Journal of Operations Management 28 (2010) 4757

Contents lists available at ScienceDirect

Journal of Operations Management

journal homepage: www.elsevier.com/locate/jom

The differential impact of product complexity, inventory level, and

conguration capacity on unit and order ll rate performance
David J. Closs a,*, Gilbert N. Nyaga b,1, M. Douglas Voss c,2
a
Department of Supply Chain Management, Michigan State University, N370 Eli Broad College of Business, East Lansing, MI 48912, United States
b
Operations and Analysis Group, College of Business Administration, 214 Hayden Hall, Northeastern University, Boston, MA 02115, United States
c
University of Central Arkansas, 210 Burdick Business Administration Building, Conway, AR 72035, United States

A R T I C L E I N F O A B S T R A C T

Article history: This paper examines the simultaneous impact of conguration capacity, inventory level, and complexity
Received 9 February 2007 on service performance as measured by unit and order ll rates in a congure-to-order environment.
Received in revised form 10 April 2009 Demand skew is treated as a control variable. A simulation model based on data from a leading
Accepted 13 April 2009
electronics manufacturer is used to test the hypotheses and identify the impact. Results suggest that
Available online 3 May 2009
there are differential direct and interactive effects of examined variables on unit and order ll rates.
2009 Elsevier B.V. All rights reserved.
Keywords:
Capacity management
Inventory management
Operations strategy
Simulation
Product complexity

1. Introduction increasing pressure to provide more product variety, reduce

Manufacturers often face the need to produce increasingly
complex products to meet more specic customer demands.
Complexity is manifested as the variety of components in the
nished product. In congure-to-order (CTO) production systems,
nal product conguration starts only after customer orders are
received. Typically, the nal product consists of multiple
components, each having multiple subcomponents with stochas-
tic attach rates and replenishment lead times. An attach rate is the
percentage of base models that contain a specic product feature
alternative. CTO systems are typical of computer, electronics and
equipment manufacturers that await customer orders and then
congure the customer specied unit using component inventory.
Therefore, in a CTO environment, product complexity increases
with increasing number of product components and sub-
components, and with increasing variation in component attach
rates and replenishment lead times. Graman and Magazine (2002)
posit that greater product complexity requires correspondingly
higher inventory levels. This is problematic for CTO rms given
* Corresponding author. Tel.: +1 517 353 6381; fax: +1 517 432 1112.
E-mail addresses: closs@bus.msu.edu (D.J. Closs), g.nyaga@neu.edu
(G.N. Nyaga), voss@uca.edu (M.D. Voss).
1
Tel.: +1 617 373 2116.
2
Tel.: +1 501 450 5320; fax: +1 501 450 5302.
inventory, and minimize cost while simultaneously offering
timely order fulllment. CTO rms must also maintain enough
conguration capacity to produce these increasingly complex
products responsively particularly when demand is skewed or
seasonal. Consequently, CTO rms face the challenge of cong-
uring complex products to meet a broader range of customer
requirements while simultaneously reducing inventory and
conguration capacity. Examples of CTO rms experiencing these
challenges of complexity include consumer and industrial
manufacturers such as Dell, General Electric, Hewlett Packard,
IBM, Motorola, and Sony.
Although prior research has examined inventory requirements
related to product complexity (Graman and Magazine, 2002),
limited attention is given to capacity as it relates to product
complexity. Indeed, a review of the literature has not identied
any work that has simultaneously examined the interactive
effects of product complexity, inventory level, and conguration
capacity on performance as measured by unit and order ll rates.
Yet, these interactions and related strategies might signicantly
inuence manufacturer performance. Generally, optimal perfor-
mance is achieved when a rm adopts a holistic view where
different processes and strategies are simultaneously evaluated
and aligned.
The notion of strategic alignment or t is well developed in the
manufacturing strategy literature. Specically, the competitive
priorities framework stipulates that rms tend to emphasize

0272-6963/$ see front matter 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.jom.2009.04.003
48 D.J. Closs et al. / Journal of Operations Management 28 (2010) 4757
certain competitive dimensions and develop manufacturing
capabilities to achieve the chosen dimensions to enhance their
market position (Hayes and Wheelwright, 1984; Boyer and Lewis,
2002; Hallgreen and Olhager, 2006). The competitive dimensions
are cost, quality, delivery, and exibility. Hayes and Wheelwright
(1984) posit that it is difcult, and potentially dangerous, for
rms to try to compete by offering superior performance along all
of the dimensions simultaneously. This strategy likely results in
performance that is inferior on each dimension relative to other
rms that devotes more resources to developing competitive
advantages in key dimension as perceived by the customer
(Hayes and Wheelwright, 1984, p. 41). Instead, a rm should
prioritize limited dimensions to emphasize and appropriately t
its practices to the chosen dimensions (Anderson et al., 1989;
Boyer and Lewis, 2002; Da Silveira, 2005). Manufacturing
strategy encompasses a sequence of decisions that enable a rm
to achieve desired manufacturing structure, infrastructure, and
set of specic capabilities (Hayes and Wheelwright, 1984). These
decisions relate to production process and control, technology,
capacity, facilities, workforce, planning, etc. An effective strategy
demonstrates consistency between competitive priorities with
major focus and corresponding decisions regarding operational
structure and infrastructure (Boyer and Lewis, 2002). According
to the framework, rms must strive to achieve congruence
between competitive priorities and rm processes and capabil-
ities. We argue that manufacturers must also consider con-
gruency in measuring performance of each dimension. This is
because use of appropriate measures lead to correct identication
of key capabilities requiring improvements and resulting in
enhanced performance. However, alignment is not always
achieved since many rms tend to focus on different strategic
and structural variables independently. As a result, they miss
opportunities to simultaneously identify effects of different
processes and align processes to improve performance. More-
over, rms may adopt inappropriate performance measures
leading to misalignment.
First, this research examines the simultaneous effects of
product complexity, inventory level, and conguration capacity
on unit and order ll rates in a CTO production system where the
nal product consists of components and subcomponents that
have stochastic attach rates and replenishment lead times.
Demand skew, as represented by seasonal spikes in customer
demand, is considered a control variable.
Second, this research assesses the relative impact of
product complexity, capacity, and inventory on unit and order
ll rates. Bourne et al. (2002) indicate that it is difcult to
identify the true drivers of individual performance measures.
Perhaps this is due to the relative impact of different factors on
performance measures such as unit and order ll rate.
Mirchandani and Mishra (2002) support this notion and found
that unit and order ll rates are inuenced by different factors.
Theoretically, unit and order ll rates should be highly
correlated if a customers order is composed of both many
SKUs and large quantities of each unit, with common drivers
inuencing both. However, given the work of Mirchandani and
Mishra (2002) the potential exists for one of the above
independent variables to affect unit and/or order ll rate
differently. Previous work has not examined this possibility
using empirically derived data.
The next section reviews past literature involving the variables
under investigation and denes the research hypotheses. The third
section discusses the simulation design. Fourth, simulation results
are presented and discussed. Fifth, the managerial implications of
these results are presented. Finally, study limitations and future
research opportunities are provided followed by concluding
remarks.
2. Literature review
The following literature review focuses on performance
measurement and then on the specic drivers of performance
measures.
2.1. Performance measurement
To be competitive, manufacturers must efciently and effec-
tively manage inventory and capacity in an environment with
increasing product complexity. Performance measurement is also
necessary for rms to identify and guide efforts to enhance
competitiveness (Van Hoek, 1998). The use of incorrect measures
may result in an inability to meet customer expectations, sub-
optimization of rm performance, and intra-rm conict (Lambert
and Pohlen, 2001). A commonly used availability service perfor-
mance measure is ll rate (Johnson and Scudder, 1999). Fill rates
take several forms. Unit ll rate is dened as the number of units
(e.g., cases) lled as a fraction of units ordered (Bowersox et al.,
2006) and is a disaggregate availability measure in that it considers
a single stock-keeping unit (SKU). Unit ll rate may also represent
the percentage of components available to completely assemble a
congured product. Order ll rate is dened as orders lled
complete as a fraction of [the] total number of orders (Zinn et al.,
2002, p. 22) and is an aggregate measure of availability in that it
encompasses other types of ll rates, such as unit ll rate.
Grifs et al. (2004) suggest performance measures (e.g., unit or
order ll rate) should be selected to properly align with rm
strategy. Focusing on an incorrect measure may lead to a mismatch
between rm strategy and service capabilities, which may
ultimately undermine competitiveness. At least three disconnects
potentially exist between rm performance measurement needs
and their performance measure choice (Grifs et al., 2004). First,
however, unlikely it may be, a rm might not measure
performance. In this case there exists a lack of information relating
system output to any existing performance goals. Second, incorrect
information might be measured. This occurs when a rm utilizes
unreliable or invalid measures. In this case, a rm has performance
measures in place but for some reason perhaps manipulation of
ll rates by employees for personal gain the information obtained
through these measures is incorrect. Third, a rm might utilize the
wrong measures. When the wrong measures are utilized, and the
problem goes unrecognized, the rm is more likely to take
incorrect actions due to a lack of knowledge regarding the true
state of organizational performance.
This research focuses on the third of these disconnects. If rm
protability depends substantially on availability of a particular
SKU, then unit ll rate is an appropriate performance measure. In
this situation, if a rm focuses on order ll rate, this important
product may not be afforded the attention it deserves. On the other
hand, if rm protability depends substantially on a limited
customer base, then it is important to provide each customer with
high product availability. Order ll rate is the appropriate
performance measure in this case because it measures the service
level provided to key customers with respect to the combination of
SKUs found in each order. In this situation, a focus on unit ll rate
would only capture a small portion of each customers order. For
example, consider the case of a full-line computer manufacturer,
which sells equipment ranging from PCs to servers through two
different channels utilizing a combination of strategies based upon
their CTO manufacturing structure.
First, the manufacturer sells equipment direct to customers. No
one customer in this channel represents a signicant contribution
to the manufacturers overall protability. However, certain
product congurations are more protable than others with most
per unit prot derived from a very small number of SKUs sold at
 D.J. Closs et al. / Journal of Operations Management 28 (2010) 4757
higher margins. In this channel, the manufacturer follows a
differentiation strategy by utilizing a CTO manufacturing structure
allowing for customization of machines to individual customer
preferences.
Second, the manufacturer also sells equipment through a
distributor network. Compared to the customer direct channel,
distributors order computers in high volumes then sell these
machines globally. Each local market demands different cong-
urations in terms of power supply, disk drives, memory, etc., to
coincide with local standards and preferences. Each distributor
order is relatively low prot margin but the relatively high
distributor volumes make each one an important customer.
Therefore, most prots for the distributor channel come from a
very small number of customers. The rm follows a focus strategy
in this channel, complimented by cost leadership, and utilizes a
CTO manufacturing structure to reduce nished goods inventory
commitment while simultaneously offering a variety of high
volume products at relatively low prices.
If we assume Paretos law applies in both of these scenarios,
then 80% of prots come from 20% of products in the customer
direct channel. Conversely, 80% of prots come from 20% of
customers in the distributor channel. To maximize performance
(protability, product availability, etc.), the rm is likely to adopt
different strategies and inventory policies. However, if these
strategies and policies are not properly aligned, then performance
will suffer. The competitive priorities framework emphasizes the
alignment of competitive dimensions and operational structure in
order to achieve superior market performance. Therefore, rms
must develop manufacturing and operational capabilities that
enhance competitive priorities and performance. This will, in turn,
inuence CTO operations. In addition, the rm will adopt different
performance measures for each channel. Performance for the
customer direct channel should be measured via unit ll rate due
to the focus on a relatively small number of very protable SKUs.
Distributor channel performance should be measured by order ll
rate due to the need to provide high service levels for each order for
a relatively small number of customers. Fig. 1 offers a framework
illustrating the proper alignment of channel (distributor vs.
customer direct), strategy (differentiation vs. focus/cost leader-
ship), structure (CTO manufacturing environment characterized by
endogenous and exogenous variables), and performance measures
(unit vs. order ll rate).
If a customers order is composed of both many SKUs and large
quantities of each unit, unit and order ll rate should be highly
correlated with common drivers inuencing both. Perhaps this
Fig. 1. Computer manufacturer model by differing channels and performance measurement needs.
49
explains why most studies investigating the drivers of ll rate only
examine an aggregate or disaggregate measure, not both
simultaneously. Prior literature has not identied any research
based on empirically derived data examining relative differences
in aggregate (i.e., order ll) and disaggregate (i.e., unit ll)
performance measures. Mirchandani and Mishra (2002) develop
optimization models to examine the effects of complexity on
disaggregate performance measures. They also draw differences
between aggregate and disaggregate measures in their work and
nd that disaggregate measures are better suited to capture the
effect of inventory shortage costs. Johnson and Scudder (1999)
examine the effects of production scheduling on item ll rate in a
make-to-order (MTO) environment. An aggregate measure, such as
order ll rate, is not examined. Baker et al. (1986), Eynan and
Rosenblatt (1996), and Hillier (1999) measure service levels on an
aggregate level.
In general, unit ll rate reects a rms ability to meet specic
product demand, while order ll rate reects the rms ability to
satisfy demand for a broad range of products. This research
considers both unit and order ll rate because they could
conceivably be driven by different factors as demonstrated by
Mirchandani and Mishra (2002). This research empirically
investigates this possibility.
2.2. Drivers of performance measures
Many manufacturing rms are faced with the need to offer an
extensive product portfolio to remain competitive (Fixson, 2005).
However, product complexity, as reected in the present work by
the number of product variants a rm produces, typically increases
the number of component and production process variations.
Complexity research has focused on inventory benets from
decreased complexity (Fisher and Ittner, 1999; Hillier, 2000),
reductions in procurement cost that result from purchase size
economies (Meyer and Mugge, 2001), and R&D effectiveness
(Meyer et al., 1997; Qiang et al., 2004) among other topics. Product
complexity has been shown to inuence holding costs (Johnson
and Anderson, 2000; Alfaro and Corbett, 2003), service levels
(Alfaro and Corbett, 2003), and delivery reliability (Mapes et al.,
1997). Lee and Tang (1997) nd that increasing product complex-
ity increases inventory and decreases service levels. Alfaro and
Corbett (2003) nd that as the number of SKUs increase, inventory
increases and service level decreases. However, Johnson and
Anderson (2000) nd that complexity has no effect on service
levels as measured by ll rate. These ndings suggest that with
50 D.J. Closs et al. / Journal of Operations Management 28 (2010) 4757
proper inventory support, CTO rms should be able to produce
increasingly complex products and meet customer service
requirements. However, there is disagreement regarding the effect
of product complexity on service levels.
Given that product complexity makes inventory management
and production processes more difcult, it is expected to degrade
service performance. This expectation is based on the potential for
increases in manufacturing order cycle time variation, production
change-overs, and time required to produce more complex
products. Further, previous research has not examined the relative
impact of product complexity on performance in terms of
aggregate and disaggregate service measures. Hypotheses 1A
and 1B examine the impact of complexity on service levels and the
potential differential impact on disaggregate and aggregate
measures of service performance:
H1A. Product complexity negatively affects unit ll rate service
performance.
H1B. Product complexity negatively affects order ll rate service
performance.
As previously discussed, product complexity requires increased
component and sub-component inventory. In this research, the
nal product consists of multiple components, each having
multiple subcomponents with stochastic attach rates and replen-
ishment lead times. A stock-out of any component or subcompo-
nent would delay nal product availability (Lu et al., 2003).
Therefore, component and sub-component inventories are neces-
sary to support productions ability to congure sufcient nal
products to meet customer demand. Graman and Magazine (2002)
posit high levels of product variety require correspondingly higher
levels of inventory. Gerchak et al. (1988) nd that a reduction in
product variety reduces inventory levels. Adequate inventory is
necessary to meet customer requirements and the rms target
service levels. Therefore, inventory is expected to have a positive
effect on service performance. Similar to complexity, previous
studies have not examined the relative impact of inventory on
performance in terms of aggregate and disaggregate service
measures. Hypotheses 2A and 2B examine the impact of inventory
on service levels and the potential differential impact on
disaggregate and aggregate measures of service performance:
H2A. Inventory level positively affects unit ll rate service per-
formance.
H2B. Inventory level positively affects order ll rate service per-
formance.
A major challenge for CTO rms is allocating conguration
capacity to meet customer demand. Several studies have examined
the problems posed by capacity constraints. For example,
Marucheck and McClelland (1992) examine implications of
varying levels of capacity utilization in an assemble to order
(ATO) environment. They argue that since conguration capacity
level is not always able to respond to demand changes, a rms
resource level becomes an operational constraint with respect to
capacity allocation. Balakrishnan et al. (1996) examine capacity
allocation problems faced by manufacture to order (MTO) rms
encountering periods when expected demand exceeds available
capacity. They demonstrate that capacity rationing, where
managers preferentially allocate available conguration capacity
depending on the marginal contribution of the product, can
effectively increase total prot for rms manufacturing seasonal
products. Bish et al. (2005) examine the use of exible capacity
management in a MTO environment to protect against demand
variability by adjusting production levels to match demand,
prioritize higher margin products, and satisfy customer demands.
They argue that such a strategy results in drastic production
swings, larger upstream order variability, higher levels of
component inventory, and greater operational costs. Bradley and
Arntzen (1999) also examine the problem of simultaneous
planning of production, capacity, and inventory in seasonal
demand characterized by quarterly demand skew. Whereas these
and other studies address the nite capacity allocation problem,
there is limited consideration of the interaction between product
complexity and capacity. Indeed, no studies examine capacity
utilization when conguring generic products with multiple
components and subcomponents in an environment characterized
by stochastic lead times and quarterly demand skew. However, it is
clear that availability of capacity necessary to congure products
to fulll customer demand has service performance implications.
Moreover, this may have different effects on aggregate and
disaggregate measures of service performance. Therefore, Hypoth-
eses 3A and 3B state:
H3A. Conguration capacity positively affects unit ll rate service
performance.
H3B. Conguration capacity positively affects order ll rate ser-
vice performance.
Demand skew is treated as a control variable since it is an
exogenous variable largely outside of rm control. Monthly,
quarterly, or annual demand skew is characteristic of many
electronic and durable products to reect the promotional and
pricing activities used to meet rm nancial reporting demands.
Demand skew is included in the following analysis where the
interaction of demand skew signicantly adds to the discussion of
the above hypothesized relationships.
3. Experimental design
This research uses an ARENA 5.0 simulation model combining
dynamic simulation with a graphics component (Kelton et al.,
2003). A 3  3  3  4 experimental design is used to generate
data and test hypotheses. The design consists of three levels of
inventory, three levels of conguration capacity, four levels of
complexity, and three levels of demand skew. There are 108
scenarios. Each scenario is replicated 30 times, resulting in 3240
observations. Based on Law and Keltons (2000, p. 512) procedure
for calculating the required number of replications to estimate
population mean at a specied precision, it is determined that 15
replications per scenario is appropriate.i However, since increased
sample size improves accuracy and condence in results, thirty
replications per scenario are performed. The model simulates a
rm manufacturing one type of machine that can be congured
using a combination of ve components. Each of the ve
components has ve variants. Model parameters are derived from
demand and conguration characteristics of a leading computer
manufacturer. The simulation network consists of component
manufacturing, sub-assembly manufacturing, and conguration
center. Components are produced at component manufacturing
and then delivered to sub-assembly manufacturing. Sub-
assembled components are delivered to the conguration center
i
According to the procedure, an approximate expression for the minimum
number of replications, na b, required to obtain an absolute error of b is given by:
q
na b minfi  n : t i1;1a=2 S2 n=i  bg. na b can be determined by iteratively
q
increasing i by 1 until a value of i is obtained for which t i1;1a=2 S2 n=i  b. For
example, to estimate unit ll rate with an absolute error, b, of 0.05 and a condence
level, a, of 90%, initial pilot runs were performed and initial mean (m) and standard
deviation (S2) were computed. From 15 replications, a S2 = 0.02 was obtained
q
resulting in i = 0.05, which satises t i1;1a=2 S2 n=i  b requirement.
 D.J. Closs et al. / Journal of Operations Management 28 (2010) 4757
Fig. 2. List of events in the simulation.
to await nal CTO assembly according to customer specications.
There is a 1-day lead time for this delivery as these processes are
co-located. Fig. 2 illustrates the sequence of events on a typical
simulated day.
At the beginning of day, replenishment shipments scheduled to
arrive (in-transit inventory) are received at the conguration
center (event 1). Inventory level is updated. New customer orders
are then received (event 2). If required components are available to
congure ordered products, the orders are lled and shipped to
customers (event 3). Replenishment orders are then created (event
4). These orders are received at the subassembly manufacturing
(event 5). If required components are available, subassembly takes
places and orders are shipped to the conguration center (event 6).
Replenishment orders are then sent to component manufacturing
(event 7) where production of components takes place (event 8).
The last event (event 9) is the collection of statistics (ll rates and
average inventory). Individual statistics are aggregated to compute
overall performance. The system returns to the rst event after the
last event is complete, repeating the same sequence until the
simulation period is over. Orders that are not fullled are discarded
and recorded as stockouts. Stochastic production and replenish-
ment lead times are imposed at different stages of the simulation
process.
There are ve components (e.g., disk, memory, power supply, I/
O card, and communications card) that are combined with a central
processing unit (CPU) to congure a salable machine. For each
component, there are ve variants available (e.g., disk component
variants are disk 1, disk 2, disk 3, disk 4, and disk 5). These variants
represent disk drives of different capacity and speed. The daily
demand for CTO computers is stochastic with a triangular
distribution (i.e., a truncated normal distribution) that is modeled
after historical demand distribution of a leading computer
manufacturer. The daily unit volume is based on the unit volume
of the manufacturer for a single conguration site. The order sizes
are also based on seasonalized daily averages. The daily production
capacity is correspondingly matched with the aggregate volume.
However, specic daily volume may exceed or be less than daily
production capacity due to actual demand variation. The aggregate
daily component demand has a triangular distribution: disk
TRIA(4,10,20), memoryTRIA(1,4,10), power supplyTRIA(1,2,4), I/
O cardTRIA(1,2,3), and communicationsTRIA(2,4,6). In ARENA,
TRIA (2,4,6) results in minimum daily demand of two, a mode of
four, and a maximum of six.
3.1. Independent variables
Four independent factors are examined: product complexity,
inventory target level, conguration capacity, and demand skew.
51
To test the effect of complexity, the number of variants available
per component is reduced from ve to four, three, and then two.
Each variant has a different attach rate. Attach rates decrease from
the rst variant to the last variant for each component as shown in
Table 1. The disk attach rates and number of component variants in
Table 1 are replicated for the other component types (i.e., memory,
I/O card, etc.). These attach rates reect a typical Pareto demand
distribution where there are a few high volume and several low
volume component variants. Thus, product complexity grows
when the possible component variants are considered. As
complexity is reduced from ve variants to four, three, and two,
it can be assumed that some, but probably not all of the volume
from the discontinued variants is merged into the remaining
variants as illustrated in Table 1. The consideration of the potential
volume reduction from reduced variant offerings is beyond the
scope of this research.
The rm employs an order-up-to inventory control policy for
the components. According to this policy, whenever component
inventory declines below the target level, an order is placed to
bring the combination of on-hand inventory and on-order up-to
the target level. The delivery lead time from component
manufacturing to sub-assembly is modeled as a TRIA (1,2,4). This
means that the minimum lead time is 1 day, the maximum lead
time is 4 days, and that most deliveries require 2 days. This
research uses three target inventory levels: 25 days, 30 days, and
35 days using a monthly forecast to predict average daily sales rate.
Inventory is reviewed every 7 days with the potential of making a
replenishment order during each replenishment cycle. While more
frequent review might change the specic performance results, it
should not change the signicance of the interactions. Initial
inventory is based on daily demand averages derived from
historical data of the electronics manufacturer. The sourcing
process is performed at the three network stages: components
manufacturing, sub-assembly manufacturing, and conguration
center.
While component availability at the component plant is
assumed to be unlimited (albeit with a stochastic lead time to
Table 1
Disk variant attach rate.
Attach rate (%) by number of disk variants
5-Variants 4-Variants 3-Variants 2-Variants
Disk 1 40 42 47 59
Disk 2 30 32 35 41
Disk 3 15 16 18
Disk 4 10 10
Disk 5 5
52 D.J. Closs et al. / Journal of Operations Management 28 (2010) 4757
get it to the sub-assembly plant), conguration capacity is modeled
at three levels: 100%, 150%, and 200%. A 100% capacity level reects
the conguration capacity utilization required to ll mean daily
demand of nal product if demand volume is spread equally across
the month and quarter (i.e., no skew). The 150% and 200% capacity
levels assume that the rm maintains buffer capacity to congure
50% and 100%, respectively, more than the average daily demand.
In essence, rms maintain buffer capacity to meet demand during
months when demand is greater than 33% of quarterly volume (i.e.,
high demand skew). Interviews with electronics industry man-
agers suggest that set-up time is not a signicant factor since there
are no substantial economies of scale in the conguration process.
Therefore, set-up time is included in the unit conguration time for
this analysis.
Demand skew is modeled at three levels: low, medium, and
high. When demand skew is low, demand in each month of the
quarter represents 33% of quarterly demand. In essence, demand is
constant across months. When demand skew is medium, monthly
demand is modeled as 20%, 30%, and 50% of quarterly demand, for
the rst, second, and third months, respectively. High demand
skew is modeled as 10%, 20% and 70% of quarterly demand for the
rst, second, and third months, respectively. These quarterly skew
levels are typical in the electronics industry due to end of quarter
sales pushes to meet quarterly nancial projections.
3.2. Dependent variables
The simulation examines customer service performance as
measured by order ll rate and unit ll rate. In effect, an order
represents a customer request for one or more machines (i.e.,
units) congured with the desired components. In order to achieve
100% order ll rate, a rm must ll all customer orders completely.
A customer order often contains requests for several different
machine congurations (i.e., SKUs). Unit ll rate indicates the
percentage of these specic units lled by the rm. Higher unit ll
rate and order ll rates are associated with better customer service.
3.3. Model verication and validation
Model verication ensures that the computerized model and its
implementation are correct while model validation ensures that
the computerized model possesses a satisfactory range of accuracy
consistent with intended application of the model (Sargent, 2000).
Four types of model validity and verication are considered. First,
conceptual model validity tests that the theories and assumptions
underlying the conceptual model are correct and that the
simulation model representations are reasonable. To ensure
conceptual model validity, managers of four different companies
were asked to analyze and evaluate each event proposed in the
model. The managers concurred that relationships proposed in the
conceptual model were reasonable.
Second, computerized model verication ensures that the
computer programming and implementation of the conceptual
model are correct. To accomplish this, the model was tested to see
if it was programmed correctly in ARENA 5.0 simulation language.
Traces were used to track entities throughout the simulation
model. A log le with every activity generated in the model was
created. Information and product ows were then tracked and
analyzed. It was conrmed that the model operated accurately and
according to proposed logic.
Third, operational validity determines whether model output
behavior has sufcient accuracy. Three techniques were used to
test for operational validity. First, degenerative tests were
completed by appropriate selection of values of the input and
internal parameters. For example, when production lead-time was
increased to replicate a constrained environment, there was an
increase in backorders and substantial decrease in service levels.
Secondly, internal validity checks were performed by making
several replications of the model to determine the amount of
internal stochastic variability. Coefcient of variability (CV) was
used to assess variability. The average CV across scenarios
indicated low variability across different replications (between
0.01 and 0.02). Third, parameter variabilitysensitivity analysis,
which involves changing the values of model input parameters to
identify the impact on the models behavior and output, was
performed. For example, when inventory levels were decreased,
service levels declined, while lead time increases resulted in
increased cycle times and in-transit inventory. Therefore, the
model achieved operational validity.
Finally, data validity ensures that data necessary for model
building, model evaluation and testing, and conducting of model
experiments are adequate and correct. Simulation data were
screened for outliers and tested using internal consistency checks.
Reviews were conducted with company experts and any observed
errors were corrected.
3.4. Statistical analyses
Multivariate analysis of variance (MANOVA) and univariate
analysis of variance (ANOVA) are used to test each research
hypotheses. For MANOVA results to be valid, dependent variables
(i.e., unit and order ll rates) should be correlated. To test this
assumption, Bartletts test of sphericity is used (Hair et al., 1998).
The test result (Bartletts = 133196.4 with 5df, p < 0.000) indicates
that use of MANOVA in this study is appropriate.
4. Analysis Of results
Table 2 summarizes MANOVA and univariate ANOVA results.
There are four main effects, six two-way interactions, four three-
way interactions, and a four-way interaction. When interpreting
MANOVA results, interaction terms must be examined rst (Hair
et al., 1998). Multivariate results indicate that ve interactions are
signicant: inventory level and product complexity (IL  PC)
(p < .001); inventory level and demand skew (IL  DS) (p < .001);
product complexity and demand skew (PC  DS) (p < .001);
conguration capacity, product complexity, and demand skew
(CC  PC  DS) (p < .05); inventory level, product complexity, and
demand skew (IL  PC  DS) (p < .001). All main effects are
signicant (p < 0.001).
Although multivariate results indicate that CC  PC  DS effect
is signicant (p < 0.001), univariate results show that the effect on
unit ll rate and order ll rate is not signicant. IL  PC  DS
signicantly affects unit ll rate (p < 0.001) but not order ll rate.
As Figs. 24 indicate, high product complexity (5-variants) results
in lower unit ll rate than less complex products irrespective of
inventory levels and demand skew. The difference is most marked
when inventory level is lowest (i.e., 25 inventory days) and
demand skew is highest. As product complexity increases,
different components are required for each nal product, which
results in increased inventory requirements. Moreover, high
demand skew is associated with a greater need for both nal
product and component inventory. Therefore, for all levels of
demand skew and product complexity, higher inventory levels
result in higher unit ll rate (Fig. 5).
The interaction between inventory level and product complex-
ity (IL  PC) signicantly inuences unit ll rate (p < 0.001). As
Fig. 6 illustrates, unit ll rate decreases as product complexity
increases for all levels of inventory. However, the decrease is
greater at 25 inventory days than at 35 inventory days, which
suggests that greater inventory provides the rm with more
exibility to meet demand when product complexity is high.
 D.J. Closs et al. / Journal of Operations Management 28 (2010) 4757 53

Table 2
Multivariate and univariate test resultsa.

Source of variation Multivariate results Univariate results

F-Ratio Deg. of freedom Sig. Partial h2 F-Ratio Deg. of freedom Sig. Partial h2

Conguration capacity (CC) 358.7 4.6262 0.001 0.186

Order ll rate 799.8 2.3132 p < 0.001 0.338
Unit ll rate 0.01 2.3132 p > 0.05 0.000

Inventory level (IL) 1667.1 4.6262 0.001 0.516

Order ll rate 2.68 2.3132 p > 0.05 0.002
Unit ll rate 5106.9 2.3132 p < 0.001 0.765

Product complexity (PC) 1889.6 6.6262 0.001 0.644

Order ll rate 6.2 3.3132 0.001 0.006
Unit ll rate 7192.7 3.3132 0.001 0.873

Demand skew (DS) 391.1 4.6262 0.001 0.200

Order ll rate 4.4 3.3132 0.001 0.003
Unit ll rate 874.7 3.3132 p < 0.05 0.358

CC  IL 0.0 8.6262 p > 0.05 0.000

CC  PC 0.0 12.6262 p > 0.05 0.000
CC  DS 0.3 8.6262 p > 0.05 0.000

IL  PC 261.2 12.6262 0.001 0.334

Order ll rate 0.46 6.3132 p > 0.05 0.001
Unit ll rate 652.7 6.3132 0.001 0.556

IL  DS 130.3 8.6262 0.001 0.143

Order ll rate 0.03 4.3132 p > 0.05 0.000
Unit ll rate 282.2 4.3132 0.001 0.265

PC  DS 65.2 12.6262 0.001 0.111

Order ll rate 0.53 6.3132 p > 0.05 0.001
Unit ll rate 138.1 6.3132 0.001 0.209

CC  IL  PC 0.0 24.6262 p > 0.05 0.000

CC  IL  DS 0.0 16.6262 p > 0.05 0.000
CC  PC  DS 0.0 24.6262 p > 0.05 0.000

IL  PC  DS 17.1 24.6262 0.001 0.062

Order ll rate 0.01 12.3132 p > 0.05 0.000
Unit ll rate 35.4 12.3132 0.001 0.119

CC  IL  PC  DS 0.0 48.6262 p > 0.05 0.000

a
Note: Source of variation column shows main and interaction variables. F-ratio column shows the F-statistic, which tests whether the effect of each independent variable is
signicant. Deg. of freedom column shows the degrees of freedom used to obtain observed signicance levels. Sig. column shows the signicance level (p-value). Partial h2
column shows the effect size with larger values indicating stronger effect. Interaction variables are abbreviated. For example, CC  IL represents interaction between
conguration capacity and inventory level. CC  IL  PC represents interaction between conguration capacity, inventory level, and product complexity. Other interactions
are represented in a similar format.

Results also indicate that two-way interactions between inventory
level and demand skew (IL  DS) and product complexity and
demand skew (PC  DS) are signicant (p < 0.001). However,
examination of marginal means for both interactions demon-
strates that the effects are negligible.
Univariate test results (see Table 2) indicate that capacity has a
signicant effect on order ll rate (p < 0.001), but not on unit ll
rate. An evaluation of marginal means indicates that increased
conguration capacity results in a marked increase in order ll
rate. Greater capacity enables more orders to be assembled to
completion. Not surprisingly, the effect of capacity on unit ll rate
is not signicant. Unit ll rate is a disaggregate measure of service
based on single units (SKUs). However, order ll rate is an
aggregate measure of service affected by availability of multiple
units (SKUs). Therefore, shortage of one SKU signicantly
inuences order ll rate. Fig. 7 illustrates that increasing capacity
from 100% to 150% results in a 4% increase in order ll rate while
raising capacity from 150% to 200% results in 1% increase in order
ll rate. This suggests that there is a level beyond which increasing
capacity results in marginal or negligible improvements in order
ll rate. Consequently, when determining capacity buffers to
manage anticipated demand surges or skew, managers need to

Fig. 3. Effect of IL  PC interaction on unit ll rate at low demand skew. Fig. 4. Effect of IL  PC interaction on unit ll rate at medium demand skew.
54 D.J. Closs et al. / Journal of Operations Management 28 (2010) 4757
Fig. 5. Effect of IL  PC interaction on unit ll rate at high demand skew.
Fig. 6. Effect of IL  PC interaction on unit ll rate.
Fig. 7. Effect of conguration capacity on order ll rate.
identify the capacity level that provides for highest order ll rate at
different levels of product complexity.
Inventory level signicantly impacts unit ll rate but not
order ll rate. This is not surprising. Order ll rate is based on the
availability of multiple SKUs and not necessarily the volume of
each SKU. Therefore, order ll rate may not be affected by
inventory level, unless inventory drops below a certain mini-
mum threshold. However, availability and volume of individual
SKUs does impact unit ll rate. Indeed, partial h2 values (partial
h2 = 0.765) for the impact of inventory level on unit ll rate
show the effect is quite pronounced. Increasing inventory days
implies that more component inventory is available to support
production, which results in a higher unit ll rate as illustrated in
Fig. 8.
Product complexity negatively impacts both order and unit ll
rates. However, partial h2 values indicate that complexity has a
stronger inuence on unit ll rate than order ll rate (partial
h2 = 0.873 and 0.006, respectively). Complex products have greater
component variation. To ensure the availability of these compo-
nents, rms must either hold inventory of each component or plan
for their timely delivery (supply), culminating in greater inventory
and/or supplier management complexity. This, in turn, affects
rms ability to ll customer orders in a timely manner.
Consequently, as complexity increases, order and unit ll rates
decrease as indicated in Figs. 9 and 10. However, the decrease in
Fig. 8. Effect of inventory level on unit ll rate.
Fig. 9. Effect of product complexity on unit ll rate.
Fig. 10. Effect of product complexity on order ll rate.
order ll rate is marginal as illustrated in Fig. 9 and by the low
partial h2 value (0.006) (see Table 2).
These results provide support for some hypotheses. Hypotheses
1A and 1B stipulate that product complexity negatively impacts
unit and order ll rate, respectively. Results show that product
complexity signicantly impacts both measures. Therefore,
Hypotheses 1A and 1B are supported. Hypotheses 2A and 2B posit
that inventory level positively impacts unit and order ll rates,
respectively. Results show that inventory level only impacts unit
ll rate. Therefore, Hypothesis 2A is supported but 2B is not
supported. Hypothesis 3A and 3B suggest that conguration
capacity positively impacts unit and order ll rate, respectively.
However, results indicate that conguration capacity impacts
order ll rate but not unit ll rate. Therefore, Hypothesis 3A is not
supported while 3B is supported.
5. Managerial implications
Research results demonstrate that product complexity, inven-
tory level, and conguration capacity differentially inuence unit
and order ll rates. Capacity signicantly affects order ll rate,
whereas inventory increases are shown to signicantly affect unit
ll rate but not order ll rate. In other words, increased inventory
increases unit ll rate. However, without adequate capacity, it does
not result in increased order ll rate. The results demonstrate the
 D.J. Closs et al. / Journal of Operations Management 28 (2010) 4757
differential and interactive effect of inventory, demand skew,
complexity, and capacity on order ll rate. These ndings have
signicant managerial and theoretical implications.
First, results demonstrate that product complexity signi-
cantly affects unit ll rate. As the number of component
variations increase, unit ll rate declines substantially. This has
two implications. First, it is much more difcult to maintain very
high unit ll rates (>98%) when there are many component and
subcomponent variations. Second, it requires substantially more
component inventory to maintain a high unit ll rate (>98%)
when the nal product has more component variants (i.e.,
products consisting of many component variants require
signicantly higher levels of component inventory). This implies
that when offering products with many variations, managers
must take into account the substantial increase in component
inventory required to maintain high unit ll rates. This will
ultimately result in higher inventory carrying cost and poten-
tially worse nancial performance (Shah and Shin, 2007).
However, do rms necessarily maintain higher inventory when
dealing with many product variations? Since CTO production
systems are designed to reduce inventory while simultaneously
improving responsiveness to diverse customer requirements, a
substantial increase in inventory may not be appropriate. Instead,
managers need to develop and adopt efcient inventory manage-
ment policies to support exible production systems and greater
responsiveness to customer demand. For example, managers
could adopt proactive demand management policies such as
offering incentives to customers to order product modules and
features that require readily available inventory. Such
approaches will ensure higher ll rates, efcient inventory
utilization, and improved performance.
Second, rms that segment their customers by prot contribu-
tion should measure customer service performance in terms of
order ll rate. Order ll rate measures the percentage of completed
orders delivered to a customer and enables rms to monitor the
extent to which they are lling orders placed by the most protable
customers and establish corrective measures if performance is
below expectations. Results indicate that increasing conguration
capacity signicantly improves order ll rate. Thus, managers
should establish efcient capacity planning, allocation, and
utilization strategies to improve service performance when
measured in terms of order ll rate. On the other hand, if rms
segment their products by prot contribution, unit ll rate should
be utilized because of its focus on single SKUs. Results indicate that
increasing inventory level signicantly improves unit ll rate
while higher product complexity reduces unit ll rate. Therefore,
managers need to maintain higher component inventories for
products with high prot contribution to realize improved
performance as measured by unit ll rate. These results suggest
that managers should adopt performance measures that align with
their rms strategic focus. Unit ll rate should be used to measure
availability of the most important products while order ll rate
should be used to measure service levels provided to the most
important customers.
Third, since higher inventory levels are necessary when
component complexity is high, managers must be willing to bear
higher inventory carrying costs in order to improve service
performance. Results show that there is an optimal inventory
level. Increasing inventory beyond this level does not markedly
improve performance. The challenge for managers is to identify
the optimal inventory level, which is likely to vary by product
and demand characteristics. While this study did not develop an
inventory optimization policy, past studies have suggested
several inventory optimization heuristics that could form a
basis for policy formulation. This research underscores the need
to optimize inventory levels by carefully assessing trade-offs
55
between service improvements with increases in inventory
levels and corresponding increase in inventory carrying cost.
Managers must also consider other factors such as demand
uncertainty, supply lead times, competitor actions, and customer
reactions to service failure when justifying the level of inventory
to maintain.
Fourth, as might be expected, demand skew and capacity
constraints exacerbate the problems introduced through complex-
ity. As quarterly skew increases, service performance declines,
which means additional assets are required in the form of
conguration capacity or inventory to improve performance.
Again, the implication is that managers must consider the cost and
service trade-offs when determining needed buffer capacity to
maintain high service performance in an environment character-
ized by demand skew. In addition, managers might consider
adopting a capacity rationing approach whereby a portion of
capacity is reserved for high priority products or customers
(Balakrishnan et al., 1996). Such an approach enables the rm to
gain from best selling product modules during peak demand (when
demand skew is high) or provide high ll rates for most protable
customers.
Finally, although results suggest that conguration capacity
signicantly improves order ll rate, this is not the case when
product complexity increases. While increasing product complex-
ity reduces the probability that the correct combination of
components is available, more conguration capacity does not
make it any more or less likely that a specic component or the
correct combination of components would be available. The
implication is that increased conguration capacity will neither
improve nor decrease service as product complexity increases.
While this might seem counterintuitive, the uncertainty intro-
duced by complexity may not be systematically reduced through
increased capacity. It is probable that availability of component
variants rather than capacity to congure nal product is more
critical in determining order ll rate. Further, the interaction
between product complexity, inventory level, and conguration
capacity was not signicant. This implies that factors not examined
in this research may drive order ll rate when complexity
increases.
Table 3 summarizes the research results relating rm strategy
to objectives. If a rm is primarily focused on key distributors with
a strong interest on enhanced order ll rate, the CTO rm should
focus on increasing conguration capacity, reduced product
complexity, or reducing demand skew. Since they all represent
main effects, these three tactics will enhance order ll rate
regardless of the other environmental variables. If the rm is
primarily focused on individual products with a strong interest on
enhanced unit ll rate, the CTO rm should focus on increasing
component inventory, reduced product complexity, or reduced
demand skew. Besides these main effects, there are some
interactive effects which require consideration of trade-offs to
identify a strategy to enhance unit ll rate. The best combined
tactic to enhance unit ll rate is through reduced product
complexity and increased inventory. The second option is reduced
demand skew and increased inventory. The third option is reduced
demand skew and reduced product complexity. The nal option is
reduced demand skew, reduced product complexity, and
increased inventory. It is interesting to note that the best
combined tactic (reduced complexity and increased inventory)
is entirely within the control of the rm. The other tactics require
the rm to change the external environment by smoothing out the
markets response to end-of-quarter expectations. Changing
these expectations have proven to be very difcult. The
synthesized results strongly indicate that rms can achieve
enhanced unit ll rates by reducing product complexity, increased
inventory levels, or combination of both.
56 D.J. Closs et al. / Journal of Operations Management 28 (2010) 4757
Table 3
Strategy recommendations based on simulation results.
Service objectives Demand Product Conguration
skew complexity capacity
Order ll rate Yes
Unit ll rate
Order and unit ll rate Yes
Order and unit ll rate Yes
Unit ll rate Yes
Unit ll rate Yes
Unit ll rate Yes Yes
Unit ll rate Yes Yes
6. Limitations and opportunities for future research
These simulation results are based on historical data from a
leading electronics manufacturer. Although the generalized model
(Fig. 1) suggests that the results should apply to different
industries, future research should extend this work by investigat-
ing the proposed relationships outside of the electronics industry.
To our knowledge this is the rst study to explicitly compare
aggregate and disaggregate performance measures in this manner
and considering the interactions of product complexity, inventory,
capacity and demand skew. Further comparisons of aggregate and
disaggregate performance measures would represent a signicant
contribution to the operations literature. These ndings suggest
that component complexity, inventory level and conguration
capacity impact unit and order ll rate differently. Component
complexity results in both a lower unit and order ll rate.
Inventory level is directly related to unit ll rate only. On the other
hand, conguration capacity directly impacts only order ll rate.
The rationale for the difference is that increasing inventory
increases the availability of individual components but it is
necessary to have all of the correct components and the capacity to
completely ll the order. While 9598% unit ll rates for
components are high, the probability of a stockout of a single
component is still quite high when there are multiple components.
Thus, the inventory level does not substantially impact the order
ll rate until unit ll rates are very high (>99%). On the other hand,
conguration capacity directly impacts only order ll rate since it
is the binding constraint for the conguration of multiple SKUs to
fulll an order. Having enough components to meet unit ll rate
objectives is of limited benet without available capacity. Due to
the requirement that multiple components must be matched with
sufcient capacity to complete an order, the inuence of inventory
is not demonstrated except on unit ll rate for the range of values
tested. The interactive effects suggest that conguration capacity
and inventory level tend to be the major drivers of order ll rate.
While the parameter values for component complexity, inventory
level, and capacity are representative of the electronics industry,
the implied coefcients of variation offer results that can be
generalized across multiple industries. Specically, industries
Inventory Management implications
level
In CTO environments, increased conguration capacity can be used
to enhance order ll rate regardless of demand skew, product
complexity, or inventory level.
Yes In CTO environments, increased inventory levels can be used to
enhance unit ll rate regardless of demand skew, product
complexity, and conguration capacity.
In CTO environments, reduced product complexity (fewer component
variations) allows the rm to enhance both order and unit ll rate
regardless of demand skew, conguration capacity, and inventory level.
However, reduced product complexity has a much stronger
effect on unit ll rate.
In CTO environments, both order and unit ll rates can be enhanced
through reduced demand skew regardless of product complexity,
conguration capacity, and inventory level. However, reduced
demand skew has a stronger effect on unit ll rate.
Yes In CTO environments, unit ll rate can be enhanced through
reduced product complexity and increased inventory.
Yes In CTO environments, unit ll rate can be enhanced through
reduced demand skew or increased inventory.
In CTO environments, unit ll rate can be enhanced through
reduced demand skew or reduced product complexity.
Yes In CTO environments, unit ll rate can be enhanced through a
combination of reducing demand skew or product complexity
and increasing inventory.
characterized by relatively high component complexity, high
seasonality, and relatively limited production exibility (either
due to limited capital equipment or labor) could benet from the
results presented. In addition to electronics, other industries that
could consider these results include toys, furniture, along with
construction and agricultural equipment. Future research may
want to consider other drivers including replenishment lead time
uncertainty, increased number of components and nished goods
variations, and possible application of pre-assembly for appro-
priate components. Future research might also extend this analysis
to situations with other industry characteristics or other perfor-
mance measures (e.g., order cycle time and protability) that
might provide further insights and comparisons with results of this
work.
From a theoretical standpoint, this research demonstrates that
it is important to examine the simultaneous impact of factors
affecting service performance in a CTO environment. Where
appropriate, it is important that researchers examine both the
aggregate and disaggregate measures of performance. Failure to
highlight the differences in these measures, and to identify which
measures apply in which context, could be misleading since there
are signicant differences in what they capture. Future research
may consider examining each of the factors in isolation to obtain
more in-depth results.
7. Conclusion
Performance measurement is an issue of great importance.
Performance measures should be chosen to elicit desired employee
behavior, equitably assess performance of functions and indivi-
duals, complement rm strategy, and even benchmark perfor-
mance against set goals or industry leaders. This study explores the
nuances of performance measurement by specically investigating
how capacity, inventory, complexity, and demand individually and
interactively affect unit and order ll rate. Results indicate that
there are differences in strength and signicance of each factor. The
ndings indicate that there are differences with regard to
improving rm performance and these differences have been
highlighted in this work.
 D.J. Closs et al. / Journal of Operations Management 28 (2010) 4757
References
Alfaro, J.A., Corbett, C.J., 2003. The value of SKU rationalization in practice: the
pooling effect under suboptimal inventory policies and non-normal demand.
Production and Operations Management 12 (1), 1229.
Anderson, J.C., Cleveland, G., Schroeder, R.G., 1989. Operations strategy: a literature
review. Journal of Operations Management 8 (2), 133158.
Baker, K.R., Magazine, M.J., Nuttle, H.L.W., 1986. The effect of commonality on safety
stock in a simple inventory model. Management Science 32 (8), 982988.
Balakrishnan, N., Sridharan, V., Patterson, J.W., 1996. Rationing capacity between
two product classes. Decision Sciences 27 (2), 185214.
Bish, E.K., Muriel, A., Biller, S., 2005. Managing exible capacity in a make-to-order
environment. Management Science 51 (2), 167180.
Bourne, M., Neely, A., Platts, K., Mills, J., 2002. The success and failure of performance
measurement initiatives. International Journal of Operations and Production
Management 22 (11), 12881310.
Bowersox, D.J., Closs, D.J., Cooper, M.B., 2006. Supply Chain Logistics Management.
McGraw-Hill, New York, pp. 1656.
Boyer, K.K., Lewis, M.W., 2002. Competitive priorities: investigating the need for
trade-offs in operations strategy. Production and Operations Management 11
(1), 920.
Bradley, J.R., Arntzen, B.C., 1999. The simultaneous planning of production, capacity,
and inventory in seasonal demand environments. Operations Research 47 (6),
795806.
Da Silveira, G.J.C., 2005. Marketing priorities, manufacturing conguration, and
business performance: an empirical analysis of the order-winners framework.
Journal of Operations Management 23 (6), 662675.
Eynan, A., Rosenblatt, M.J., 1996. Component commonality effects on inventory
costs. IIE Transactions 28 (2), 93104.
Fisher, M.L., Ittner, C.D., 1999. The impact of product variety on automobile
assembly operations: empirical evidence and simulation analysis. Management
Science 45 (6), 771786.
Fixson, S.K., 2005. Product architecture assessment: a tool to link product, process,
and supply chain design decisions. Journal of Operations Management 23 (34),
345369.
Gerchak, Y., Magazine, M.J., Gamble, A., 1988. Component Commonality with
Service Level Requirements. Management Science 34 (6), 753760.
Graman, G.A., Magazine, M.J., 2002. A numerical analysis of capacitated postpone-
ment. Production and Operations Management 11 (3), 340357.
Grifs, S.E., Cooper, M., Goldsby, T.J., Closs, D.J., 2004. Performance measurement:
measure selection based upon rm goals and information reporting needs.
Journal of Business Logistics 25 (2), 95118.
Hayes, R.H., Wheelwright, S.C., 1984. Restoring our Competitive Edge: Competing
Through Manufacturing. Wiley, New York.
Hair, J.F., Anderson, R.E., Tatham, R.T., Black, W.C., 1998. Multivariate Data Analysis,
5th ed. Prentice Hall, New Jersey.
Hallgreen, M., Olhager, J., 2006. Differentiating manufacturing focus. International
Journal of Production Research 44 (18/19), 38633878.
57
Hillier, M.S., 1999. Component commonality in a multiple-period inventory model
with service level constraints. International Journal of Production Research 37
(12), 26652683.
Hillier, M.S., 2000. Component commonality in multiple-period, assemble-to-order
systems. IIE Transactions 32 (8), 755766.
Johnson, M.E., Anderson, E., 2000. Postponement strategies for channel derivatives.
International Journal of Logistics Management 11 (1), 1935.
Johnson, M.E., Scudder, G., 1999. Supporting quick response through scheduling of
make-to-stock production/inventory systems. Decision Sciences 30 (2), 441
467.
Kelton, D.W., Sadowski, R.P., Sadowski, D.A., 2003. Simulation With Arena, 3rd ed.
McGraw-Hill, New York.
Lambert, D.M., Pohlen, T.L., 2001. Supply chain metrics. International Journal of
Logistics Management 12 (1), 119.
Law, A.M., Kelton, W.D., 2000. Simulation Modeling and Analysis, 3rd ed. McGraw-
Hill, New York.
Lee, H.L., Tang, C.S., 1997. Modeling the costs and benets of delayed product
differentiation. Management Science 43 (1), 4053.
Lu, Y., Song, J., Yao, D.D., 2003. Order ll rate, lead time variability, and advance
demand information in an assemble-to-order system. Operations Research 51
(2), 292310.
Mapes, J., New, C., Szwejczewski, M., 1997. Performance trade-offs in manufactur-
ing plants. International Journal of Operations and Production Management 17
(10), 10201040.
Marucheck, A., McClelland, M., 1992. Planning capacity utilization in an assemble-
to-order environment. International Journal of Operations and Production
Management 12 (9), 1838.
Meyer, M.H., Mugge, P.C., 2001. Make platform innovation drive enterprise growth.
Research-Technology Management 44 (1), 2539.
Meyer, M.H., Tertzakian, P., Utterback, J.M., 1997. Metrics for managing research
and development in the context of the product family. Management Science 43
(1), 88111.
Mirchandani, P., Mishra, A.K., 2002. Component commonality: models with pro-
duct-specic service constraints. Production and Operations Management 11
(2), 199215.
Qiang, T., Mark, A.V., Ragu-Nathan, T.S., Bhanu, R.N., 2004. Measuring modularity-
based manufacturing practices and their impact on mass customization
capability: a customer-driven perspective. Decision Sciences 35 (2),
147168.
Sargent, R.G., 2000. Verication, validation, and accreditation of simulation models.
In: Proceedings of the Winter Simulation Conference.
Shah, R., Shin, H., 2007. Relationships among information technology, inventory,
and protability: an investigation of level invariance using sector level data.
Journal of Operations Management 25 (4), 768784.
Van Hoek, R., 1998. Measuring the unmeasurable: measuring and improving
performance in the supply chain. Supply Chain Management 3 (4), 187193.
Zinn, W., Mentzer, J.T., Croxton, K.L., 2002. Customer-based measures of inventory
availability. Journal of Business Logistics 23 (2), 1943.