to or larger than the required LOS.
Sherbrooke (1992) defines the stock level s as:
s ˆ OH ‡ DI ÿ BO; …2†

where:
OH = number of units of stock on hand;
DI = number of units of stock due in from repair or re-supply; and
BO = number of units being back ordered.
The stock level s is a constant, the other variables are non-negative random
numbers. Any change in one of them causes a change in another. The
simplicity of Formula (2) is caused by the fact that the batch sizes for ordering,
repair, re-supply and demand are all equal to one.
If DI # s-1 demand will be followed by a fill, because OH > 0. For DI s, there
is no stock on hand. So the expected fill-rate Fp(s) follows from Formula (3):
IJOPM Fp …s† ˆ ProbfDI ˆ 0g ‡ ProbfDI ˆ 1g ‡ . . . ProbfDI ˆ s ÿ 1g
…3†
20,6 ˆ ProbfDI s ÿ 1g;

where Prob{DI = i} is the steady-state probability of having i units of stock due

in. According to Sherbrook (1992) this is a Poisson probability with mean mT,
668 where mT follows from Palm's theorem:
If demand for an item is a Poisson process with annual mean m and if the time for repair or
resupply for each failed unit is independently and identically distributed according to any
distribution with mean of T years, then the steady-state probability distribution for the
number of units in repair or resupply is a Poisson distribution with mean mT.

So, for DI we have (i = 0, 1, . . .):

ProbfDI ˆ ig ˆ …mT†i exp…ÿmT†=i! …4†

Substitution of Formula (4) into Formula (3) enables the expected fill-rate of a
service part to be calculated, if stock level, average monthly demand, and
average re-supply time (in months) are given.
If the number of service parts is very high and consumption per item is low,
it makes sense to combine the items into a group and base the analysis on the
demand of the group, rather than on each of the items. Then the formula for the
LOS becomes:
X
LOS ˆ CGg FGg 100%; …5†

where:
LOS = overall level of service (0 LOS 100);
g = group number;
CGg = consumption rate (0 CGg 1), i.e. consumption for group g divided
by total consumption; and
FGg = fill-rate, (0 FGg 1), i.e. the fraction of demand that is met from
stock on the shelf.

6. Application of the model

The foregoing model has been applied to the assortment of service parts for a
certain family of products. European consumption was given over a period of
six months. Application of Framework 1 divided them into about 30,000
functional items (Class E) and 20,000 cosmetic items (Class D). Then
management decided that all functional items have to be stocked, a simple and
straightforward answer to the first basic question, ``Which items have to be
stocked?''.
``How many units have to be stocked, and where?'' was the second question
posed. Treating each of the 30,000 functional service parts in the way outlined
in Section 5.4 would be very time-consuming. Moreover, it would produce a tool
that is hard to implement and support. Hence, it was decided to apply the Stocking
framework only to the most frequently demanded service parts, and treat the strategy for
service parts less frequently demanded as groups. This idea led to the service parts
following procedure:
(1) Division of the assortment into 11 groups, on the basis of demand
frequency (see Table III).
669
(2) Analysis of each service part in Group 1 and Group 2 in the way outlined
in Section 5.5, via Formula (1). These parts are the so-called fast movers.
(3) Analysis of the service parts in Groups 3-11, per group, on the basis of
Equation (5). These are the so-called slow movers.

6.1. Stocking individual items in every national warehouse

The calculation of the LOS for each of the fast-moving items, i.e. those in
Groups 1 and 2, is illustrated by looking at the service part with the highest
consumption. This one is labelled p = 1. Its consumption in six months equals
5,410 units. As there are 80 national warehouses (NW) in Europe, this means an
average of about 12 units per month per NW. The fractional consumption rate,
C1, for this item is 5,410/524,668 = 0.0103. The replenishment time from the
ESC to each of the NWs is assumed to be distributed with a mean value of three
days, or 0.1 month. Hence, in Equation (4), substituting mT=12 0.1 = 1.2, we
get for the number of units due in:
ProbfDI ˆ ig ˆ …1:2†i exp…ÿ1:2†=i! …6†

Next, the fill-rate for a given value of the stock level s is considered. With
Equation (6) we can calculate Fp(s) as given in Formula (3). Some results
(s = 0, 1, . . . ,7) are listed in Table IV. This table shows that in order to acquire
an expected fill-rate of at least 0.99, a stock level of five units in each of the 80
NWs is needed. As the price of this item is $35, the total investment in stock for

Demand frequency Total consumption

per part over six No. of items in a over six months Average price
Group months (units) group (units) ($)

1 > 1,000 69 125,885 140

2 501-1,000 104 71,211 231
3 401-500 52 23,497 240
4 301-400 85 28,961 231
5 201-300 194 46,711 271
6 101-200 470 65,594 322
7 51-100 777 53,780 310
8 11-50 3,613 78,088 312
9 4-10 5,407 27,199 407
10 1-3 2,719 3,742 407 Table III.
11 0 15,562 0 493 Grouping of service
Total 29,052 524,668 parts for Europe
IJOPM Stock F1(s) numerical Contribution to
20,6 level (s) F1(s) in formula value LOS

0 0 0
1 Prob{DI = 0} 0.301 0.0031
2 Prob{DI = 0} + ... + Prob{DI = 1} 0.662 0.0068
3 Prob{DI = 0} + ... + Prob{DI = 2} 0.879 0.0091
670 4 Prob{DI = 0} + ... + Prob{DI = 3} 0.966 0.0099
5 Prob{DI = 0} + ... + Prob{DI = 4} 0.992 0.0102
Table IV. 6 Prob{DI = 0} + ... + Prob{DI = 5} 0.998 0.0103
Expected fill rate and 7 Prob{DI = 0} + ... + Prob{DI = 6} 0.999 0.0103
LOS for a single part
(part number 1) Notes: LOS = level of service

this item will be 5 80 $35 = $14,000. Then the service part is expected
to contribute approximately 0.992 0.0103 100% = 1.02% to the overall
LOS. The foregoing calculation has to be done for all the 173 parts in Groups
1 and 2.

6.2. Stocking groups of parts in every national warehouse

This type of calculation will be illustrated by means of Group 3 in Table III.
This group contains 52 service parts with a demand between 401 and 500 units
during the period of six months. The average price per item is $240. Total
consumption equals 23,497 units. Hence, the average demand per month and
per NW is m = 0.94 units (23,497/(52 6 80)). Once again we assume that
demand is a Poisson process and the replenishment time from the ESC to the
NWs is on average three days, i.e. T = 0.1 month. Consequently, the number of
units due in has a steady-state Poisson probability distribution with mean
value mT = 0.094.
Column 2 of Table V shows the expected fill-rate for s = 0, 1, 2 and 3 units. If
management decides that an expected fill-rate of 0.910 is sufficient for Group 3,
each NW should keep one unit in stock. Roughly, this will require an
investment of 1 52 80 $240 = $998,400. As CG3 = 23,497/524,668 = 0.045,
the contribution by this group to the overall LOS amounts 0.910 0.045
100% = 4.1%.

6.3. Stocking groups of items in a central national warehouse

So far the focus has been on stocking service parts in every NW. However,
often the average monthly consumption (AMC) per NW per item is far less than

Stock level (s) FG3(s) FG7(s) FG9(s)

0 0.000 0.000 0.000

Table V. 1 0.910 0.931 0.657
Expected fill rate for 2 0.996 0.998 0.933
Groups 3, 7 and 9 3 0.998 0.991
1. In that case, stocking items in just one central warehouse in a country rather Stocking
than in each NW, is preferable. For example, in Group 7 the AMC per country strategy for
per service part is 0.72. The average replenishment time from the ESC to a service parts
central NW is three days, or 0.1 month. So the number of units due in has a
steady-state Poisson probability distribution with mean value 0.072. Column 3
of Table V shows the fill-rate for stock levels s = 0, 1 and 2 units. A (central)
stock level of 1 unit in every country for every item in this group will result in 671
an expected fill-rate of 0.931. This requires an investment of 1 777 16
$310 = $3,900,000. As the CG7 = 53,780/524,668 = 0.103, the contribution of this
group to the overall LOS is 0.931 0.103 100% = 9.54%. The feasibility of
this mode of stocking parts depends on the time that is allowed for service
parts to arrive at the customer. Usually, this allowed time is stated in the
contract with the customer.

6.4. Stocking groups of items only in the ESC

For many service parts demand is so low that even central stocking in the
countries is not acceptable. For these items, stocking at the ESC or not stocking
at all, are the only possibilities. Items of Group 9, for example, have a European
AMC of 27,199/(5407 6) = 0.84. It is important to note that in this case the
average replenishment time from vendors to ESC is crucial. It is assumed to be
equal to 0.5 month. Consequently, the number of units due in has a Poisson
distribution with mean value 0.5 0.84 = 0.42. Column 4 of Table V shows the
fill-rate at the ESC for stock levels from 0 to 3 units. A stock level of two units
calls for an investment of around 2 5407 $407 = $4,400,000; it results in an
expected fill-rate of 0.933, a CG9 = 0.052 (27,199/524,668) and an overall
contribution to the LOS equal to 0.933 0.052 100% = 4.9%. The impact of
the accepted service response time is even stronger than in the previous case.
Despite a very low demand, some items do have to be stocked in the countries,
or even in local warehouses. Information on response time, however, was not
available during this investigation.

6.5. A spreadsheet for the whole assortment

The foregoing sections showed how the model has to be applied in various
cases, corresponding to the intensity of the demand process. In this way it
becomes clear where the service parts have to be stocked and in what quantity.
The cases treated correspond to the boxes of the framework drawn in Figure 3.
To facilitate the calculations needed for the whole assortment, i.e. the 11 groups
with in total 29,052 items, a spreadsheet has been developed. It enables EDIAP
to calculate the investments in stock and consequently the interest costs of
stock holding, necessary to acquire a pre-specified Level Of Service. Transport,
warehousing and handling costs are not taken into account. As a matter of fact
these costs are not negligible, but it can be shown (Van Goor et al., 1994) that for
service part distribution by EDIAP interest costs are most important.
With the spreadsheet three scenarios have been investigated:
IJOPM (1) Scenario 80-0-20. The overall LOS has to be at least 95 per cent,
20,6 approximately 80 per cent of total customer demand has to be met
instantaneously by the national warehouses and 20 per cent by the ESC.
(2) Scenario 50-0-50. The overall LOS has to be at least 95 per cent,
approximately 50 per cent of total customer demand has to be met
instantaneously by the national warehouses and 50 per cent by the ESC.
672
(3) Scenario 100-0-0. The overall LOS has to be at least 95 per cent,
approximately 100 per cent of total customer demand has to be met
instantaneously by the national warehouses.
Table VI shows the most important results. We note some interesting effects:
. As the actual total investment in stocks was $107.3 million, the model
shows that a reduction of about $33 million is possible in case of
Scenario 80-0-20.
. For Scenario 50-0-50, total investment in stock is much lower than for
Scenario 80-0-20. This can be understood by realising that central
stocking is cheaper than stocking in national warehouses. Compared
with the actual investments, a decrease of $58 million seems possible.
The ``price'' to be paid for this saving is the longer average lead-time of
slow movers.
. Delivering all parts from NW (Scenario 100-0-0) requires a large amount
of stock. The average lead-time for fast movers increases somewhat, but
for the slow movers the average lead-time reduces to half the value that
holds for Scenario 80-0-20.

7. Conclusions
A framework has been presented by means of which the three basic questions
for service parts inventory control for a two-echelon network of warehouses

Scenario Scenario Scenario

Description 80-0-20 50-0-50 100-0-0

Investment in fast movers 9,148,200 9,148,200 7,226,800

Investment in slow movers 64,842,000 40,129,100 420,732,300
Total investment 73,990,200 49,277,300 427,959,100
Average lead time of fast movers to customers
(days) 0.54 0.54 0.66
Average lead time of slow movers to customers
(days) 1.22 1.90 0.56
LOS for parts delivered by the national
warehouses (per cent) 96.7 98.5 95.1
LOS for parts delivered by the ESC (per cent) 90.2 90.5 ±
Table VI. Overall LOS (per cent) 95.4 95.1 95.1
Overview of results for
three scenarios Note: LOS = level of service
with some 50,000 items can be answered. It is based on a division of the items Stocking
into groups of decreasing importance, followed by the calculation of logistic strategy for
parameters that hold for an entire group. Justification of this approach can be service parts
found in management's desire to keep things simple. The framework has been
implemented into a spreadsheet programme. Three scenarios have been
analysed, each differing in the degree in which service parts are to be delivered
directly from local warehouses ± 50, 80 and 100 per cent. In all scenarios, the 673
required customer service level is 95 per cent.
Some interesting conclusions can be drawn from this study:
(1) The best solution to a logistics problem, here the analytic hierarchy
process, is not always acceptable to management.
(2) Service parts managers consider consumption expressed in units as
more important than consumption expressed in money.
(3) The answer to the first basic question: Which items have to be stocked?
depends on the criticality of the item, i.e. the consequences to the
customer if an item is needed and not available. This leads to the
distinction between Vital, Essential and Desirable service parts.
Practitioners, however, easily reduce this to only two classes: Vital plus
Essential, and Desirable.
(4) Service response time and functionality primarily determine criticality
in the company being studied.
(5) Service response time turned out not to be a suitable criterion for
classifying items as vital or desirable, because an item can belong to
different systems, each with its own contracted response time.
(6) If the assortment of service parts to be managed is large, it is better to
limit the number of classes to two, namely functional items and cosmetic
items.
(7) The decision not to stock cosmetic items can reduce investments in
stock considerably.
(8) In order to answer the second basic question: How many units have to be
stocked and where? we have used two factors: usage in units, and price.
Service response time is important too. But for the same reasons as
mentioned in 5, this factor cannot be included.
(9) Service response time has a huge impact on required investments in
stock.
(10) Analysis of three scenarios reveals that the developed tool is capable of
reaching a desired service level at much lower inventory costs.
Note
1. The work reported here is based on a project to obtain an MSc degree in Industrial
Engineering. At the Eindhoven University, such a project takes eight to nine months.
IJOPM References
20,6 Botter, R. (1996), ``Service parts: run or brake down!'', Report of a special assignment, Eindhoven
University of Technology (in Dutch).
Botter, R. (1997), ``An integrated European stocking strategy for spare parts'', MSc thesis,
Eindhoven University of Technology.
Dekker, R., Kleijn, M.J. and De Rooij, P.J. (1998), ``A spare parts stocking policy based on
674 equipment criticality'', International Journal of Production Economics, Vols 56/57 Nos 1-3,
pp. 69-77.
Fortuin, L. and Martin, H. (1999), ``Control of service parts'', International Journal of Operations &
Production Management.
GruÈnwald, H.J. and Fortuin, L. (1992), ``Many steps towards zero inventory'', European Journal of
Operational Research, Vol. 59, pp. 359-69.
Hill, R.M., Omar, M. and Smith, D.K. (1999), ``Stock replenishment policies for a stochastic
exponentially-declining demand process'', European Journal of Operational Research,
Vol. 116, pp. 374-88.
Saaty, T.L. (1980), The Analytic Hierarchy Process, McGraw-Hill, New York, NY.
Sherbrooke, C.C. (1992), Optimal Inventory Modeling of Systems, Wiley, New York, NY.
Shibuya, T., Dohi, T. and Osaki, S. (1998), ``Optimal continuous review policies for spare part
provisioning with random lead times'', International Journal of Production Economics,
Vol. 55, No. 3, pp. 257-71.
Silver, E.A., Pyke, D.F. and Peterson, R. (1998), Inventory Management and Production Planning
and Scheduling, 3rd ed., Wiley, Chichester.
Van Goor, A.R., Ploos van Amstel, M.J. and Ploos van Amstel, W. (1994), Physical Distribution:
Thinking in Added Value, Stenfert Kroese, Leiden (in Dutch).
Vereecke, A. and Verstraeten, P. (1994), ``An inventory management model for an inventory
consisting of lumpy items, slow movers and fast movers'', International Journal of
Production Economics, Vol. 35 Nos 1-3, pp. 379-89.