Modularity, product variety, production volume, and component sourcing

Journal of Operations Management 20 (2002) 549–575

Modularity, product variety, production volume, and componentsourcing: theorizing beyond generic prescriptions

F. Salvadora,b,∗, C. Forzab, M. Rungtusanathamaa Arizona State University, Department of Management, P.O. Box 874006, Tempe, AZ 85287, USA

b Università di Padova, Dipartimento di Tecnica e Gestione dei sistemi industriali, Stradella S. Nicola 3, 36100 Vicenza, Italy

Received 8 August 2001; received in revised form 26 October 2001; accepted 11 March 2002

Abstract

Research in operations management suggests that firms can mitigate the negative impact of product variety on operationalperformance by deliberately pursuing modularity in the design of product family architectures. However, modularity is nota dichotomous property of a product, as different types of modularity can be embedded into a product family architecture.The present paper explores how manufacturing characteristics affect the appropriate type of modularity to be embedded intothe product family architecture, and how the types of modularity relate to component sourcing. The study is based on aqualitative research design involving a multiple case study methodology to examine six product families belonging to sixEuropean companies. The themes derived through case analyses are synthesized in the form of empirical generalizations.Insights from these empirical generalizations are subsequently developed into two propositions explaining why and underwhat conditions these empirical generalizations might hold for a product family outside of the original sample. The theoreticalresults formalize, first of all, a type of modularity (i.e. combinatorial modularity) not currently described in literature. Second,the theoretical propositions suggest that when the desired level of product variety is low (high) relative to total productionvolume, component swapping modularity (combinatorial modularity) helps to maximize operational performance. Finally,the complexity of component families outsourced to suppliers and the geographical proximity of component family suppliersaffect the extent to which the product variety–operational performance trade-off can be mitigated through modularity.© 2002 Elsevier Science B.V. All rights reserved.

Keywords:Case study research; Product variety; Sourcing; Modularity; Supply chain management; Operational performance

1. Introduction

In an effort to better respond to heterogeneouscustomer needs, many firms find it appropriate toincrease product variety, i.e. the number of differentproducts offered to customers (Pine II, 1993). In do-ing so, firms are convinced that they maximize thefit between product offerings and customer desires,which can allow them to defend, if not increase, their

∗ Corresponding author.E-mail address:[email protected] (F. Salvador).

market shares. While this conscious decision mightallow a firm to better align what it offers in the marketto customer requirements, such a decision tends toalso present the firm with a number of challenges withrespect to the performance of its operations. In fact,as product variety increases, a firm would experiencelower performance of its internal operations becauseof higher direct manufacturing costs, manufactur-ing overhead, delivery times, and inventory levels(Anderson, 1995; Child et al., 1991; Fisher and Ittner,1999; Flynn and Flynn, 1999; Forza and Salvador,in press; Kotteaku et al., 1995; Miller and Vollmann,

0272-6963/02/$ – see front matter © 2002 Elsevier Science B.V. All rights reserved.PII: S0272-6963(02)00027-X

550 F. Salvador et al. / Journal of Operations Management 20 (2002) 549–575

1985; Prasad, 1998). Likewise, as product varietyincreases, there would likely be an increase in the va-riety of at least some purchased product components(Fisher et al., 1999), especially when vertical integra-tion is low. Consequently, suppliers may experiencediseconomies due to component variety, with potentialnegative impact on component prices, delivery times,and component inventory levels (Krishnan and Gupta,2001; McCutcheon et al., 1994). From the firm’s per-spective, therefore, a trade-off exists between productvariety and operational performance, which includes,in this study, performance of its internal operations,as well as its component sourcing performance.

Both research and practice commonly suggest thatfirms may mitigate this trade-off by deliberately pur-suing modularity in designing their final product ar-chitectures, which obtains final product configurationsby mixing and matching sets of standard components(Starr, 1965). At the crux of this suggestion, is the no-tion that modularity permits firms to increase productvariety without incurring substantial negative impacton operational performance.

This notion has encouraged, at least, two paral-lel streams of research, with one stream in the op-erations management/management science domain(Erens and Verhulst, 1997; Feitzinger and Lee, 1997)and the second in the design theory/engineeringmanagement domain (Huang and Kusiak, 1998;Pahl and Beitz, 1984). While research in opera-tions management/management science has generallyleaned towards understanding how internal operationschange, or should be managed, when firms modu-larize their product architectures, research in designtheory/engineering management has focused morecentrally on the issues of how to formally define whata modular product is, how a product architecture can,or should be modularized, and how design activi-ties change, or should be managed, in the design ofmodular products.

While both operations management/managementscience research and design theory/engineering man-agement research have contributed substantially tocurrent comprehension of how and why product mod-ularity may alleviate the product variety–operationalperformance trade-off, one limitation is that onlysporadic attempts have been made to integrate resultsacross the two domains for richer theoretical andpragmatic insights (He et al., 1998; Ulrich, 1995).

As such, we have an opportunity to bridge thesetwo research streams and to respond to the growingawareness of the interdependence between productdesign and operations strategy (Fine, 1998; Hoekstraand Romme, 1992).

In this research, we aim to extend our understandingof how the product variety–operational performancetrade-off can be reduced by borrowing from the designtheory/engineering management discipline the funda-mental argument that different “types of modularity”can be embedded into a product family architecture.More precisely, we develop, articulate, and justify aset of theoretical propositions explicating the condi-tions that make a given type of modularity better thananother, in reducing the product variety–operationalperformance trade-off. To this end, we first review, inSection 2, pertinent literature related to the issues un-der investigation. We then describe, inSection 3, thequalitative research design involving a multiple casestudy methodology. InSection 4, we present briefprofiles of each case, highlighting how the productvariety–operational performance trade-off is mani-fested for the product families included in our study.We devoteSections 5 and 6to report the results ofacross-case analyses.Section 5 describes and de-fines a type of modularity that we observed from oursampled product families, and which is not currentlypresent in published typologies.Section 6reports aset of empirical generalizations relating the types ofmodularity to product variety to component sourc-ing. In Section 7, we articulate and justify theoreticalpropositions drawing on the insights from the empir-ical generalizations. We conclude inSection 8withsuggestions for theory and practice.

2. Background

Researchers and practitioners alike have sought,for a long time, to identify interventions to reduce theproduct variety–operational performance trade-off.Among many suggestions, one commonly consideredintervention revolves around modular product design,or simply, modularity. The basic idea behind mod-ularity is to “. . . design, develop, and produce. . .parts which can be combined in the maximum num-ber of ways” (Starr, 1965, p. 38), thereby enhancingthe compatibility between product variety require-

F. Salvador et al. / Journal of Operations Management 20 (2002) 549–575 551

ments and operational performance for discrete prod-ucts (Hoekstra and Romme, 1992; Karmarkar andKubat, 1987). Research on how to reduce the productvariety–operational performance trade-off throughmodularity in the design of products can be furtherclassified into one of the two disciplinary areas: oper-ations management/management science and designtheory/engineering management.

2.1. Modularity research in operationsmanagement/management science

In general, operations management/managementscience research has treated modularity as a means toincrease commonality across different product vari-ants within a product family, i.e. to allow for thesame component(s) to be used in multiple productvariants, and when feasible, in all product variants(Evans, 1963). As such, one motivation underlyingthe operations management research stream has beento understand the benefits of component commonalityon operational performance, as well as the variousfactors that might affect these benefits (Baker et al.,1986; Collier, 1981; Fisher et al., 1999; Karmarkarand Kubat, 1987; Mather, 1986; Sheu and Wacker,1997; Vakharia et al., 1996).

In addition, research in operations manage-ment/management science has highlighted that mod-ularity in product design may allow for the design ofa loosely coupled production system in which differ-ent subassemblies can be made independently, andthen rapidly assembled together in different ways,given technical constraints, to build the final productconfigurations (Ernst and Kamrad, 2000; Novak andEppinger, 2001). One important consequence of thisincreased flexibility in the allocation of manufactur-ing tasks, as many scholars have observed, is thatfirms can more effectively pursue a postponementstrategy (Feitzinger and Lee, 1997; Lee and Tang,1997; Mather, 1986; van Hoek, 2001; van Hoek andWeken, 1998).

2.2. Modularity research in designtheory/engineering management

While research in operations management tends totreat modularity as given, one primary focus of mod-ularity research in design theory/engineering manage-

ment is epitomized by the question of how to im-plement modularity into the design of product fami-lies with multiple variants (Erens and Verhulst, 1997;Erixon, 1996; Gonzalez-Zugasti et al., 2000, Huangand Kusiak, 1998; Jiao and Tseng, 1999; Parnas, 1971;Parnas et al., 1985; Stevens et al., 1974). The moti-vating perspective underlying this research pursuit isthe realization that a modular product family archi-tecture, by mapping specific functional requirementsto specific product components, facilitates design ac-tivities. Accordingly, changes in a specific functionalrequirement would affect only a given product com-ponent and not the entire product family architec-ture (Huang and Kusiak, 1998; Pahl and Beitz, 1984;Suh, 1990; Ulrich, 1995), and would, therefore, easeproduct configuration activities both in the design andin the manufacturing realms.

Furthermore, research in design theory/engineeringmanagement has explored the properties a modularproduct family may display, implicitly suggesting thatthere are different types of modularity (seeTable 1).Pahl and Beitz (1984)hinted at multiple propertiesthat may differentiate a module from another, explor-ing in particular the notion of module function, dis-tinguishing between basic, special, and adaptive mod-ules.Ulrich and Tung (1991)proposed a classificationfor types of modularity based on the geometric con-figurations of a product family architecture. In fact,their typology captures different possible approachesto combining modules, distinguishing between variantand common modules.Ulrich (1995), more recently,proposed a typology that relied on the property of theinterface among modules as the classification criterion.

Finally, we should note that modularity research indesign theory/engineering management has also beenconducted for reasons unrelated to product variety.For example, research in this domain has sought todecompose complex design problems into more easilymanageable sub-problems (Baldwin and Clark, 2000von Hippel, 1990; Sundgren, 1999) or to investi-gate the impact on product performance (Kusiak andHuang, 1996).

2.3. Synthesis

Although, modularity research in the two domainshas mostly advanced independently of one another(Swaminathan and Tayur, 1999), there have been some


Table 1Typologies of modularity

References Classification criterion Types of module/modularity

Pahl and Beitz (1984) Stability of the functionallocated to the component

Basic and auxiliary modules implementfunctions that are common throughout theproduct familySpecial modules implement complementaryand task-specific functions that do not needto appear in all the product variantsAdaptive modules implement functionsrelated to the adaptation to other systemsand to marginal conditions

Ulrich and Tung (1991) How the final productconfiguration is built

Component swapping modularity: productvariants are obtained by swapping one ormore components ( , , ) on thecommon product body ( )Fabricate-to-fit modularity: product variantsare obtained by changing a continuouslyvariable feature ( , , )within a given componentBus modularity: product variants areobtained by matching any selection ofcomponents from a set of component types( , , ) with a component that hastwo or more interfaces ( )Sectional modularity: product variants areobtained by mixing and matching in anarbitrary way a set of components (, , ),as long as they are connected at their interfaces

Ulrich (1995) Nature of the interfacebetween components

Slot modularity: interfaces between

different components are different (, )

Sectional modularity: all the components

are connected via identical interfaces ()Bus modularity: special case of sectionalmodularity where there is a singlecomponent, the bus( ),performing the connection function

recent attempts to bring the two domains together.Some of these attempts have generated theoreticaldiscussions on the interdependencies between productdesign and manufacturing decisions in planning prod-uct families with multiple variants (Erens and Verhulst,1997; Forza and Salvador, 2002a; Ishii et al., 1995;Shirley, 1990; Ulrich, 1995). Others have adopted amore quantitative orientation, developing algorithms

and procedures for designing modular product fam-ilies while considering manufacturing constraints(Garg, 1999; Gupta and Krishnan, 1999; He et al.,1998; Krishnan and Gupta, 2001; Krishnan et al.,1999; Raman and Chhajed, 1995). With the excep-tion of Duray et al. (2000), none of the publishedresearches bridging product design and operationsmanagement issues attempted to develop insights on


the implications of implementing different types ofmodularity on the product variety–operational perfor-mance trade-off. Neither has there been any attemptto discuss the conditions under which a given typeof modularity may be more appropriate than another,provided that they are viable.

The present research aims to fill this gap in the lit-erature. First, our intent is to understand, whether ornot the type of modularity that should be embeddedinto a product family architecture relates to key man-ufacturing variables from the product–process matrix(Hayes and Wheelwright, 1984), namely, productvariety level (i.e. the number of final product vari-ants within a given product family) and productionvolume (i.e. total yearly production output of theproduct family), and how these relationships affectthe operational performance (e.g. inventory, deliv-ery times, manufacturing overhead costs, etc.) of theproduction system assembling the product family.Our first research question, as such, can be stated asfollows.

What are the relationships, if any, among the typesof modularity, product variety level, production vol-ume, and operational performance within a givenproduct family?

Second, as previously observed, since product vari-ety tends to affect component suppliers as well, espe-cially when vertical integration is low, and, therefore,demands increased flexibility from the supply chain,we intend to explore a second related question as fol-lows.

Given the decision to allocate to suppliers themaking of components that come in multiplevariants, how is the type of modularity embed-ded in the product family architecture related tothe final assembler’s component sourcing per-formance [e.g. component inventories and sourc-ing lead-times] and, more generally, operationalperformance?

3. Research design

To develop theoretical and pragmatic insights intothe researched problem, we employed a qualitative re-search design involving six case studies. The multiplecase study approach, as noted byEisenhardt (1989)

andVoss et al. (2002), is well suited for the empiricaldevelopment of testable theories and like other quali-tative research methods, “(is) particularly oriented to-wards exploration, discovery, and inductive logic. . . ”(Patton, 1990, p. 44).

3.1. Unit of analysis and level of analysis

The unit of analysis, in this study, is the productfamily, defined as (a) a set of final products thatare offered by a single company; (b) are partiallysubstitutable in their demands, possessing underly-ing similarities in their functionality; and (c) sharethe same common design and assembly process(Gupta and Krishnan, 1998; Meyer et al., 1997).Given the research agenda, the product family is anappropriate unit of analysis, since modularity in prod-uct design is essentially a property of product setsoffered by a company (Shirley, 1990) that typicallyaddress heterogeneous, but interrelated, customerneeds (Sanchez, 1999), and that can be produced bythe same process, thus allowing for economies ofscope (Garud and Kumaraswamy, 1995).

Specifying the product family precisely is an im-portant task that needs to be complemented with thesecond task of specifying the hierarchical level withinthe product family at which measurements are to betaken (seeAcademy of Management Review, April1999). In this context, we have to clearly denote thehierarchical level of the bill of materials defining theproduct family architecture at which we wish to makestatements about, and related to, the type of mod-ularity. Consider, for example, the typical personalcomputer (PC) comprising a set whose constituentsare the motherboard inside a case, mouse, keyboard,printer, and monitor. In the language of the bill ofmaterials, the PC is at level 0, whereas the mother-board, mouse, keyboard, printer, and monitor are allat level 1. If we were to specify the PC as the prod-uct family and we are interested in the type of mod-ularity embedded into the PC architecture at level 0,then the type of modularity that the PC embeds is“slot modularity.” However, we may be interested inthe modularity property of the PC at a lower level ofthe bill of materials, say at the level of the mother-board, and if such were the case, then the prevailingtype of modularity, we would observe, becomes “busmodularity.”


3.2. Reference population and sampling

Besides specifying the unit of analysis and the levelof analysis, it is also important to identify clearly whatthe reference population is. Doing so, helps to con-trol for variation within cases and implicitly sets theboundaries of the theoretical insights that will eventu-ally be developed (Eisenhardt, 1989; Yin, 1988).

First, the reference population, given the unit ofanalysis, is necessarily constrained to product familieswith the potential to embed modularity in the designof the final product architecture. As such, we consideronly discrete products that can be built by assemblingtwo or more components. Second, we include in thereference population only firms selling durable goods,either for the industrial market, or for the consumermarket. Durable, typically more expensive, goodsare more likely to be chosen through quasi-rational,rather than prevalently emotional, decision processes.As such, it becomes more critical for the firm to op-timize the product variety–operational performancetrade-off, rather than to bias the customer buying deci-sion through marketing actions. Third, in order to con-trol for variation in the manufacturing environment,as well as in its relationships to entities upstream anddownstream within the supply chain, we consider onlyfinal assemblers (or product integrators, according tothe practitioner jargon) in the reference population.Finally, we include, as part of the reference popu-lation, only product families with a relatively largeproduction volume (>10,000 units), so as to excludemanufacturing environments with extremely low pro-duction volume. In fact, when production volume isvery low there is generally very little margin to im-prove sourcing performance and, therefore, exploringmodularity issues related to sourcing performance, inthis context, is of little practical relevance.

Having delimited the reference population, wethen identify the specific cases using two samplingstrategies: criterion sampling and stratified purpose-ful sampling (Patton, 1990). Whereas the criterionsampling approach increases the chance of selectinginformation-rich cases whose study will illuminatethe issues under study (Patton, 1990), the stratifiedpurposeful sampling approach aims to maximizeobserved variance among selected cases.

In executing the criterion sampling approach, weconsider two criteria—firm size and market presence.

The first criterion leads to the selection, from the ref-erence population, of firms with at least 1000 employ-ees so as to improve the opportunities to interact withprofessionally trained managers, reducing languagebarriers and simplifying interaction during qualitativeinterviewing. The second criterion selects only thosefirms with an international market presence, using thiscriterion as a proxy for the firm’s ability to develop,produce, and market a successful product family. Mar-ket success for durable goods would reduce the chancethat product families and operations strategies of theselected firms were fatally flawed, while increasing thelikelihood of getting important insights on the topicsof interest.

In addition, we stratify the sampling from the ref-erence population in two ways: by industry and byconsidering simultaneously product variety level andproduction volume. We considered three different in-dustries: transportation vehicles, telecommunicationequipment and food processing machines. Havingmore than one industry in the sample reduces the riskof developing theoretical insights that are bounded toa specific industry or a specific type of product. Withineach industry stratum, we select two cases—one casewith low product variety level and high productionvolume and a second case with high product varietylevel and low production volume. This stratificationapproach is consistent with our research questionsand is motivated by the product–process matrix byHayes and Wheelwright (1984)and byMcCutcheonet al. (1994).

The sampling grid, along with the six selected cases,is identified inTable 2. FromTable 2, we can see thatproduct families with high product variety level andlow production volume tend to be more complex thantheir counterparts with low product variety level andhigh production volume, since they tend to be com-prised of a larger number of elementary parts (e.g.nuts, o-rings, bushings, etc.). Intuitively, this makessense since a product whose architecture has many el-ementary parts offers a larger potential for generatingvariants than one with fewer elementary parts. How-ever, a complex product does not necessarily mean thatthe product must offer a large number of variants (e.g.Ford Model T). As such, product variety and com-plexity, as defined here, are indeed intertwined and di-rectly related for cases with high product variety leveland low production volume. On the other hand, for


Table 2Sampling grid

Low product variety level, high production volume High product variety level, low production volume

Industry

Transportation vehiclesTrendy-moped Heavy-truck

Product Mopeds Product TrucksFamily turnover 165 Family turnover N/ANo. of variants 64 No. of variants 2500Production volume 55000 Production volume 30000No. of variants/product volume (×1000) 1.16 No. of variants/product volume (×1000) 83.3

Telecom equipmentCustom-phone Multiplexer

Product Cell-phones Product MultiplexersFamily turnover 350 Family turnover N/ANo. of variants 250 No. of variants >1000Production volume 2000000 Production volume 30000No. of variants/product volume (×1000) 0.12 No. of variants/product volume (×1000) >30

Food processing equipmentMicrowave Techoven

Product MW ovens Product Convection ovensFamily turnover 52 Family turnover 38No. of variants 280 No. of variants 300Production volume 320000 Production volume 10000No. of variants/product volume (×1000) 0.8 No. of variants/product volume (×1000) 30.0

cases with low product variety level and high produc-tion volume, the product architecture can have eithera few or many elementary parts. More importantly, inreality, it would be impossible to identify any productfamily with over 10,000 variants that is comprised ofonly a relatively few elementary components.

3.3. Data collection

Table 3 provides an overview of the procedureswe follow for the collection and subsequent analysesof data. In collecting field data, we rely primarilyon qualitative, open-ended interviews with key or-ganizational informants, including the departmentheads of product development, of manufacturing,and of purchasing. Because a firm’s operations strat-egy and its approaches to managing the productvariety–operational performance trade-off are largelythe outcomes of decisions by top and middle man-agers, interviewing these upper-level managers is,therefore, a legitimate way to directly tap their men-tal models. Once key themes begin to emerge from

the first round of interviews, we augment the datasources with ancillary data sources (Denzin, 1978),while contemporarily triangulating the informationobtained from manager interviews.

Furthermore, during the first round of data collec-tion, we make a conscious decision to not performin-depth case-by-case analyses of the interview data soas to “avoid imposing meaning from one participant’sinterviews on the next” (Seidman, 1998, p. 96). In-stead, in the second round data collection, we dynam-ically adjust follow-up interview protocols in order togain maximum insights into the themes emerging fromfirst round interviews.

3.4. Case analyses

Once the interviews are concluded, we performintra-case analyses adopting the coding techniquesrecommended byStrauss (1987). We first cluster in-terview data into large conceptual categories (opencoding) and subsequently identify sub-categories (ax-ial coding) according to an indented coding scheme

556F.

Sa

lvad

or

et

al./Jo

urn

al

of

Op

era

tion

sM

an

agem

en

t2

0(2

00

2)

54

9–

57

5


(Rubin and Rubin, 1995, pp. 248–249). The purposeof these analyses is to gain a clear understandingof the different contexts embedded within the casesand to identify the main themes and/or variablesrelevant to the research agenda. We subsequently an-alyze across the six cases, looking for patterns amongdifferent themes and/or variables, as suggested byRunkel (1990).

3.5. Empirical generalizations and theoreticalpropositions

The themes derived through case analyses are thensynthesized in the form of empirical generalizations,i.e. isolated statements summarizing observed unifor-mities of relationships between two or more variablesalong the sample (Merton et al., 1959). Empirical gen-eralizations, by definition, are not theoretical proposi-tions or laws (Merton et al., 1959; Zetterberg, 1954)since they are not backed by a fully deployed theo-retical basis (Wallace, 1971). Therefore, in order todevelop theoretical propositions from empirical gen-eralizations, we would need to: (1) state the boundaryconditions for theoretical propositions; (2) definerelevant constructs that play a role in the theoreticaldevelopment; and (3) provide explanations for the pro-posed relationships among constructs (Dubin, 1969).

4. Case profiles

In this section, we provide a brief outline of the sixfirms, highlighting specifically the market and envi-ronmental forces behind the need to reduce the prod-uct variety–operational performance trade-off in theconsidered product families. All six firms are multi-national companies doing business in, at least, theEuropean market.

4.1. Case 1: Trendy-moped

This particular firm is a major European firm op-erating in the moped market. The product familyunder study—which we label Trendy-moped—is tar-geted at the sports moped market segment. In 1996,Trendy-moped had three competitors, offering five tosix rival product families. By the end of year 2000,the number of competitors had increased to 13, with

approximately 40 rival product families, all com-peting for the same market segment. The growingnumber of competitors, especially the entrance oflow-cost far-east manufacturers, has led to escalatingprice competition, which the firm expects to becomeeven fiercer in the next few years. Because price isan important factor in the typical customer’s pur-chasing decision, and the product is expected to bereadily available at the dealership, these factors forcethe industry, as a whole, to respond to the marketin a make-to-stock fashion. In turn, the industry isrequired to hold finished product stocks in the distri-bution channel, making it extremely difficult to keepcosts low.

4.2. Case 2: Heavy-truck

This particular firm is a global player in the truck(light, medium, heavy, and quarry/construction) indus-try. The product family under study—which we labelHeavy-truck—is a “heavy road truck” product fam-ily. Deregulation of the European transportation in-dustry had resulted in the focalization of logistic ser-vice providers on specific transportation missions and,consequently, the requirement for differentiated vehi-cles. At the same time, deregulation had led to theconsolidation of logistic service providers, increasingtheir bargaining power, and putting higher pressureon prices. From the customers’ perspective, reliable,rather than short, delivery times are critical.

4.3. Case 3: Custom-phone

This particular firm is a fast-growing actor in thecell-phone industry, operating in Europe and in thefar-east. The product family under study—which welabel Custom-phone—is the “cellular-phones” prod-uct family complying with the GSM standard. Inorder to gain a foothold in a market dominated bya few technology leaders, the firm’s strategy is tosupport the marketing campaigns of mobile commu-nication providers by customizing cell phones to theneeds and to the specific technical features of eachprovider’s telecommunication infrastructure. Thiscustomization strategy, however, cannot come at thecost of substantial price increases, since competitorsare mass producers with low unit production costs.


Furthermore, short delivery times, given the very dy-namic cell-phone market, are mandatory for the firm.

4.4. Case 4: Multiplexer

The firm is a global leader in telecommunicationsequipment. The product family under study—whichwe label Multiplexer—is a data transmission unitdeployed in building the telecommunications infras-tructure of the so-called “information highways”.Liberalization of the European telecommunicationsmarket had initiated radical changes not only tothe competitive scenario for the telecommunicationsservices industry, but also to the supplying firmsof telecommunications equipment. Customers nowrequire promptly available, relatively inexpensive,tailored solutions that allow them to rapidly and ef-ficiently adapt to the effervescent and competitivetelecommunications services market.

4.5. Case 5: Microwave

This particular firm is a diversified multinationalcompany serving the home appliances industry.The product family under study—which we labelMicrowave—is a microwave oven product familycommercialized in the European market. The firmbegan to offer greater product variety during the lastdecade for various competitive reasons. The firmwanted firstly to escape from the pressure of low-cost,entry-level products imported from far-east manufac-turers, and secondly to respond to the requirementsof more and more powerful retailers, who wish todifferentiate their retail product offerings from eachother. At the same time, powerful distributors hadbeen exerting increasing pressures on price and ser-vice levels that, operating the industry according toa make-to-stock fashion, tended to translate into fi-nal goods inventory. The firm is, therefore, squeezedbetween the requirement for variety and the require-ments for competitive price.

4.6. Case 6: Techoven

This particular firm is part of a multinational groupin the food service equipment industry. The productfamily under study—which we label Techoven—is aconvection oven product family. The highly diversi-

fied needs of food service equipment customers (e.g.hotels, restaurants, canteens, etc.), both in terms of thekind of cooking capability required and oven capacity,as well as the multiple commercial brands owned bythe firm, forced the firm to develop a very articulatedproduct range. At the same time, pressure on deliv-ery times is extremely high for two reasons. Deliverytimes can affect whether or not a customer places anorder with the firm or with a competitor. Also, whenan existing oven has to be replaced due to wear or age,the replacement unit, in general, needs to be availablein a very short time period.

5. A new type of modularity

From the within-case and across-case analyses, weidentified four significant empirical findings, the firstof which is presented and discussed here separatelyfrom the remaining three. We do so because this firstempirical finding was not posed in our original re-search agenda. Another reason is that this empiricalfinding provides the foundation upon which we derivethe remaining three empirical findings most directlyrelevant to the two research questions.

More particularly, the first empirical finding pro-vides insights into a new type of modularity that, al-though present in practice, is not currently captured inpreviously articulated typologies. Before we presentand discuss this new type of modularity, it is impera-tive that we provide precise definitions for a numberof key terms that are incorporated into our explana-tion. These definitions, we trust, should help to avoidconfusion from terminological ambiguities (e.g. howis a “part” different from a “component”?) and to pre-vent misunderstandings regarding the unit of analysisand the level of analysis.

5.1. Definitions

In the ensuing description and discussion, we distin-guish between “components” and “product familyvariants”, considering components of a product familyto be all parts brought together in the final assemblystage, including those parts required for functiona-lity reasons and any optional add-on parts. Differentsets of components can, therefore, be combined inthe final assembly stage into different product familyvariants.


Consider, for example, a bicycle, assembled fromsuch components as the gear set, the brake system, theseat, wheels, the fork, and the handlebar. In this ex-ample, a component can be a one-piece element, suchas the fork, or an entire subassembly made of manyparts, such as the gear set. If multiple gear set vari-ants are offered, the different gear set variants makeup what we call a “component family”, with each gearset variant being referred to as a “component familyvariant”. This notion of the component family is simi-lar to that of “replaceable component set” advanced byGupta and Krishnan (1999)and that of “module type”proposed byChakravarty and Balakrishnan (2001).When the component family consists of only one op-tion (i.e. there is only one component family variant),then the component family degenerates into a “com-mon component”, meaning that the label “componentfamily” is no longer applicable.

From these definitions, we can now discriminatebetween the unit of analysis and the level of analysisin this study. Whereas the product family constitutesthe unit of analysis whose behavior we wish to de-scribe, explain, and predict for purposes of theory de-velopment, the level of analysis, or the level at whichwe observe the type of modularity embedded withinthe product family, is the components making up theproduct family architecture.

5.2. Component swapping modularity andcombinatorial modularity

With these terms and definitions, we can com-pare and contrast two of the six product families—Custom-phone and Heavy-truck—to illustrate anddefine “combinatorial modularity.”

In Custom-phone, product family variants differ pri-marily in terms of the faceplate and a few other de-tails (e.g. the aerial shape), while the PCB, back panel,battery, display, and keyboard remain identical acrossproduct family variants. In terms ofPahl and Beitz’s(1984) typology, all components essentially imple-ment basic functions. ApplyingUlrich’s (1995) ty-pology further reveals that the interfaces between thebasic product body and different component families(e.g. the external faceplate and the aerial) are differ-ent, with the product family embedding essentially slotmodularity. Finally, at the schematic level, componentvariants from a component family are swapped over

a basic product body made of a number of commoncomponents, or component swapping modularity perUlrich and Tung (1991).

Conversely, in the Heavy-truck product family,the various components (gearbox, rear axle, chassis,cabin, engine, front brakes, etc.) are all componentfamilies that are to be combined, subject to a fewtechnological constraints, to obtain product familyvariants. Two product variants within Heavy-truckmay, in fact, differ in terms of all components. LikeCustom-phone, the Heavy-truck product family com-prises of components implementing whatPahl andBeitz (1984)refer to as basic functions. ConsideringUlrich (1995), since different component familieshave different interfaces and cannot be arbitrarilyconnected to one another (e.g. the cabin cannot beconnected with the rear axle; the cabin and the enginedo not connect to the chassis in the same manner),the most appropriate type of modularity describingthe product family is slot modularity, not sectionalmodularity, and not bus modularity. But, in terms ofUlrich and Tung’s (1991)typology, none of the fourtypes of modularity apply. Hence, a type of modular-ity that would describe the product family architectureof Heavy-truck appears to be missing fromUlrichand Tung’s (1991)typology.

Comparing the product architectures betweenCustom-phone and Heavy-truck reveals that the keydifference between the two is the ratio of commoncomponents to component families. Whereas all com-ponents in Custom-phone, with a few exceptions (e.g.the external faceplate), are common components, inthe case of Heavy-truck, virtually all componentsare component families, meant that no basic productbody can be observed. This significant difference hasimplications for both product design and operationsmanagement. For product design, the presence of acommon body allows the adoption of integral (i.e.non-modular) designs for that particular portion ofthe product, with potentially positive implications onproduct performances (Ulrich and Seering, 1989). Foroperations management, the number of common com-ponents compared to component families has impor-tant implications in terms of component commonalityand, hence, for operational performance (seeSection2.1). Reflecting on this key difference, the ratio ofcommon components to component families, a typeof modularity that is the complement of component

560F.

Sa

lvad

or

et

al./Jo

urn

al

of

Op

era

tion

sM

an

agem

en

t2

0(2

00

2)

54

9–

57

5


swapping modularity can be inferred. Labeling thiscomplement as “combinatorial modularity”, we canmore formally characterize combinatorial modularityas follows.

1. All components making up a product family variantbelong to component families, meaning that eachcomponent itself is a variant.

2. Each component family interfaces with a subset ofother component families, with the interface beingstandardized by pairings of component families.The interface refers to a set of rules that constrainthow two components are to connect and to interact(Parnas, 1971; Baldwin and Clark, 2000).

3. The interface between two component families isdependent upon the specific coupling of componentfamilies, but is independent of the specific com-ponent variants selected from the two componentfamilies that need to be combined.

From this characterization, we can make three ob-servations. First, it is evident that (1) and (2) are inher-ited from the definition of slot modularity, of whichcombinatorial modularity is a special case. Second,by stating the converse of (1): “only one of the com-ponents making up a product variant is a componentfamily”, we can state a formal definition of compo-nent swapping modularity that is consistent with thatsuggested byUlrich and Tung (1991). Componentswapping modularity and combinatorial modularity,therefore, have to be considered as two extremes, di-vided by a spectrum of situations with a gradually in-creasing incidence of component families to commoncomponents—all of which fall in the general case ofslot modularity (seeFig. 1). Finally, we must realizethat combinatorial modularity represents an ideal, par-ticularly given (3). Pragmatically, technological con-straints may make it very difficult to achieve interfacestandardization across all possible pairings of inter-facing component family variants.

6. Empirical generalizations

Besides uncovering a new type of modularity,within-case and across-case analyses also generatedthree empirical insights into the relationships amongthe type of modularity, product variety, and compo-nent sourcing decisions.

6.1. Types of modularity, product variety level, andproduction volume

For Trendy-moped, Custom-phone, and Microwave,the three product families with low product varietylevel and high production volume, product variety ap-pears to be obtained mostly through component swap-ping modularity. For each case, it is possible to iden-tify: (a) a set of common components that are invari-ant across different final product configurations—thebasic product body; and (b) a set of components thatcan be swapped in order to generate product variants(seeTable 4).

When the requirements for product variety arenot too stringent, firms that deploy a batch processin the final assembly stage, such as Custom-phone,Trendy-moped, and Microwave, appear to derive twobenefits from embedding component swapping mod-ularity into their product family architectures. First,this type of modularity allows the three firms to con-centrate on economies of scale in manufacturing thebasic product body. Consider, for example, this com-ment by Custom-phone about the PCB, a key basicproduct body component.

This level two code, which is the PCB with allthe components mounted on it, is combined withmany aesthetic variants, so that it ends up beingshared by hundreds of [product] variants. The as-sembled PCB is all the same throughout all vari-ants. Given our volumes [this] allowed us to heavilyinvest in automated equipment. . . we have SMDlines capable of mounting up to 35,000 componentsper hour [and] testing lines capable of fully testing3000/4000 PCBs per day.

Likewise, at Microwave, the standardization ofthe frame justified the installation of an automatedwelding line that provided a reasonable return on in-vestment and, more importantly, did not affect ovenaesthetics and user interface.

Second, it appears that component swappingmodularity enables the three firms to postpone thegeneration of product family variants until the final as-sembly stage without negatively affecting operationalperformance. For Custom-phone, which operates on amake-to-order basis, postponement actually translatesinto shorter delivery times, while at Trendy-mopedand Microwave, which operate on a make-to-stock

562F.

Sa

lvad

or

et

al./Jo

urn

al

of

Op

era

tion

sM

an

agem

en

t2

0(2

00

2)

54

9–

57

5


basis, postponement translates into decreased inven-tory risks and/or shorter average delivery times. Asone Trendy-moped manager said:

It is not by chance that the factory performs only as-sembly, it is our very approach saying: let’s shift allproduct customization at the last ring of the chain,. . . giving ourselves an asset that allows us to as-semble as near as possible to the market. . . mini-mizing risk of understock or unsold final inventory.

In summary, final assemblers deploying batch pro-cessing and facing moderate levels of product varietycan benefit most readily from embedding componentswapping modularity into their product family archi-tecture. Component swapping modularity, to be pre-cise, appears to provide these firms with the specificmeans to extend the capability of what is basically amass production system and make it compatible withthe requirement for moderate levels of product variety,without serious penalties to productivity.

Conversely, for the remaining three product fami-lies with high product variety level and low productionvolume (i.e. Heavy-truck, Multiplexer, and Techoven),product variety appears to be achieved by embed-ding combinatorial modularity into the product fam-ily architectures. In each case, the product family isstructured fundamentally as combinations of compo-nent families, with different paired component fami-lies having different interfaces, and with the interfacesbetween paired component families tending to remainstable.

The case of Multiplexer deserves additional expli-cation since one might think, from a strictly technicalpoint of view, that the embedded type of modularity isbus modularity, since the connection between the var-ious groups of components and other external devicesis effected through a bus. However, careful evaluationof how product family variants are generated revealsthat different choices available from the key compo-nent families (aggregates and tributaries) may actuallyrequire different variants of the back panel component.Moreover, since the back panel provides connectionbetween the machine and other telecommunicationsequipment, different requirements in terms of suchconnections may also necessitate different versionsof the back panel. Therefore, because the individualback panel variant cannot accept any combination ofcomponent families, and the back panel itself may

be chosen in multiple variants based on customers’needs, combinatorial modularity, not bus modularity,is a more accurate description of the type of modu-larity embedded in the Multiplexer product family.

For all three firms, the decision to embed combina-torial modularity into the product family architectureis primarily driven by the need to offer customers amuch larger set of choices than the number of prod-uct family variants made available when only a fewcomponents are swappable. To satisfy this need, allthree firms sacrificed the opportunity to standardizeand maintain a common product body in the prod-uct family architecture. Instead, they maximize thenumber of choices by defining component familiesthat can vary according to customer requirements. Atthe same time, so as to minimize the increase in thenumber and levels of inventoried items, especiallywhen the combining of different components duringassembly also requires an interfacing component tochange as well, all three firms chose to standardizeas much as possible the different interfaces betweenpairs of component families that have to be matchedin final assembly. Standardizing the interfaces amongcomponents does not, however, mean that all the in-terfaces between all pairs of component families arestandardized to a single interface, for a standardizedinterface that would work for all possible compo-nent family pairings would be a property of sectionalmodularity.

A manager at Techoven, for example, pointed outthat standardizing the interface between pairs of com-ponent families results in lower inventory levels visa vis a non-modular product with the same level ofproduct variety.

Differences between oven doors for different brandsare [basically] restricted to the oven handle. . .

At one time, different handle variants had differ-ent sockets [connecting them to the locking mech-anism], so that when we had to assemble a certaindoor variant, we needed the specific locking mech-anism interface that could slot that specific handle. . . This meant that we had to carry more line inven-tories of different locking mechanisms—door han-dle interfaces. . . Then we looked at our basic doorlocking mechanism, and we imposed that the han-dle socket be standardized for all handle variants.Now all the brands have the same locking mech-


anism and defining the front door brand variant isdependent [for a given height] only on the handle. . . The basic idea is to have standardized, plug-gable interfaces among the different components.

Similarly, at Heavy-truck, in the past, as differentgear variants were selected, the interfaces for controlsignals between gear, engine, and cabin had to change.This led to assembly problems, since the manner inwhich the different components had to be connected byoperators changed with the specific component vari-ants changed. Instead, by introducing a unique circuitconnecting electronic control units that are scatteredthroughout the various vehicle components alleviatesassembly problems and quickens the final assemblyoperations.

Hence, when the requirements for product varietyare high, combinatorial modularity appears to allowthe three firms in our study to reduce the necessity ofcarrying a larger number and a higher level of inven-tories via interface standardization between pairs ofcomponent families. In doing so, these firms are ableto move the processing step responsible for the prod-uct family variants upstream from the final assemblystage to the preceding stage, where component familyvariants are manufactured—essentially, doing the op-posite of what postponement suggests. Furthermore,in all three firms, combinatorial modularity enablesthem to economically justify and implement, for thefinal assembly stage, a mix-model production systeminstead of a batch system that, in turn, reduces finalassembly time and provided faster response to hetero-geneous customer requirements. For example, a Mul-tiplexer manager stated:

We have a very, very high variety of final prod-uct configurations [more than 10,000 product vari-ants], which we obtain by combining certain typesof PCBs [approximately 1000 component familyvariants], each one performing a certain function. . . we do this because we want to cut the lead-timethe customer sees. . .

In conclusion, when we compare and contrastTrendy-moped, Custom-phone, and Microwave toHeavy-truck, Multiplexer, and Techoven, we can ob-serve the following empirical generalization about therelationship between the types of modularity and thevariety:volume ratio.

Empirical generalization 1: When product varietylevel is low and production volume is high, the ap-propriate type of modularity is component swappingmodularity, whereas when product variety level is highand production volume is low, then the appropriatetype of modularity is combinatorial modularity.

6.2. Types of modularity and component sourcing

The within-case and across-case analyses also ledto the identification of two empirical generalizationsrelating the types of modularity to component sourc-ing or the external allocation of manufacturing tasksto either firm-owned or independent supplying enti-ties. Since all six firms are basically final product as-semblers, many, if not all, component manufacturingactivities are performed by external suppliers that areeither sister units belonging to the same firm, or plantsowned by other firms. What appears to be novel, how-ever, are the complex interactions between the typesof modularity, i.e. component swapping modularity orcombinatorial modularity, and the way firms manageto reduce the negative impact of component variety onmanufacturing and sourcing performance.

For Custom-phone, Trendy-moped, and Microwave,we can make two common observations that are sum-marized inTable 5. First, all three firms allocated themanufacturing for swappable components to suppli-ers who tended to be located in closer proximity tothe final assembly facility than those suppliers man-ufacturing the basic product body components. Sec-ond, all three firms were able to exert more pressureand stronger influence over the swappable componentfamily suppliers than over the basic product body com-ponent suppliers. For example, consider the followingquote from Custom-phone:

In order to customize the cell-phone body in termsof color and decals, and for product packaging, werely on suppliers that work with us in real time,with delivery times of 15 and, sometimes, 7 days.We know them [i.e. cell-phone body suppliers] welland we control them well. . . our agreement withthem is that they have to produce from time to timebased on what we need, even working through thenight . . . and, you know, this is a good area for plas-tic molding and painting [i.e. there are many poten-tial suppliers nearby]. Their proximity is a logistic


Table 5Component swapping modularity, component sourcing, and operational performance

Trendy-moped Custom-phone Microwave

Main swappablecomponents families

External body Front body Door and control panel

Supplier proximity Near (nearby region) Near (nearby region) Near (same region)Ability of the final assembler

to exert control overthe supplier

High (suppliers owned by thefinal assembler)

High (suppliers easilysubstitutable and very smallcompared with the finalassembler)

High (suppliers small comparedto the final assembler; suppliersabsorb a significant portion offirm production)

Impact on operationalperformance

Very short sourcing lead-timesfor swappable componentsvariants

Very short sourcing lead-timesfor custom swappablecomponents

Very short sourcing lead-timesfor swappable componentvariants

advantage, as we can save time and money in trans-portation. . .

By locating swappable component suppliersnearby and over whom they can influence strongly,Custom-phone, Trendy-moped, and Microwave wereable to reduce the suppliers’ response and deliverytimes, while postponing the generation of productvariants until the final assembly stage—a benefitechoed by Trendy-moped:

Saying that we [the final assembler] have great flex-ibility [in generating product variants], but then thatour suppliers are not flexible, is nonsense, becausethis would mean that my input inventory wouldskyrocket. Therefore, our job is to ensure that. . .

all suppliers responsible for components generat-ing customization. . . are [located] as near as pos-sible to the final assembly facility in terms of time.[If suppliers are not near, and] we were to keepthese components in stock. . . [this] would gener-ate a foolish proliferation of codes, such as paintedplastics. Because we need extremely short sourcinglead-time,. . . we require suppliers to react with a1-week delivery time.

Similarly, in the case of Custom-phone, the abilityof the supplier of the cell-phone external body (i.e. theswappable component) to respond quickly enabled thefirm to contemporarily issue purchase orders and makeorders to the cell-phone external body component sup-pliers and to the internal automated PCB assemblyoperations, respectively. This ability to do so, in turn,made it possible for both the cell-phone external bodyand the PCB to be available quickly for final assembly.At Microwave, since the frame is highly standardized

and is manufactured by a highly automated process,it is critical for suppliers of aesthetics-related parts tokeep pace with this frame manufacturing process anddeliver with short lead-times.

Based on this discussion, we can, therefore, de-rive the following empirical generalization aboutcomponent swapping modularity and componentsourcing.

Empirical generalization 2: Firms that choosecomponent swapping modularity limit the negativeimplications of product variety on operational per-formance by relying on component family supplierslocated near their final assembly facilities and whichtend to be smaller or directly controlled by the finalassembler.

On the contrary, when we consider Heavy-truck,Techoven, and Multiplexer, the three product familiesembedding combinatorial modularity, a number of ob-servations about component sourcing decisions, dif-ferent from those for Custom-phone, Microwave, andTrendy-moped emerges (seeTable 6). First, with com-binatorial modularity, in principle, the larger the num-ber of different component families and the larger thenumber of different variants within each componentfamily, the larger the number of different final prod-uct configurations that can be obtained. Yet, all threefirms strive to keep, as low as possible, the number ofdifferent component families to be allocated to exter-nal supplying entities. Techoven, in fact, is in the pro-cess of shrinking the number of different componentfamilies and says:

Today, just to give you an idea, our ovens are builtfrom about 400 component [families], without


Table 6Combinatorial modularity, component sourcing, and operational performance

Heavy-truck Multiplexer Techoven

No. of componentfamilies

Low (a relatively low number ofcomponent families is gatheredfor final assembly)

Low (a relatively low number ofcomponent families is gatheredfor final assembly)

Decreasing (the firm is trying toreduce from 400 to 50 thenumber of components that haveto be brought together in finalassembly)

Component familymodularization

Yes (cabin, engine, transmission,axles all embed some modulardesign concepts)

Yes (component families aredesigned according with thehourglass concept)

Increasing (some componentfamilies, as the steam generationmodule, embed componentswapping modularity)

Relationship with componentfamilies suppliers

Intense (multiple year purchaseagreements involving alsoproduct development issues)

Intense (suppliers are part of thesame group, but they also workas independent contractors andare heavily involved in productdevelopment)

Developing (the firm is shiftingfrom a arm’s length approach toa relation-building approach)

Impact on operationalperformance

Relatively short componentdelivery times, simpler internalflows in final assembly plant

Relatively short componentdelivery times, provided thatelementary electroniccomponents are available

Relatively short componentdelivery times, simpler internalflows in final assembly plant

counting nuts, bolts, and screws. . . The future onwhich we are working. . . that will [still] be a futureof strong product modularity [and] is a situation inwhich these 400 component families will be shrunkto about 50 component families. . . [Techoven].

At Multiplexer, the relatively large number of fi-nal product configurations (>1000) are essentially as-sembled from just 10 component families. Likewise,for Heavy-truck, only a few component families—thecab, transmission, pre-assembled chassis, etc.—are as-sembled together to create approximately 2500 finalproduct variants.

Of course, reducing the component families countin a given product family inevitably increases the av-erage number of sub-components that make up eachcomponent family. In other words, for a given prod-uct family variant, putting together a smaller numberof component families for final assembly means thatthese component families, themselves, are more com-plex. Again, Techoven illuminates why it should seekto reduce the number of component families and howthis pursuit affects operational performance.

If I make [the component families] more complex,then the final assembly lead-time is drastically cut. . . But if the component [family] is made insidethe plant then the overall lead-time is higher. . .

The idea works if my module comes from outsidethe plant. [Otherwise] we are wasting resources ininventory, resources in terms of money and people,in optimizing inventory management, in optimizingmaterial flows inside the factories. Obviously whenwe go from 400 to 50 component families, then wehave great simplification [in final assembly]. It is amatter of flexibility, as we will obtain with a smallnumber of modular components all the product vari-ants we need. . . The front door is a perfect mod-ule, as it is complex, heavy and expensive. It is amodule because if a supplier arrives and delivers itto me, I just have to install it and the game is done![Techoven].

Likewise, by reducing the count of component fam-ilies, Multiplexer was able to simplify final assemblyoperations. At Heavy-truck, the same reasoning leadsto the following comment.

If we took upon us all the complexity deriving fromthe variability of components in a heavy truck vehi-cle, then we will die from this complexity. Hence,we made a drastic decision. . . since we are notable to manage this complexity well, we have todelegate it [Heavy-truck].

Therefore, while there would be simplification forthe final assembler, it appears that the component


family suppliers would be subjected to greater oper-ational complexity, higher costs, and lower deliveryperformance. In fact, one could argue that by allocat-ing the manufacturing task for component families tosuppliers, the three final assemblers in our study havesimply shifted the product variety–operational per-formance trade-off up the supply chain. As multiplemanagers at these firms have pointed out, the poten-tial transfer of the trade-off from the final assemblersto the component family suppliers can be overcomethrough the pursuit of modularity in the componentfamilies that are allocated to suppliers.

The groups we source from feeder plants are madeby combining a certain set of common intermediateobjects into a wider set of end items [components],. . . according to the classical hourglass concept.Then we expand these end items [components] intofinal product configuration [Multiplexer].Another important module we consider for out-sourcing is the steam generator, the heating elementconnected to the oven cavity. . . it is a case filledwith water with resistors placed inside it. . . IfI sell it in the USA it comes with a 208 V resis-tor, in Europe with a 400 V resistor, and so on,depending on the country. . . simpler it is impos-sible! . . . All the other features are standardized[Techoven].The entire heavy vehicles engines’ product rangehas been designed so that they are modularized andmanufactured in a single plant. Even engines withdifferent displacements share the same concept, andhence they can be made with the same machines. . . They also share many components. . . can becustomized by changing certain components as thecompressor: no compressor, low compression rate,high compression rate, variable geometry blades,etc. [Heavy-truck].

Finally, we can observe that the allocation of themanufacturing tasks for component families alongwith the requisite necessity of seeking modularity inthese component families moves the nature of therelationship with the supplier away from one that ischaracterized by unilateral decisions, as we observedin the cases of Trendy-moped, Custom-phone, andMicrowave, towards a more collaborative, bilateralrelationships. This quote from Heavy-truck reiteratesthis observation:

We wanted to equip our product line with an op-tional automated gear. . . The electronics control-ling the gear must consider the behavior of manysystems in the truck. . . the braking system, thepneumatic suspension system, the engine controlsystem, etc.. . . , all of which can come in multiplevariants. In developing the automated gearbox, oursupplier told us “we cannot do it by ourselves, weare not able to consider all these variables”. . . sowe formed a joint venture [Heavy-truck].

The formal joint venture arrangement betweenHeavy-truck and the automated gearbox supplier fa-cilitated the definition of a standard electronic gearmanagement system that works as a standard in-terface between the automated gear and the othervarying subsystems within the truck. In this partic-ular incident, the interdependencies between truckfeatures and specific gear features were minimizedto just the gear torque, which is unavoidably relatedto the engine power that is desired. All other gearvariants (e.g. the external case material, the integratedhydraulic retarder, the number of gears, etc.) de-pended exclusively on specific, customer-defined gearfeatures.

In summary, for combinatorial modularity, the fol-lowing empirical generalization appears to be sup-ported by the case analyses.

Empirical generalization 3: Firms that choosecombinatorial modularity limit the negative implica-tions of product variety on operational performanceby reducing the overall number of component fami-lies defining the final product architecture, by work-ing with suppliers to modularize the respectivelyallocated component families, and by setting up bi-lateral relationships with suppliers of componentfamilies.

7. Theoretical development

In this section, we build upon the insights from theempirical findings with the purpose of deriving whyand under what conditions the three empirical general-izations might hold for a product family outside of theoriginal sample. In doing so, we state the boundariesof our theory development, identify and define, whereneeded, relevant constructs that inform our theoretical


development, and provide a rationale for the proposedtheoretical relationships among these constructs.

7.1. Boundary conditions

For the purpose of theory development, we delimitthe scope of the propositions to discrete manufactur-ing firms of product families: (i) whose productionvolumes are large enough to justify their manufac-ture on a dedicated assembly line; (ii) whose productfamily architectures can generally embed slot modu-larity; and (iii) whose constituent components are ex-ternally sourced so as to allow firms to focus theirinternal operations on final assembly. These bound-ary conditions are consistent not only with the speci-fications of the reference population and sample, butalso with the unit of analysis and the level of analysisthat guided the within-case and across-case analyses.Furthermore, these boundary conditions exclude firmswith discrete product families that have very low pro-duction volumes and the need to consider interactionsbetween multiple product families made on the samefinal assembly process.

7.2. Types of modularity, product variety level, andproduction volume

Recall, first of all, from our discussion inSection 1that product variety generally has a negative impact onoperational performance. Second, recall that empiricalgeneralization 1 inSection 6suggests a relationshipbetween types of modularity, product variety level,production volume, and operational performance.Third, recall, fromSection 5, that slot modularity canbe more precisely described as a spectrum anchoredby the two extreme cases of component swappingmodularity and combinatorial modularity (Fig. 1),with the spectrum being defined by the ratio of com-mon components to component families. Synthesizingthese three insights suggest that the appropriate po-sition of a product family within the slot modularityspectrum (more descriptively, the component swap-ping modularity–combinatorial modularity spectrum)that would minimize the negative impact of productvariety on operational performance is impacted by thetwo manufacturing variables of product variety leveland production volume.

7.2.1. Product variety levelTo understand how and why product variety level

impacts the appropriate positioning of a product fam-ily within the slot modularity spectrum, we need tounderstand two interrelated points. First, consider thata product (e.g. an automobile) can generally be de-scribed as a vector of attributes, including exteriorcolor, engine power, interior color, safety devices, etc.(Lancaster, 1971; Dobson and Kalish, 1988; Ramanand Chhajed, 1995). Deciding to offer a particular carmodel, say Honda Accord, in multiple variants wouldinvolve two decisions: (a) what are the attributes thatcustomers are allowed to express choices (e.g. the cus-tomer can choose the exterior color, but not the en-gine size)? and (b) what are levels of these attributes(e.g. exterior color: four choices of blue, red, silver,or white)?

In making these decision, economic consumer the-ory informs us of a phenomenon known as “satiation”(Lancaster, 1979, pp. 147–149), i.e. as the numberof levels offered for a given attribute increases, themarginal increase in the consumer utility decreases.For example, consider the extreme scenario whereina car manufacturer increases the number of color op-tions over which the customer can choose from zero(i.e. only color is available) to five to 1000 colors.The customer, consequently, experiences greater util-ity up to a point beyond which there would not be anysignificant impact on consumer utility. Extending thisphenomenon to the current context suggests that of-fering 1000 product family variants based on a singleattribute (scenario A: for example, 1000 cars differ-entiated by exterior color) may not generate for theconsumer the same level of utility as offering 1000product family variants based on combining choicesamong multiple attributes (scenario B: for example,1000 different cars differentiated by choosing from 10exterior colors, four engines, five interior colors, andfive stereo brands). In fact, scenario B, in this case,results in higher consumer utility than scenario A.More generally, we can, therefore, state that for anygiven level of consumer utility, the product varietylevel is positively and directly related to the numberof product attributes allowed to vary (point 1). Hence,if customers were demanding greater product variety,it would be better, from a consumer utility perspec-tive, to allow more attributes to vary than to increasethe number of levels within a few attributes.


Second, by definition of modular product archi-tecture, there is a one-to-one mapping of productfunctions, which are the engineering equivalents ofproduct attributes, onto components. As such, with re-spect to slot modularity, consistent with boundary con-dition (ii), changes to the number of levels of a productattribute would impact only a single, individual com-ponent (Ulrich, 1995). More specifically, this impliesthat the number of levels of a product attribute is di-rectly tied to the number of component family variants,while the number of product attributes allowed to varyis directly tied to the number of component families(point 2). Realize, therefore, that by relaxing bound-ary condition (ii) and considering product familieswhose architecture can generally embed not only slotmodularity, but also sectional modularity, then prod-uct attributes may not necessarily be strictly related tocomponents, but may, instead, be related to the waythey are connected. For example, consider three sofasthat are modular, connected via two corner pieces toform a U-shape or an S-shape depending on customerpreference. In this case, the “sofa layout” product at-tribute corresponds not to components that vary, but tothe way the components, or sofa pieces, are connected.

When we consider points 1 and 2 together, we can,therefore, infer that for any given level of consumerutility, the product variety level is positively and di-rectly related to the number of component families(which as defined comprises multiple component vari-ants). Hence, when demands for product variety arenot excessive, many components can be fixed to definea common body while a few become component fam-ilies, with the appropriate type of modularity leaningmore towards component swapping modularity thancombinatorial modularity. However, as customers de-mand greater and greater product variety, most, if notall, components should become component familiesand, as such, the appropriate type of modularity ap-proaches combinatorial modularity.

By directly translating the demand for product va-riety into the ratio of component families to commoncomponents (which, de facto, specifies the positionof the product family along the component swappingmodularity–combinatorial modularity spectrum), slotmodularity allows the exploitation of the satiationphenomenon to maximize internal operational perfor-mance, essentially by minimizing the negative impactof providing product variety on operational perfor-

mance. For both component swapping modularityand combinatorial modularity, this minimization isachieved essentially by limiting variety at the compo-nent level to only what is strictly required to match agiven level of consumer utility, so as to derive max-imal benefits from component commonality. In thecase of component swapping modularity, economiesof scale derive maximal benefits from the commonbody, while focusing attention on how to mitigatethe negative operational impact of a relatively fewswappable components. In the case of combinatorialmodularity, the minimization of the negative impactof product variety on operational performance derivesfrom the fact that we can achieve the same level ofproduct variety (e.g. 1000 car configurations) by hav-ing to manage for fewer variants within componentfamilies. Furthermore, because product attributes mapdirectly onto specific components whose variationsdo not propagate to other components, the complexityand costs of recovering from incorrectly specifyingmarket demand for a product attribute are reduced.

7.2.2. Production volumeProduction volume, like product variety level, also

impacts the positioning of a product family along theslot modularity spectrum and, hence, the appropriatetype of modularity to embed within the product fam-ily architecture. The explanation as to how and whyis relatively straightforward. When the product fam-ily production volume is high, there is a greater in-centive to pursue component commonality, because ofthe potentially high magnitude of scale economies. Infact, this incentive further supports the specification ofa common product body, consistent with componentswapping modularity. At the same time, since productfamily variants can be generated simply by swappinga few components on a mass-produced product body,component swapping modularity becomes an effec-tive means to allow repetitive, high volume productionsystems to basically deliver some product variety.

On the contrary, when the production volume of aproduct family tends to be low relative to the numberof product family variants available within the sameproduct family, the relative advantage of sharing acommon product body throughout a product fam-ily tends to decrease. The benefits from economiesof scale production of a standardized product body,consistent with component swapping modularity,


consequently diminish. As such, the differential dis-advantage in terms of economies of scale from pur-suing combinatorial modularity instead of componentswapping modularity becomes reduced and, perhaps,trivialized.

7.2.3. Conceptual synthesisIntegrating the above arguments concerning the ap-

propriate position of a product family within the slotmodularity spectrum, in light of product variety level,production volume, and consequences for operationalperformance, we can advance the following proposi-tion.

Proposition 1. As the product variety level increases(decreases) and/or production volume decreases(in-creases), the type of modularity that maximizes opera-tional performance in building product family variantsmoves away from component swapping modularity(combinatorial modularity) and towards combinato-rial modularity (component swapping modularity).

7.3. Types of modularity, component familycomplexity, and geographical proximity (to suppliers)

Recall, firstly, from the earlier discussion inSection1, that product variety tends to increase componentfamily variety (Fisher et al., 1999), affecting the per-formance of not only the final assembler, but also com-ponent family suppliers as well (Krishnan and Gupta,2001). Recall also that empirical generalizations 2 and3 in Section 6, in essence, suggest how a final assem-bler, given the positioning of its product family withinthe slot modularity spectrum, can mitigate the negativeimpact of component family variety on componentsourcing performance and, consequently, its opera-tional performance, via component sourcing decisionsrelating to two key variables—component familycomplexity and geographical proximity to suppliers.

Before we develop the theoretical rationale forwhy these two variables can mitigate the productvariety–performance trade-off, given the position ofthe product family within the slot modularity spec-trum, two additional points need to be clarified. First,we have to realize that the final assembler’s objectivein pursuing product variety is to deliver product fam-ily variants as quickly as possible and as efficiently as

possible. In light of boundary condition (iii), one wayto achieve this objective is to maintain high inventorylevels of the component family variants. However,doing so would significantly increase manufacturingcosts. So, in order to avoid increasing manufacturingcosts in pursuing product variety, the final assemblermust strive to maintain minimal inventory levels ofcomponent family variants without negatively im-pacting the ability to deliver product family variantsquickly (point A).

Second, the positioning of a product family withinthe slot modularity spectrum defines, a priori, the por-tion of the supply chain that would be directly affectedby the demands for product variety (point B). In theextreme case of component swapping modularity, onlyone swappable component supplier would be directlyaffected, whereas, in the extreme case of combinato-rial modularity, all component family suppliers wouldbe directly affected.

7.3.1. Component family complexityThe first variable, component family complexity,

intends to capture not only the number of compo-nent family variants within a component family, butalso the number of parts making up each componentfamily variant, on average, and the innovativeness ofthe component family technology. The importance ofcomponent family complexity in mitigating the nega-tive impact of component family variety on componentsourcing performance and, consequently, operationalperformance can be considered at the two extremes ofthe slot modularity spectrum.

7.3.1.1. The case of component swapping modularity.At the end anchored by component swapping modu-larity, the product family architecture is divided, bydefinition, into common body components and swap-pable components. For component body components,the objective of keeping inventory levels low, whileensuring quick delivery from common body compo-nent suppliers (as stated in point A) can be read-ily attained via long-term purchasing agreements thatspecify component pricing, delivery schedule, etc. Forswappable components, such an approach is riskier,since it is difficult, if not impossible, to specify in ad-vance not only what specific component variants arerequired, but also how many and when. Hence, in thecase of swappable components, the ability of these


suppliers to react quickly and deliver with short no-tice becomes of critical importance (implied by pointB), since being able to do so necessarily impacts theability of the final assembler to deliver product familyvariants quickly without carrying high inventory lev-els of various swappable components.

To enable and facilitate quick response from swap-pable component suppliers, low component familycomplexity is a key factor for three reasons. First,it is intuitively obvious that the speed at which aswappable component variant can be made is a directfunction of the number of parts defining the compo-nent. Second, the less innovative the technology em-bedded into the swappable component and the fewerthe number of parts, the greater the likelihood of find-ing small, substitutable suppliers with relatively lowbargaining power vis a vis the final assembler andover whom the final assembler can demand shorterdelivery times without paying a premium. Third, andmore generally, the fewer the number of parts in aswappable component, the less complex the interfaceis likely to be between the swappable component andthe common product body, which, as a result, re-duces the risk of having a complex coupling requiringexcessive processing time during final assembly.

7.3.1.2. The case of combinatorial modularity.At theopposite end of the slot modularity spectrum (i.e. com-binatorial modularity), all component families, by def-inition, are allowed to vary while the interface betweenspecific pairs of component families is standardized.In this case, all suppliers are directly affected by thepursuit of product variety (point B) and must be takeninto consideration to help the final assembler attain itsobjective of delivery product variety quickly and effi-ciently.

Component family complexity remains a criticalfactor and keeping component family complexity lowremains theoretically sound for the same reasons asprovided in the case of component swapping modu-larity. However, we can provide three arguments as towhy high rather than low component family complex-ity is preferred when the position of a product familywithin the slot modularity spectrum approaches thatof combinatorial modularity.

First, the benefits of interface simplification andquick delivery, when component family complexityis low, may be more than offset by the operational

challenges that a final assembler faces in the case ofcombinatorial modularity. In particular, when com-ponent family complexity is low, more componentfamilies needs to be combined during final assem-bly than when component family complexity is high.Compare, for example, a situation in which we definea product by 100 component families that have to becombined during final assembly to another situationin which we now define the same product as only 10component families that have to be combined duringfinal assembly. By increasing component family com-plexity, the final assembler gains from reducing thecomplexity of materials flow within the final assemblyprocess, the number and complexity of connectionsto be performed during final assembly, the overheadcosts of manufacturing planning and control, etc.However, given boundary condition (iii), these gainscome at the expense of the component family suppli-ers, since product variety is effectively translated intohigher component variety.

Second, the benefit of shorter component deliverytimes from low component family complexity mayalso be offset by the degradation of the delivery re-liability of the overall supply chain.Ceteris paribus,component variety lowers component delivery reli-ability, or the ability of the component supplier todeliver on time, largely because of the complexityinduced within supplier operations by component va-riety (Fisher et al., 1999). Otherwise, the componentfamily supplier is forced to increase its componentfamily variants inventory in order to ensure deliveryreliability. In the case of combinatorial modularity,since all component family suppliers have to delivermultiple component family variants, and since overallsupply chain delivery reliability, assuming indepen-dence across suppliers, is the product of all individualsuppliers’ delivery reliabilities, the more componentfamily suppliers (i.e. low component family com-plexity) there are, the worse off the overall supplychain delivery reliability. Stated in terms of compo-nent family complexity, there is direct relationshipbetween component family complexity and overallsupply chain delivery reliability.

However, high component family complexity nec-essarily implies, for a given product variety level, ahigher average number of component variants percomponent family. This, in turn, has the potential toaffect component delivery reliability from specific


suppliers. Consequently, in order to mitigate this po-tential disadvantage, the complexity associated withthe manufacture of component family variants needsto be reduced. Modularity, this time at the compo-nent level rather than at the product family level,offers a way to do so. Furthermore, when the com-ponent family is not an off-the-shelf item, but isspecifically designed to fit within the final productarchitecture, the necessity of embedding modularityinto the component family would naturally encouragegreater collaboration between the final assembler andthe component family supplier, particularly duringproduct development. As a result, the nature of therelationship between the final assembler and compo-nent family suppliers is also affected by componentfamily complexity.

Third, recall that in the case of component swappingmodularity for which component family complexityis low, the less innovative the technology embeddedinto component families outsourced to suppliers themore favorable the operational performance resultswould be for the final assembler. But, when innova-tive technology that might affect component designbecomes available, the final assembler, in order toavoid shifting bargaining power towards the specificcomponent family supplier, can choose to standardizethis technology in such a way that would anticipatevarying customer requirements and make the compo-nent part of the common product body. This allows thefinal assembler to also maintain low component fam-ily complexity consistent with component swappingmodularity. However, as we move towards the caseof combinatorial modularity, component family com-plexity, in terms of the innovativeness of componenttechnology, is, by definition and on average, highersince an increasing portion of product components arecomponent families. In fact, the fewer the number ofcomponent families, the more difficult it becomes tonot have technologically complex component families.

7.3.2. Geographical proximity (to suppliers)The second variable, geographical proximity, refers

to the physical distance (and, hence, logistical dis-tance) between component suppliers and the final as-sembler. Generally speaking, geographical proximity(to suppliers) is always desirable in terms of compo-nent sourcing since it allows for the opportunity toreduce sourcing lead-times. In fact, since component

variety has negative effects on supplier delivery per-formance, geographical proximity may provide themeans to counterbalance this operational disadvan-tage for final assemblers. More specifically, for a finalassembler that responds to product variety require-ments in a purchase-to-order fashion, geographicalproximity reduces delivery times of components fromsuppliers to the final assembler. Moreover, for a fi-nal assembler that purchases components based uponforecasts of product variety requirements, the shortersourcing lead-times for component families reducesuncertainty in materials planning and, subsequently,lower component inventory risks.

Like component family complexity, the importanceof geographical proximity in mitigating the negativeimpact of component family variety on componentsourcing performance and, consequently, operationalperformance can also be considered at the two ex-tremes of the slot modularity spectrum. In the caseof component swapping modularity, the geographicalproximity of the suppliers of a few swappable compo-nents translates directly into one of the two advantagesalready mentioned.

In the case of combinatorial modularity, the bene-fits of geographical proximity should generally hold aswell. However, geographical proximity is not of equalcritical concern across all component family suppliers,but is of primary concern for suppliers of componentfamilies with relatively longer throughput times. Thesesuppliers are essentially external bottlenecks that con-strain the overall delivery performance of the final as-sembler and force the final assembler to lengthen thematerials planning time horizon. As a result, placingthem as close as possible to the final assembler canovercome some of these operational disadvantages.

7.3.3. Conceptual synthesisIntegrating the above arguments about component

family complexity and geographical proximity, in lightof the positioning of the product family with the slotmodularity spectrum and the consequences for oper-ational performance, we can advance the followingproposition.

Proposition 2. As the type of modularity embedded inthe product family architecture moves away from com-ponent swapping modularity towards combinatorialmodularity(or conversely, from combinatorial modu-


larity towards component swapping modularity), theextent to which the negative effect of component familyvariety on operational performances can be mitigateddepends upon:

A. The extent to which the component family com-plexity of component families can be increased(re-duced).

B. The extent to which the component family supplierwith the longest throughput time can be located ingeographical proximity relative to other suppliersof component families.

(The extent to which all suppliers of componentfamilies can be located in geographical proximityrelative to the suppliers of common components tothe final assembler.)

8. Conclusion

The issue of the interdependence among productdesign, process design, and supply chain design hasbeen recognized and brought to the attention of schol-ars and managers as early asHoekstra and Romme(1992). Since then, this issue has become prominentlydiscussed inFine (1998). Despite the undeniable ap-peal and importance of this issue to both science andpractice, we know very little about how decisions inproduct design, process design, and supply chain de-sign should be coordinated to maximize operationaland supply chain performance.

To further our understanding of the interdependenceamong product, process, and supply chain design, thepresent research explores the implications of modu-larity in terms of such manufacturing characteristicsas the level of product variety and the production vol-ume of the final assembly process. Furthermore, thepresent research ties the decision to embed a specifictype of modularity into the product family architectureto component sourcing decisions that relate to sup-plier selection and supplier location, with consequentimplications for buyer–supplier relationships.

The empirical findings and theoretical developmentpresented here, while enhancing our understanding ofthe intricacies of coordinating product, process, andsupply chain design, suggest several promising re-search opportunities that can be pursued. First, futureresearch should seek to test the theoretical propositionsposed in this paper via case-based and/or survey-based

research designs. A second possibility is for future re-search to consider the issue of multiple product fam-ilies, which essentially relaxes one of the boundaryconditions and allows enrichment of current theoret-ical propositions and/or generation of new theoreti-cal propositions. Yet, a third research opportunity isto integrate the results here with more formal opera-tions research perspectives (e.g. the product line de-sign problem (Yano and Dobson, 1999)). A fourth andfinal research opportunity is to consider the implica-tions of the theoretical propositions in the context ofother operations management prescriptions (e.g. JIT,Lean Manufacturing, TQM, etc.), as well from otherfunctional and theoretical perspectives (e.g. organiza-tional theory, strategic management, etc.).

Pragmatically, we are reminded that research in pro-fessional schools needs to provide practical theory thatwould not only advance science, but would also informthe practice of the profession (van de Ven, 1989). Thetheoretical results reported here should provide guid-ance to managers facing the need to coordinate prod-uct, process, and supply chain decisions. In fact, suchadvice could be portrayed as a roadmap for managersto traverse as they navigate through decisions aboutwhat type of modularity would be most appropriate,what components to outsource and to which suppli-ers, and how should the relationship with suppliers bestructured.

Acknowledgements

We thank the College of Business at Arizona StateUniversity and the Fondazione Gini for the generousfinancial support that made this research possible. Wealso thank the firms and the numerous managers whodevoted their time and experience to our research. Weacknowledge and appreciate the insights provided byour colleagues, particularly Manfredi Manfrin. Finally,we thank Jack Meredith and three anonymous refereesfor their insightful comments to earlier drafts of thispaper.

References

Anderson, S.W., 1995. Measuring the impact of product mixheterogeneity on manufacturing overhead cost. AccountingReview 70 (3), 363–387.


Baker, K.R., Magazine, M.J., Nuttle, H.L.W., 1986. The effectof commonality on safety stock in a simple inventory model.Management Science 32 (8), 982–988.

Baldwin, C.Y., Clark, K.B., 2000. Design Rules. Harvard BusinessSchool Press, Cambridge, MA.

Child, P., Diedrichs, R., Sanders, F., Wisniowski, S., 1991. SMRforum: the management of complexity. Sloan ManagementReview 33 (1), 73–80.

Chakravarty, A.K., Balakrishnan, N., 2001. Achieving productvariety through optimal choice of module variations. IIETransactions 33 (7), 587–598.

Collier, D.A., 1981. The measurement and operating benefits ofcomponent part commonality. Decision Sciences 12 (1), 85–96.

Denzin, N.K., 1978. The Research Act: A Theoretical Introductionto Sociological Methods, McGraw-Hill, New York.

Dobson, G., Kalish, S., 1988. Positioning and pricing of a productline: formulation and heuristics. Marketing Science 7 (2), 107–125.

Dubin, R., 1969. Theory building. Collier McMillian, The FreePress, New York.

Duray, R., Ward, P.T., Milligan, G.W., Berry, W.L., 2000.Approaches to mass customization: configurations and empiricalvalidation. Journal of Operations Management 18 (6), 605–625.

Eisenhardt, K.M., 1989. Building theories from case research.Academy of Management Review 14 (4), 532–550.

Erens, F., Verhulst, K., 1997. Architectures for product families.Computers in Industry 33 (2-3), 165–178.

Erixon, G., 1996. Design for modularity. In: Huang, G.Q. (Ed.),Design for X: Concurrent Engineering Imperatives. Chapman& Hall, New York, pp. 356–379.

Ernst, R., Kamrad, B., 2000. Evaluation of supply chain structuresthrough modularization and postponement. European Journalof Operational Research 124 (3), 495–510.

Evans, D.H., 1963. Modular design: a special case in nonlinearprogramming. Operations Research 11, 637–643.

Feitzinger, E., Lee, H.L., 1997. Mass Customization at HewlettPackard. Harvard Business Review 75 (1), 116–121.

Fine, C.H., 1998. Clockspeed-Winning Industry Control in theAge of Temporary Advantage. Perseus Books, Reading, MA.

Fisher, M.L., Ittner, C.D., 1999. The impact of product varietyon automobile assembly operations: empirical evidence andsimulation analysis. Management Science 45 (6), 771–786.

Fisher, M., Ramdas, K., Ulrich, K., 1999. Component sharingin the management of product variety: a study of automotivebraking systems. Management Science 45 (3), 297–315.

Flynn, B.B., Flynn, J.E., 1999. Information-processing alternativesfor coping with manufacturing environment complexity.Decision Sciences 30 (4), 1021–1052.

Forza, C., Salvador, F., 2002a. Managing for variety in the orderacquisition and fulfillment process: the contribution of productconfiguration systems. International Journal of ProductionEconomics 76 (1), 87–98.

Forza, C., Salvador, F. Product Configuration Systems: Impactson Operations and Customer Service. Computers in Industry,in press.

Garg, A., 1999. An application of designing products and processesfor supply chain management. IIE Transactions 31 (5), 417–429.

Garud, R., Kumaraswamy, A., 1995. Technological andorganizational designs for realizing economies of substitution.Strategic Management Journal 16 (Special Issue), 93–109.

Gonzalez-Zugasti, J.P., Otto, K.N., Backer, J.D., 2000. A methodfor architecting product platforms. Research in EngineeringDesign 12 (2), 61–72.

Gupta, S., Krishnan, V., 1998. Product family-based assemblysequence design methodology. IIE Transactions 30 (10), 933–945.

Gupta, S., Krishnan, V., 1999. Integrated component and supplierselection for a product family. Production and OperationsManagement 8 (2), 163–182.

Hayes, R., Wheelwright, S.C., 1984. Restoring Our CompetitiveEdge: Competing through Manufacturing. Wiley, New York.

He, D., Kusiak, A., Tseng, T., 1998. Delayed productdifferentiation: a design and manufacturing perspective.Computer-Aided Design 30 (2), 105–113.

Hoekstra, S., Romme, J., 1992. Integral Logistic Structures:Developing Customer-Oriented Goods Flow, McGraw Hill, NewYork.

Huang, C.C., Kusiak, A., 1998. Modularity in design of productsand systems. IEEE Transactions on Systems, Man andCybernetics 28 (1), 66–77.

Ishii, K., Juengel, C., Eubanks, C., 1995. Design for productvariety: key to product line structuring. In: Proceedings of theASME Ninth International Conference on Design Theory andMethodology.

Jiao, J., Tseng, M.M., 1999. A methodology for developingproduct family architecture for mass customization. Journal ofIntelligent Manufacturing 10, 3–20.

Karmarkar, U.S., Kubat, P., 1987. Modular product design andproduct support. European Journal of Operational Research 29,74–82.

Kotteaku, A.G., Laois, L.G., Moschuris, S.J., 1995. The influenceof product complexity on the purchasing structure. Omega23 (1), 27–39.

Krishnan, V., Singh, R., Tirupati, D., 1999. A model-basedapproach for planning and developing a family oftechnology-based products. Manufacturing and ServiceOperations Management 1 (2), 132–156.

Krishnan, V., Gupta, S., 2001. Appropriateness and impact ofplatform-based product development. Management Science47 (1), 69–84.

Kusiak, A., Huang, C., 1996. Development of modularproducts. IEEE Transactions on Components, Packaging andManufacturing Technology: Part A 19 (4), 523–538.

Lancaster, K., 1971. Consumer Demand: A New Approach.Columbia University Press, New York.

Lancaster, K.J., 1979. Variety, equity and efficiency. ColumbiaUniversity Press: New York and Guildford.

Lee, H.L., Tang, C.S., 1997. Modeling the costs and benefits ofdelayed product differentiation. Management Science 43 (1),40–53.

Mather, H.F., 1986. Design, bills of materials, and forecasting: theinseparable threesome. Production and Inventory ManagementJournal 27, 90–107.


McCutcheon, D.M., Raturi, A.S., Meredith, J.R., 1994. Thecustomization-responsiveness squeeze. Sloan ManagementReview 35 (2), 89–99.

Merton, R.K., 1959. Notes on problem-finding in sociology. In:Merton, R.K., Broom, L., Cottrell Jr., L.S. (Eds.), SociologyToday: Problems and Prospects. Basic Books, New York.

Meyer, M.H., Tertzakian, P., Utterback, J.M., 1997. Metrics formanaging research and development in the context of theproduct family. Sloan Management Review 43 (1), 88–111.

Miller, J.G., Vollmann, T.E., 1985. The hidden factory. HarvardBusiness Review 63 (5), 142–150.

Novak, S., Eppinger, S.D., 2001. Sourcing by design: productcomplexity and the supply chain. Management Science 47 (1),189–204.

Pahl, G., Beitz, W., 1984. Developing size ranges and modularproducts. In: Wallace, K. (Ed.), Engineering Design. The DesignCouncil, London, UK, pp. 315–361.

Parnas, D.L., 1971. Information distribution aspects of designmethodology. In: Proceedings of the IFIP Conference, Vol. 1.pp. 339–344.

Parnas, D.L., Clements, P., Weiss, D., 1985. The modularstructure of complex systems. IEEE Transactions of SoftwareEngineering 11 (3), 259–266.

Patton, M.Q., 1990. Qualitative Evaluation and Research. SagePublications, Thousand Oaks, CA.

Pine II, J.B., 1993. Mass Customization: The New Frontierin Business Competition. Harvard Business School Press,Cambridge, MA.

Prasad, B., 1998. Designing products for variety and how tomanage complexity. Journal of Product and Brand Management7 (3), 208–222.

Raman, N., Chhajed, D., 1995. Simultaneous determinationof product attributes and prices, and production processesin product-line design. Journal of Operations Management12 (3/4), 187–204.

Rubin, H.J., Rubin, I.S., 1995. Qualitative Interviewing: The Artof Hearing Data. Sage Publications, Thousand Oaks, CA.

Runkel, P.J., 1990. Casting Nets And Testing Specimens: TwoGrounded Methods of Psychology. Praeger, New York.

Sanchez, R., 1999. Modular architectures in the marketing process.Journal of Marketing, 63 (Special Issue) 92–111.

Seidman, I.E., 1998. Interviewing as Qualitative Research.Teachers College Press, New York.

Sheu, C., Wacker, J., 1997. The effects of purchased partscommonality on manufacturing lead-time. International Journalof Operations and Production Management 17 (7-8), 725–745.

Shirley, G.V., 1990. Models for managing the redesign andmanufacture of product sets. Journal of Manufacturing andOperations Management 3 (2), 85–104.

Starr, M.K., 1965. Modular-production: a new concept. HarvardBusiness Review 43 (6), 131–142.

Stevens, W.P., Myers, G.J., Constantine, L.L., 1974. Structureddesign. IBM Systems Journal 13 (2), 115–139.

Strauss, A.S., 1987. Qualitative Analysis for Social Scientists.Cambridge University Press, Cambridge, UK.

Suh, N.P., 1990. The Principles of Design. Oxford UniversityPress, New York.

Sundgren, N., 1999. Introducing interface management in newproduct family development. Journal of Product InnovationManagement 16 (1), 40–51.

Swaminathan, J.M., Tayur, S.R., 1999. Managing design ofassembly sequences for product lines that delay productdifferentiation. IIE Transactions 31 (11), 1015–1026.

Ulrich, K., Tung, K., 1991. Fundamentals of product modularity.In: Proceedings of the 1991 ASME Winter Annual MeetingSymposium on Issues in Design/Manufacturing Integration.

Ulrich, K., Seering, W.P., 1989. Synthesis of schematic descriptionsin mechanical design. Research in Engineering Design 1 (1),3–18.

Ulrich, K., 1995. The role of product architecture in themanufacturing firm. Research Policy 24 (3), 419–440.

Vakharia, A.J., Parmenter, D.A., Sanchez, S.M., 1996. Theoperating impact of parts commonality. Journal of OperationsManagement 14 (1), 3–18.

van de Ven, A.H., 1989. Nothing is quite so practical as goodtheory. Academy of Management Review 14 (4), 486–489.

van Hoek, R.I., Weken, H.A.M., 1998. The impact of modularproduction on the dynamics of supply chain. InternationalJournal of Logistics Management 9 (2), 35–50.

van Hoek, R.I., 2001. The rediscovery of postponement: a literaturereview and directions for research. Journal of OperationsManagement 19 (2), 161–184.

von Hippel, E., 1990. Task partitioning: an innovation processvariable. Research Policy 19 (5), 407–418.

voss, C., Tsikriktsis, N., Frdich, M., 2002. Case research inoperations management. International Journal of Operations andProduction Management 22 (2), 195–219.

Wallace, W.L., 1971. The Logic of Science in Sociology. AldineAtherton, Chicago, IL.

Yano, C.A., Dobson, G., 1999. Profit-optimizing product linedesign, selection, and pricing with manufacturing costconsideration. In: Ho, T.-H., Tang, C.S. (Eds.), Product VarietyManagement-Research Advances. Kluwer Academic Press,Boston, MA, pp. 145–176.

Yin, R.K., 1988. Case Study Research: Design and methods. SagePublications, Newbury Park, CA.

Zetterberg, H., 1954. On Theory and Verification in Sociology.Bedminster Press, Totowa, NJ.

Documents

Modularity, product variety, production volume, and component sourcing