Вы находитесь на странице: 1из 13

Integrating Closed-Loop Supply Chains

and Spare-Parts Management at IBM


Moritz Fleischmann Jo A. E. E. van Nunen
Rotterdam School of Management/Faculteit Bedrijfskunde, Erasmus University Rotterdam, PO Box 1738,
3000 DR Rotterdam, The Netherlands
Ben Grve
Global Service Logistics, ITS, IBM Nederland N.V., Johan Huizingalaan 761, 1066 VH Amsterdam, The Netherlands
meischmann@fbk.eur.nl jnunen@fbk.eur.nl graeve@nl.ibm.com
This paper was refereed.
IBM is among the pioneers recognizing the benets of closed-loop supply chains that inte-
grate product returns into business operations. We worked on a project exploiting product
returns as a source of spare parts. Key decisions concern what recovery opportunities to
use, the channel design, and coordinating alternative supply sources. Our analytic inventory-
control model and a simulation model showed that procurement-cost savings largely out-
weigh reverse logistics costs and that information management is essential. These ndings
provide a basis for signicantly expanding the usage of the novel parts supply source,
thereby cutting procurement costs.
(Industries: computer, electronic. Inventory, production: policies. )
T
he traditional view of a supply chain, reected
in many textbooks and company Web sites, is a
linear structure, conveying goods from suppliers to
manufacturers, wholesalers, retailers, and nally to
the consumer. Yet, many examples prove this picture
to be simplistic. Todays companies cooperate as com-
plex, general networks rather than one-dimensional
chains. These networks include several types of goods
owing what would be upstream in the traditional
supply chain, such as returns of overstocks, ser-
vice parts, and reusable packaging. Managing these
product ows is called reverse logistics.
A class of reverse goods ows that is gaining par-
ticular importance is returns of used products at the
end of their normal life cycles. Companies are dis-
covering that used products are a valuable resource.
A range of recovery options, including refurbishing,
remanufacturing, and recycling, may allow compa-
nies to recapture part of the original value added or
the value of materials, thereby extending their prots
(Thierry et al. 1995). Additional benets concern
improved corporate citizenship and compliance with
environmental regulation. These factors combined
offer original equipment manufacturers (OEM), spe-
cialized remanufacturers and recyclers, and logistics
service providers many business opportunities.
At the same time, academic interest in reverse logis-
tics is growing, as documented in an expanding body
of literature (Guide and Van Wassenhove 2003). The
European Union has recognized the impact of reverse
logistics by sponsoring a major ve-year research
collaboration (REVLOG 2002) of which our study is
a part.
Initially, researchers focused on understanding the
particular logistics requirements of upstream ows
of goods. In the meantime, the segregation of ows
of goods into forward and reverse ows has made
room for a more comprehensive approach that con-
siders an ensemble of interrelated inbound and out-
bound ows, called closed-loop supply chains. IBMs
situation underlines the necessity of such a holistic
perspective. For example, it integrates returned used
Interfaces 2003 INFORMS
Vol. 33, No. 6, NovemberDecember 2003, pp. 4456
0092-2102/03/3306/0044
1526-551X electronic ISSN
FLEISCHMANN, VAN NUNEN, AND GRVE
Spare-Parts Management at IBM
equipment, such as PCs, larger computers, network
servers, and printers as a source for spare parts. While
rotable spares systems have long been common prac-
tice, they typically concern ows on a part level only.
Exploiting returned used machines as a source of
spares is fairly novel. While this source may produce
important savings, it also entails major challenges. In
particular, machine returns are uncertain and there-
fore difcult to coordinate with other sources of parts.
The goal of our project was to help IBM master these
challenges and establish dismantling of used equip-
ment as a regular source of spare parts.
Reverse Logistics at IBM
The electronics industry is among the most impor-
tant to develop closed-loop supply chains for value-
added recovery. Its ever-expanding market volumes
and short product life cycles result in the disposal of
huge amounts of used products. Electronic waste has
also been a prime target of environmental regulation
(ICT Milieu 2003, European Union 2000). At the same
time, modular product designs and the limited extent
of mechanical wear and tear make electronic products
and components candidates for reuse.
IBM has been among the pioneers recognizing and
exploiting the opportunities to recapture valuable
products and components from streams of postcus-
tomer equipment. Today, IBMs business activities
include various reverse product ows (IBM 2000):
Considering the value-recovery potential, the most
attractive class is end-of-lease equipment. Leases cur-
rently account for some 35 percent of IBMs hardware
output. In addition, IBM has established take-back
programs in several countries in North America,
Europe, and Asia that allow business customers to
return used products for free or after paying a small
fee. The most traditional class of reverse ows is
rotable spare parts. Keeping spares in a closed loop
by repairing defective parts that have been replaced
in the eld has been a key element of IBMs service
business. In our study, we examined the interface
between this spare-parts system and the return stream
of used machines.
Given its experience, IBM has also been impor-
tant in shaping industrywide take-back systems of
consumer electronics in response to environmental
regulations. Because individual manufacturers estab-
lishing collection systems is inefcient for low-value
consumer returns, cooperative industrywide solu-
tions prevail. In The Netherlands, for example, IBM
supports a system organized by the Dutch associa-
tion of information and communication technology
producers to comply with product take-back legisla-
tion (ICT Milieu 2003). In the United States, IBM has
established a different system for consumer returns.
IBM customers can purchase a recycling service with
any new PC, which allows them later to return the
equipment by UPS to a recycling center, which either
prepares it for donation to charities or breaks it down
into recyclable materials (IBM 2000).
Recognizing the growing importance of reverse
logistics ows, IBM created a business unit in 1998,
Global Asset Recovery Services (GARS), to manage
product returns worldwide. Its main goal is to
manage the disposition of returned items to maximize
the value recovered. GARS operates some 25 facili-
ties all over the globe that collect, inspect, and assign
returns to appropriate recovery options (Figure 1).
It assesses which equipment may be remarketable,
either as is or after refurbishment. IBM operates
nine refurbishment centers worldwide, each dedi-
cated to a specic product range. Internet auctions,
both on IBMs own Web site and on public sites, have
become an important channel for selling remanufac-
tured equipment. Yet, ever shorter product life cycles
are limiting the reusability of equipment as a whole.
Therefore, IBM is considering reuse on a component
level as an alternative that may help maximize rev-
enue from returns. In particular, by dismantling used
equipment, it can obtain spare parts for the service
network (our subject in this paper). In addition, it can
sell generic components on the open market. GARS
breaks the remainder of the return stream into recy-
clable material fractions, which it sells to external
recyclers. In this way, IBM limits landlling to a few
percent of the overall input volume.
The Dismantling Channel
By dismantling used equipment to extract valuable
components, such as hard-disc assemblies, cards, and
Interfaces
Vol. 33, No. 6, NovemberDecember 2003 45
FLEISCHMANN, VAN NUNEN, AND GRVE
Spare-Parts Management at IBM
Dis-
mantling
Reusable
materials
market
Disposal
of non-
reusable
remainder
Broker
market
Machines Parts Materials
Customers
Return
center
Refurbish-
ment
Municipa-
lity
Service
network
Recycling-
subcon-
tractor
Charities
Figure 1: IBMs disposition of returned equipment includes reuse options on a machine, parts, and material
level. In this way, it recaptures value and, at the same time, limits landll volumes. The gure shows the product
ows between the different return dispositions.
boards as spare parts, IBM exploits the fact that
spare parts generally have much longer economic and
technical life cycles than the corresponding original
products. Service contracts, which generate a need
for parts, extend years beyond the period during
which a product type is manufactured or may be
remarketable as a whole. Moreover, individual com-
ponents of a returned machine may still function well
even though the machines performance as a whole is
unacceptable.
IBMs dismantling operation came into being as an
incidental, opportunity-driven activity. When prepar-
ing obsolete end-of-lease equipment for scrapping,
the engineers preserved a few high-value compo-
nents, such as expensive cards or hard discs, which
they knew the service business could reuse. They
added these components to the stock of defective
replacement parts awaiting testing and repair in the
spare-parts loop. Because the parts-management sys-
tem did not support this spontaneous supply, plan-
ners had to take an administrative detour to make
these parts visible to the system. They kept track of
these parts as a distinct stock. Whenever the plan-
ning systems MRP engine issued an order for this
part type on the repair or the new-buy channel, they
cancelled the order manually and replaced it with an
order on the dismantling stock (Figure 2).
Gradually, IBM has recognized dismantling as
an opportunity for large savings. Disassembling
a returned machine to extract a part is fairly
inexpensive. The main costs of this supply source are
for thorough testing to ensure that all parts entering
the service network meet quality standards. Yet these
costs are generally much lower than those for buying
a new part, sometimes 80 percent lower. Compared
to the traditional repair channel, the savings are
less dramatic, yet still nonnegligible because parts
from dismantling are not necessarily defective, unlike
returned replaced parts, and therefore provide a
higher yield. The dismantling channels attractiveness
from a cost perspective is countered by its uncertain
availability, which complicates reliable planning.
46
Interfaces
Vol. 33, No. 6, NovemberDecember 2003
FLEISCHMANN, VAN NUNEN, AND GRVE
Spare-Parts Management at IBM
Dismantling
New buy
Repair
?
Machine
returns
Scrap Scrap
Service
network
Returns of
defective parts
Figure 2: By dismantling returned machines, IBM obtains an alternative source of spare parts, in addition to
buying new parts and repairing defective parts. IBM retrieves valuable parts by dismantling returned machines
that it cannot reuse as a whole, tests and possibly upgrades them, and then feeds them into its service network.
As volumes grew, the limitations of IBMs initial ad
hoc dismantling policy became apparent. Above all,
its omission on the existing information technology
system hampered its systematic exploitation. Work-
ing around the systems limitations by intervening
manually clearly was not a satisfactory solution. Fur-
thermore, the policy suffered from a lack of commu-
nication and coordination between the dismantling
operation and parts planning. Operations could dis-
mantle parts that eventually would turn out not to be
needed. At the same time, because it had little infor-
mation about needed parts it could easily miss dis-
mantling opportunities. Planners, meanwhile, might
order expensive new parts even though they could
be obtained at little cost by dismantling returned
machines. In the best case, such an order for new parts
would be cancelled. In the worst case, the new part
would eventually have to be scrapped and the invest-
ment lost.
Given these shortcomings, IBM initiated our project
in cooperation with its Global Logistics Development
Group in Amsterdam. It asked us to develop a pro-
posal for the systematic integration of dismantling as
a regular source in spare-parts planning.
Lessons Learned from Literature
Several streams of research documented in the
literature provide insight for tackling this problem.
The most straightforward link points to the theory
of repairable inventory systems, which has brought
about a rich body of literature since the 1960s
(Nahmias 1981). This work models networks of stock-
ing points of repairable items, such as spare parts.
Given a probabilistic demand process and repair
capabilities, it determines appropriate stock levels for
each item at each location. Sherbrooke (1968) did
some of the seminal work in this eld. IBM uses
related approaches for determining the stock levels of
its repairable spare parts.
A key assumption in these models is that item
returns are generated by failures in the eld and
therefore trigger immediate demands for replace-
ments. This reects the closed-loop character of
rotable spare-parts systems. However, IBMs disman-
tling source violates this assumption. Its supply stems
from end-of-lease equipment or environmental take-
back, which are largely independent of spare-parts
demand. Therefore, managing the dismantling source
Interfaces
Vol. 33, No. 6, NovemberDecember 2003 47
FLEISCHMANN, VAN NUNEN, AND GRVE
Spare-Parts Management at IBM
goes beyond the scope of traditional repairable-
inventory theory.
In recent years, several authors have developed
stochastic inventory models with item returns that are
closer to our setting. We distinguish single-level mod-
els that assume returns are added directly to the ser-
viceable stock and bilevel models in which returns
may be stocked before being recovered. Within the
former class, Heyman (1977) analyzes the disposal of
returns to optimize the trade-off between additional
inventory carrying costs and procurement savings.
Inderfurth (1997) addresses the disposal decision
and procurement policy simultaneously. In a related
paper, Fleischmann and Kuik (2003) showed that a
model with independent demand and returns can,
under certain conditions, be reduced to a conven-
tional inventory model, such that methods and results
from traditional inventory theory are applicable to the
return case. We used this model as a basic tool in ana-
lyzing IBMs situation.
The option to postpone the recovery operation until
one actually needs the returned items signicantly
complicates the inventory-control problem. In these
bilevel models, optimal policies have a simple struc-
ture in only a few special cases (Inderfurth 1997).
Therefore, researchers have proposed several heuris-
tics for this setting (van der Laan et al. 1999). Kies-
mller and Minner (2001) and Mahadevan et al. (2003)
proposed simple approximation formulas for setting
the control parameters, analogous to the usual nor-
mal demand approximations for standard inventory
models (Silver et al. 1998, Chapter 7).
In addition to these models, which explicitly con-
sider item returns, we found analogies with inventory
models from other contexts helpful. Particularly use-
ful were two-supplier models and models with price
discounts. Two-supplier models consider the coor-
dination of orders from a cheaper, yet slower sup-
plier with those from a faster, yet more expensive
one. Typically, the latter one serves as an emergency
source when stock levels drop too low (Moinzadeh
and Nahmias 1988). Similarly, for IBM, dismantling,
repair, and new buy represent alternative supply
sources that it needs to coordinate. Yet, the trade-off
is different, because the cheapest source (dismantling)
is also the fastest one.
Several authors have considered the impact of tem-
porary price discounts on inventory-control decisions
(Silver et al. 1993, Moinzadeh 1997). Discounts, in
general, result in higher stock levels. The basic trade-
off is between additional inventory costs and future
procurement savings. This setting has much in com-
mon with the situation we observed at IBM. Product
returns for dismantling can be interpreted as tem-
porary procurement opportunities with a price dis-
count. Yet, the correspondence is not perfect, because
returned products that are not recovered must be dis-
posed of and because dismantled parts must be tested
and possibly repaired before they can be reused.
Structuring the Problem
In our project, we subdivided the overall goal of
establishing a systematic dismantling source into
three core issues: (1) designing the dismantling
channel, (2) choosing which parts to retrieve from a
given machine, and (3) coordinating dismantling with
the other sources of parts.
To address the rst issue, we had to dene the pro-
cesses each part type from a returned machine would
have to undergo to be considered reusable. The gen-
eral steps include the actual disassembly, testing, and
possible repair or upgrading. In particular, we had to
take into account for which part types dismantling is
technically feasible at all. In general, IBM engineers
consider dismantling for electronic components, such
as expensive cards or hard-disc assemblies, which
often may be reused as is once they have been tested.
In contrast, they do not consider parts that are subject
to mechanical wear, such as keyboards, which can-
not be readily recovered. In addition, they take into
account design changes that require older parts to
be upgraded. For our project, the engineering depart-
ment specied the technical processing steps required
for various part types.
From a logistics perspective, we had to coordi-
nate the individual processing steps to enhance parts
ow and to locate intermediate inventory buffers.
We observed no signicant economies of scale with
respect to lot size in any of the processing steps,
and capacity limitations appeared to be negligible.
Therefore, maintaining inventories in the dismantling
48
Interfaces
Vol. 33, No. 6, NovemberDecember 2003
FLEISCHMANN, VAN NUNEN, AND GRVE
Spare-Parts Management at IBM
channel served primarily to postpone processing
expenses until parts were needed. In this light,
the most sensible location for an inventory buffer
appeared to be in front of the main value-adding
activity, namely, testing and potential upgrading.
We therefore considered two alternative channel
designs, pull versus push dismantling. In the pull
case, the rm builds up a stock of dismantled parts
against which it can place test orders when it needs
parts, in analogy with the traditional repair chan-
nel. In the push case, the rm tests dismantled parts
as they become available and adds them directly to
the serviceable stock. With the rst option, the rm
benets from postponing the investment for testing,
thereby reducing its opportunity costs and the risk
of testing parts that it no longer needs. With the sec-
ond option, the rm avoids holding defective parts
in stock and reduces the throughput time, which
may reduce safety stock. We evaluated the trade-off
between the two alternatives.
Concerning the second core issue, we had to specify
which parts to disassemble when a returned machine
arrives. While for a small-scale operation IBM could
rely on the responsible engineers intuition, growing
volumes call for a more systematic analysis of the
nancial consequences. Because machines for disman-
tling do not generally arise exactly when IBM needs
them, it must make a trade-off between immediate
costs and expected future savings. Because disassem-
bly costs were low, we considered each part indepen-
dently. Following this approach, we recommended
disassembling those parts for which the costs for
disassembly and testing discounted to the expected
moment of use are lower than the cost of disposal
plus the cost of an alternative future supply dis-
counted to the same moment. This rule comes down
to specifying a maximum expected time in stock, or
equivalently, a maximum stock level beyond which
the rm will dispose of parts (Heyman 1977).
We remark that one needs to be careful about the
holding-cost rates in this context. In our discussions,
some people argued that dismantling would become
unattractive if the salvaged parts were kept in the
expensive serviceable stock. Their point was that
this stock is valued against the weighted average
cost over all supply channels. However, a closer look
reveals that holding costs based on this valuation
are not appropriate for evaluating the trade-off of
using or not using a given dismantling opportunity.
Furthermore, in assessing expected future savings of
a current dismantling decision, the rm should mea-
sure savings relative to the supply source it would
have used otherwise. The share of each source in the
total supply can be estimated based on life-cycle plan-
ning information. Finally, we note that internal trans-
fer prices may distort the rms view of the savings
potential of the dismantling source. We observed con-
troversial debates, for example, how to deal with the
remaining book value of a returned lease machine.
We argue that the rm should decide whether or not
to recover a given part based exclusively on direct
nancial implications. Following a supply-chain per-
spective, we did not take department boundaries
into account in our analysis. This discussion actually
underlines the need for a global policy on the dispo-
sition of returns that carefully evaluates the potential
benet of alternative recovery options. In the present
study, we focused on dismantling though and took
the disposition as a given.
The third core issue that we addressed was coordi-
nating dismantling with other supply sources, notably
buying new parts and repairing replaced defective
parts. In particular, we had to decide how to reect
the dismantling supply in the order decisions for
those sources. Given the uncertainty in the disman-
tling channel, some planners we talked to were reluc-
tant to rely on any dismantling input before it was
actually available, thus advocating a reactive policy.
Yet, others argued for a proactive approach, seeking
to reduce stock levels by cutting orders to account
for expected supply from dismantling. The key quan-
tity in this context is the dismantling volume that
becomes available during the effective lead time of
other channels, which is typically several weeks.
Proactive coordination depends critically on the
information available about future dismantling
opportunities. In our project this information was
scarce. For IBM, forecasting the exact timing of
machine returns was difcult. Even for leased equip-
ment, customers typically have many options that
inuence the actual return date, for example, they
can extend the contract or purchase the machines.
Interfaces
Vol. 33, No. 6, NovemberDecember 2003 49
FLEISCHMANN, VAN NUNEN, AND GRVE
Spare-Parts Management at IBM
Moreover, the planners did not know beforehand
whether a particular machine would be available for
dismantling and exactly which parts it contained.
Because of these uncertainties, we assessed the poten-
tial benet of more reliable forecasts of returns. To
do so, we analyzed alternative reactive and proac-
tive coordination strategies, supposing that the corre-
sponding information was available.
Analytic Model and Simulation
Our approach to addressing the above issues was
twofold. First, we developed a basic inventory-control
model with product returns, which we could solve
analytically. Second, we built a simulation model that
captures more details of the problem. In both cases,
we considered the service network to be a single
aggregate stock point. How to allocate inventory to
the different locations in the network is beyond the
scope of our project.
The essence of integrating the dismantling chan-
nel into IBMs parts-planning process is about man-
aging an exogenous inbound product ow. The most
basic version of this problem is a single stock point
that faces stochastic demand and stochastic returns.
In our project, this corresponds with the case that
IBM uses all dismantling opportunities, tests disas-
sembled parts immediately upon arrival, and has
only one alternative source of supply, namely, buying
new parts. Mathematically this situation comes down
to a standard single-item inventory model in which
demand may be positive or negative. We proved that
in this situation a conventional reorder-point order-
up-to (s, S) order policy minimizes long-run aver-
age costs (1) if demand and returns are indepen-
dent, (2) if recovering parts from returns is faster than
ordering new parts, and (3) any unmet demand is
back-ordered (Fleischmann and Kuik 2003). While the
rst two conditions hold in the case of IBM, the third
does not. In a stockout situation, IBM expedites a
supply order, which mathematically corresponds to
a lost sales model. Yet, for simplicity, we used the
back-order model as a proxy in our analysis. Speci-
cally, because xed order costs played a subordinate
role in our case, we used the special form of an
(S 1, S) base-stock model. That means that when
the inventory position (stock on hand plus stock on
order minus back-orders) drops below a certain target
level S at a review instant, the rm places an order
with a supplier to return the inventory to level S
(Appendix).
Using results from queueing theory, we computed
the stationary probability distribution of the inven-
tory position. Specically, the inventory position at a
review epoch is distributed as S plus a random walk
on the nonnegative integers. Moreover, as in standard
inventory models with backlogging, we have that
the net stock is distributed as the inventory position
minus demand during a lead time plus review period,
which are stochastically independent. This allows us
to calculate the long-run average costs for our model
as a function of the target base-stock level S and thus
to compare different strategies for setting this target.
While this basic model captures much of the
essence of IBMs dismantling source, it is certainly
a simplication of the real system. In particular, we
have not included the following aspects explicitly:
that IBM has two important supply sources in addi-
tion to dismantling instead of just one, namely, new
buy and repair; that disassembled parts could be
stocked before being tested; that the dismantling
channel and the repair channel have imperfect yields;
and that IBM expedites orders to satisfy demand in
a stockout situation. The complexity of including all
of these aspects in an analytic model is prohibitive.
We therefore developed a simulation model to test the
robustness of our ndings. We included the following
elements (Figure 3):
Machine return process: Used machines containing a
particular part arrive at random moments in small
batches of varying size. Lacking reliable data, we
opted for a stationary model. Instead of modeling
the product life cycle explicitly, we ran the simu-
lation with different parameter settings to represent
different stages in the life cycle.
Dismantling selection: The decision whether to
retrieve a particular part from a returned machine or
scrap it depends on current stock levels and expected
future demand.
Dismantling process: Economies of scale are negli-
gible and, because machines arrive in small batches,
capacity restrictions are loose. Therefore, we assumed
a xed processing time and a constant cost per unit.
50
Interfaces
Vol. 33, No. 6, NovemberDecember 2003
FLEISCHMANN, VAN NUNEN, AND GRVE
Spare-Parts Management at IBM
Stock of
serviceable
parts
Scrap
Stock of
retrieved
parts
Demand
for parts
Orders for new parts
Testing
Dismantling
Machine
returns
Scrap
On order
Orders for testing
Selection
Stock of
repairable
parts
Scrap
Repair
Repair orders
Returned
defective
parts
No
return
part
Figure 3: Our simulation model reects the interaction between the dismantling channel (middle) and the exist-
ing new-buy channel (top) and repair channel (bottom). The three channels feed the inventory of serviceable
parts. The repair channel and the dismantling channel include additional intermediate stock points. The model
encompasses three sets of decisions (marked in italics), namely, selecting which returned machines to disman-
tle, deciding when to test retrieved parts, and adjusting the order strategies for the new-buy and repair sources.
Dismantling stock: Retrieved parts may be stocked
before testing or tested immediately, depending on
the dismantling channel design.
Test/upgrade process: We again assumed a xed pro-
cessing time and constant unit costs. Parts pass the
test process with a xed probability and are scrapped
otherwise.
Serviceable stock: We modeled the serviceable stock
in the network as a single inventory level.
New-buy orders: The system issues orders for new
parts periodically. They have a xed lead time and
constant unit costs.
Defective parts return process: Each demand triggers
the return of a defective part with a constant prob-
ability.
Repairable stock: Defective parts remain in inventory
until the system issues a repair order.
Repair process: In analogy with the test process, we
assumed a xed repair yield, a xed processing time,
and constant unit costs.
Demand process: In analogy with the return process,
demand for parts occurs at random moments at a sta-
tionary rate. Demand during a stockout situation is
fullled immediately but at a penalty cost.
The model encompasses three sets of decisions,
namely, choosing whether or not to dismantle a
returned machine, deciding when to test retrieved
parts, and adjusting order strategies for other sources,
that is, repair and new-buy. For each decision,
we evaluated the different candidate policies. To
this end, we compared their average annual costs,
consisting of procurement costs (new-buy order costs
and processing costs for dismantling, testing, and
repair), holding costs (serviceable stock, repairable
Interfaces
Vol. 33, No. 6, NovemberDecember 2003 51
FLEISCHMANN, VAN NUNEN, AND GRVE
Spare-Parts Management at IBM
stock, and possibly dismantled stock), and stockout
penalty costs.
Findings and Recommendations
In the analytic model, we compared three alternative
policies for new-buy orders, that is, three alternative
strategies for setting the base-stock target S. The rst
always uses the optimal base-stock level. The second
is a reactive policy that does not take the dismantling
coverage into account in determining the base-stock
target. This comes down to using the optimal target
of the case without dismantling throughout. The third
alternative is a netting policy that reduces the target
level of the case without dismantling by the expected
dismantling supply during the order lead time plus
review period (Figure 4).
In our simulation model, we compared six alterna-
tive policies, based on two alternative channel designs
and three alternative coordination mechanisms with
the original sources (new-buy and repair). The chan-
nel designs differ in whether they stock disman-
tled parts (pull) or test them immediately (push).
In the pull case, the system generates test orders
according to a base-stock policy. The three coordina-
tion approaches are analogous to those in the ana-
lytic model, namely, optimal coordination, reactive
approach, and netting. As a benchmark, we also com-
pared the six policies to a situation without any
dismantling supply (Figure 5).
The results of our two models provide a basis for
assessing the different options for integrating dis-
mantling as a source into IBMs parts planning. In
all the cases we analyzed, the procurement costs
largely outweighed inventory-related costs. Therefore,
IBM should maximize its exploitation of dismantling
opportunities. We found that the time a dismantled
part spent in inventory almost never reached the crit-
ical level, beyond which disposal would have been
preferable. Even when we limited the demand hori-
zon for dismantling to one year, the actual storage
time never exceeded three months.
These observations hold for all the scenarios we
investigated. Naturally, the overall reduction in cost
IBM could achieve by dismantling would depend on
the differences in unit-processing costs. Yet, this factor
did not change the dominance of procurement costs
over inventory-related costs. Only when dismantling
supply rose as high as 90 percent of demand did
inventory costs exceed procurement-cost savings. This
observation is in line with the results of Teunter and
Vlachos (2002).
Based on our ndings, we recommended that IBM
make use of all dismantling opportunities as long
as it had any uncovered demand. Holding costs are
hardly relevant to this decision. Scrapping reusable
parts makes sense only at the end of a parts life
cycle when retrieved parts risk not being used at all.
In particular, the rm should stop dismantling once
it has enough parts to cover the remainder of the
service horizon. Engineering changes are also a factor
to consider.
Our results also show that advance information on
returns may allow IBM to reduce its inventory of
parts signicantly by better coordinating its sources
of supply (Figures 4 and 5). A reactive approach tak-
ing dismantled parts into account only after they have
arrived consistently yields the highest costs in all the
cases we analyzed. The size of the cost advantage
to be obtained from proactive coordination depends
on several factors but can easily approach 50 percent.
This percentage tends to increase with the lead times
of the original channels, the coverage provided by dis-
mantling, and the volume of demand. As expected,
a simple netting approach of reducing orders from
the more expensive sources by the expected supply
from dismantling during their lead time is overly
optimistic, because it treats returns as a perfectly reli-
able source. Uncertainty in the return process causes
the service level to drop. Yet, from a cost perspec-
tive, netting turned out to be surprisingly close to the
optimal policy in many cases and consistently outper-
formed reactive strategies. We achieved optimal coor-
dination in our models by reducing the original order
sizes by somewhat less than the expected dismantling
supply. That is, we used a certain safety margin to
buffer against return uncertainty. The size of this mar-
gin increases with the variability of machine returns.
Our results highlight the benets of reliable fore-
casting information on the returns of used equipment.
Potential starting points for building up this type
of information include historical data, installed-base
52
Interfaces
Vol. 33, No. 6, NovemberDecember 2003
FLEISCHMANN, VAN NUNEN, AND GRVE
Spare-Parts Management at IBM
0
10
20
30
40
50
60
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Dismantling coverage
Average
cost
per
period
Optimal policy
Netting policy
Reactive policy
Figure 4: These graphs illustrate the outcomes of our analytic model (see Appendix for parameter values). They
show the average inventory and back-order costs as a function of the dismantling coverage (in percentage of
demand) for three alternative order policies, namely, the theoretically derived optimal policy, a reactive policy
that ignores parts from dismantling until they arrive, and a netting policy that subtracts the expected dismantling
output from new-buy orders. The results show that neglecting future dismantling input entails excessive stock
levels whereas a simple netting rule is close to optimal.
data, and product life-cycle data. Our models pro-
vide the basis for a cost-benet analysis of such an
information system. In addition to statistical forecast-
ing methods, advanced information technology may
allow companies to actually monitor or even control
product returns.
Whether the rm used a push strategy or a pull
strategy for dismantling had a fairly limited impact
on cost in all of the cases we analyzed. For the pull
strategy, the average time parts spent in dismantled
stock was very short. In general, postponing the test-
ing of dismantled parts yielded a slight cost advan-
tage in the case of reactive coordination, and it did so
sometimes in the case of netting. On the other hand,
a push channel slightly outperformed a pull channel
for the optimal policy. In all cases, however, the effect
of the channel design is clearly subordinate to the
impact of the coordination strategy.
Given the very limited postponement of testing
even in the case of a pull design, we recommended
testing all dismantled parts immediately. By elim-
inating the intermediate stock location, this policy
yields additional benets in terms of administration
and handling. Moreover, managing such a push pol-
icy is easier than optimizing the exact timing of the
test operation for each part. However, the rm may
benet from slightly modifying its dismantling policy
at the end of a parts life cycle. When it is unsure
whether it will ever need a particular part again, it
may make sense to put aside a number of untested
parts as a cheap additional backup.
Conclusions
We presented our ndings to IBMs internal ser-
vice logistics consulting group, which participated in
Interfaces
Vol. 33, No. 6, NovemberDecember 2003 53
FLEISCHMANN, VAN NUNEN, AND GRVE
Spare-Parts Management at IBM
0
200
400
600
800
1000
1200
1400
Original Push
reactive
Pull
reactive
Push
netting
Pull
netting
Push
optimal
Pull
optimal
Original Disman-
tling
A
v
e
r
a
g
e

a
n
n
u
a
l

i
n
v
e
n
t
o
r
y

c
o
s
t
0
10000
20000
30000
40000
50000
60000
70000
80000
90000
100000
Shortage cost
Holding cost
dismantled
Holding cost
repairables
Holding cost
serviceables
Total inventory cost
Procurement cost
A
v
e
r
a
g
e

a
n
n
u
a
l

t
o
t
a
l

c
o
s
t
Figure 5: This chart illustrates the results of our simulation model (see Appendix for parameter values). The
two rightmost columns depict total costs (procurement costs plus inventory-related costs) without dismantling
and with a dismantling coverage of 30 percent. The procurement costs are identical across all strategies that
make use of dismantling. The remaining columns break up the inventory-related costs in more detail. The left-
most column shows the case without dismantling. The other columns correspond to six alternative strategies for
integrating the dismantling source, based on two alternative channel designs (push or pull testing) and three
alternative coordination mechanisms (reactive, netting, and optimal). The results show that dismantling leads
to savings in procurement costs that are much larger than the potential increase in inventory-related costs. In
addition, the coordination strategy has a larger costs impact than the channel design.
designing an IBM-wide parts-planning system and
in managing the interface with the asset-recovery
business unit. Comments from our project partners
in this group were very positive. They appreciated
the systematic analysis of the underlying logistics
processes, which they felt helped them identify and
understand the different design alternatives. They
stated that our results provided a good basis for
integrating the dismantling source into spare-parts
management in a well-structured way. In addition,
our analysis of the alternative dismantling channel
designs has triggered an ongoing discussion on sim-
ilar issues in the repair channel. Our project partners
also believed that our clear statements concerning
the dominance of procurement costs over inventory-
related costs would help reduce skepticism within
the organization about the benets of dismantling
despite its inherent uncertainty. Moreover, the bene-
ts of coordination, which our study clearly revealed,
provided them with another strong argument for an
overall integrated parts-planning system.
In the meantime, IBM has made the disman-
tling operation the responsibility of the asset-recovery
organization. In line with our recommendation, it
retrieves parts through dismantling based on the fore-
casted uncovered demand within a certain time hori-
zon. Moreover, it tests all these parts before putting
them in stock. By moving from the original case-
by-case decision to this systematic rule, IBM has
increased the overall dismantling volume to mil-
lions of parts. In particular, it has extended its dis-
mantling portfolio from a few high-value parts to
54
Interfaces
Vol. 33, No. 6, NovemberDecember 2003
FLEISCHMANN, VAN NUNEN, AND GRVE
Spare-Parts Management at IBM
high-volume commodity parts. In this way, it has
reduced the average costs for parts and increased the
value it extracts from used equipment. Overall bene-
ts amount to several million US dollars annually. In
view of these results, IBM is incorporating disman-
tling as a regular source in its new parts-management
system.
What can managers learn from our experience?
We believe that the following lessons may be useful.
First, rms can exploit product returns as a source
of spare parts. To date, many reverse-logistics sys-
tems focus on recovering products as a whole. Yet,
ever-decreasing product life cycles limit the market
opportunities for this option. In contrast, service busi-
nesses have much longer horizons and, therefore,
have greater opportunities for extracting value from
product-return streams. This is particularly true when
sourcing options for spare parts become limited late
in the service horizon. Second, rms need to allocate
returns carefully to maximize the value recovered.
For example, the value of a products components
may exceed the value of the product as a whole.
Instead of using simplistic priority rules, the rm
should carefully analyze the nancial results of all
available reuse options. Third, information manage-
ment is key to creating an efcient closed-loop supply
chain. Many companies accept product returns as a
given and react to them only when they have them in
their hands. We have illustrated that a proactive per-
spective that recognizes returns as valuable resources
yields tangible benets. To obtain these benets, com-
panies must understand their return ows and be
able to anticipate them. While they have become very
sophisticated in forecasting demand, few companies
are collecting systematic data on product returns.
Forecasting returns is but a rst step. Even greater
opportunities come into reach for companies that use
information to actively manage their returns rather
than accept them as a given. Advances in information
technology, including data logging, radio-frequency
identication, and remote sensing provide ever more
powerful means for pursuing these opportunities.
Appendix
Denition of the (S1, S) inventory model with prod-
uct returns.
Notation
D
n
= net demand (i.e., demand returns) in
period n.

n
=inventory position at the beginning of period n
after ordering.
Y
n
=net stock at the end of period n.
S =order-up-to level.
c
l
=holding cost per item per time.
c
|
=back-order cost per item per time.
t =order lead time.
We assume (D
n
) to be a sequence of indepen-
dent, identically distributed random variables with
E[D
n
] > 0. Then, (
n
) forms a Markov chain with the
following transition rule:

n+1
=max(S,
n
D
n
).
Moreover, we have the relation
Y
n
=
nt

i=0
D
ni
,
where the two operands on the right-hand side are
independent. This allows us to evaluate the long-run
average system costs, dened as
lim
n
E|c
l
max(0, Y
n
) +c
|
max(0, Y
n
)].
Parameter settingsFigure 4
D
n
=D
+
n
D

n
.
D
+
n
Poisson(10).
D

n
Poisson(\), with \ [0; 9.75].
c
l
=$1/item/period.
c
|
=$50/item/period.
t =5.
Parameter settingsFigure 5
Part-demand process (unit/month): Poisson(100).
Part-return probability: 0.8.
Machine-return-batch arrival (unit/month):
Poisson(8.569).
Machine-return-batch size: Uniform|1, 10].
Test yield after dismantling: 0.7.
Repair yield: 0.5.
Lead time for new-buys: 2 months.
Lead time for repairs: 1.5 months.
Lead time for dismantling: 0.25 months.
Unit order cost for new-buys: $100.
Interfaces
Vol. 33, No. 6, NovemberDecember 2003 55
FLEISCHMANN, VAN NUNEN, AND GRVE
Spare-Parts Management at IBM
Unit repair cost: $50.
Unit dismantling cost: $40.
Unit stockout cost: $110.
Annual holding-cost rate: 20%.
References
European Union. 2000. Proposal for a directive of the European Par-
liament and of the Council on Waste Electrical and Electronic
Equipment. COM/2000/347. Retrieved November 4, 2002, http:
//europa.eu.int/eur-lex/en/com/pdf/2000/en_500PC0347_
02.pdf.
Fleischmann, M., R. Kuik. 2003. On optimal inventory control with
independent stochastic item returns. Eur. J. Oper. Res. 151(1)
2537.
Guide, V. D. R., Jr., L. N. Van Wassenhove, eds. 2003. Business
Aspects of Closed-Loop Supply Chains. The Carnegie Bosch Insti-
tute International Management Series, Vol. 2. Carnegie Mellon
University Press, Pittsburgh, PA.
Heyman, D. P. 1977. Optimal disposal policies for a single-item
inventory system with returns. Naval Res. Logist. Quart. 24(3)
385405.
IBM. 2000. Environment and well-being. Progress report. Retrieved
August 6, 2002, http://www.ibm.com/ibm/environment.
ICT Milieu. 2003. The ICT take-back system. Retrieved June 13,
2003, http://www.ictmilieu.nl/ictinzamelgb.htm.
Inderfurth, K. 1997. Simple optimal replenishment and disposal
policies for a product recovery system with leadtimes. OR
Spektrum 19(2) 111122.
Kiesmller, G. P., S. Minner. 2001. Simple expressions for nding
recovery system inventory control parameters. Working paper,
University of Magdeburg, Magdeburg, Germany.
Mahadevan, B., D. F. Pyke, M. Fleischmann. 2003. Periodic review,
push inventory policies for remanufacturing. Eur. J. Oper. Res.
151(2) 536551.
Moinzadeh, K. 1997. Replenishment and stocking policies for
inventory systems with random deal offerings. Management Sci.
43(3) 334342.
, S. Nahmias. 1988. A continuous review model for an inven-
tory system with two supply modes. Management Sci. 34(6)
761773.
Nahmias, S. 1981. Managing repairable item inventory systems:
A review. L. B. Schwarz, ed. Multi-Level Production/Inventory
Control Systems: Theory and Practice. North-Holland, Amster-
dam, The Netherlands, 253277.
REVLOG. 2002. The European Research Network on Reverse
Logistics. Retrieved August 6, 2002, http://www.fbk.eur.nl/
OZ/REVLOG/.
Sherbrooke, C. C. 1968. METRICA multi-echelon technique for
recoverable item control. Oper. Res. 16(1) 103121.
Silver, E. A., D. F. Pyke, R. Peterson. 1998. Inventory Management
and Production Planning and Scheduling, 3rd ed. John Wiley and
Sons, New York.
, D. J. Robb, M. R. Rahnama. 1993. Random opportunities for
reduced cost replenishments. IIE Trans. 25(2) 111120.
Teunter, R. H., D. Vlachos. 2002. On the necessity of a disposal
option for returned items that can be remanufactured. Internat.
J. Production Econom. 75(3) 257266.
Thierry, M. C., M. Salomon, J. A. E. E. van Nunen, L. N. Van
Wassenhove. 1995. Strategic issues in product recovery man-
agement. California Management Rev. 37(2) 114135.
Van der Laan, E. A., M. Salomon, R. Dekker, L. N. Van Wassenhove.
1999. Inventory control in hybrid systems with remanufactur-
ing. Management Sci. 45(5) 733747.
Daniel H. Ransdell, Vice President Hardware
Service & Support, IBM Corporation, 3039 Cornwallis
Road, Research Triangle Park, North Carolina 27709,
writes: As the worldwide executive for Hardware
Service & Support in IBM Corporation, I fully support
the referenced article. The project as described and
executed in co-operation with the team from Erasmus
University in The Netherlands has been valuable to
our organization. The article also helped to validate
our current parts strategy and, as you know, IBM has
been very active in the eld of Reverse Logistics.
The efciency of our Spare Parts Management in
IBM is very important in support of our customer
satisfaction and the nancial performance of our
Services business. As described in the article, the
authors systematic analysis has helped us to assess
different alternatives for structuring our re-utilization
process and has helped us to address the proper IT
system requirements. This process involves dozens of
warehouses and repair sites worldwide, thus truly a
global operation.
56
Interfaces
Vol. 33, No. 6, NovemberDecember 2003

Вам также может понравиться