The Pitfalls of Subsystem Integration: When Less Is MoreStylianos Kavadias, Cheryl Gaimon Scheller College of Business, Georgia Institute of Technology, Atlanta, Georgia 30332 {[email protected],

MANAGEMENT SCIENCEVol. 59, No. 3, March 2013, pp. 659–676ISSN 0025-1909 (print) ISSN 1526-5501 (online) http://dx.doi.org/10.1287/mnsc.1120.1592

© 2013 INFORMS

The Pitfalls of Subsystem Integration:When Less Is More

Sanjiv EratRady School of Management, University of California, San Diego, La Jolla, California 92093, [email protected]

Stylianos Kavadias, Cheryl GaimonScheller College of Business, Georgia Institute of Technology, Atlanta, Georgia 30332

{[email protected], [email protected]}

In various industries end-product manufacturers acquire core subsystems from upstream technology providerfirms and focus primarily on efficient end-product integration. We examine the strategic interactions between atechnology firm that introduces a new subsystem and the respective end-product manufacturers (“integrators”).We analyze how the fraction of end-product functionalities prepackaged into the subsystem impacts the optimalintroduction strategy and the relative value appropriation power across the industries. Offering a subsystemthat performs many end-product functions has a dual effect on the provider’s profits. On the positive side, theprovider extracts a higher ease-of-use rent from the integrators because of the easier/cheaper integration. Onthe negative side, such subsystems may curtail the adopters’ ability for competitive differentiation and renderadoption less valuable. We discuss the role of subsystem functionality in value appropriation in technologymarkets, and we highlight the perils of subsystem overintegration.

Key words : technology introduction; technology licensing; technological value appropriation; subsystemfunctionality

History : Received February 19, 2009; accepted February 21, 2012, by Kamalini Ramdas, entrepreneurship andinnovation. Published online in Articles in Advance November 5, 2012.

1. IntroductionMany industries exhibit a trend of disintegrationover time. They start off vertically integrated, andgradually they evolve toward a multitier disinte-grated structure (Christensen 1994, Christensen et al.2002). In such vertically disintegrated value chains,the downstream manufacturers assume the role ofend-product integrators; they assemble and integratemany of the end-product subsystems, be it large com-ponents or process technologies, which are developedand introduced by upstream technology providers.These new disintegrated industrial structures pose anew set of challenges for the participating firms.

The following example illustrates a key challengethat upstream technology providers and downstreamintegrators face, the former when they introduce anew subsystem technology, and the latter when theyadopt it. In the mid-1980s, one such disintegratedvalue chain consisted of approximately 30 large carpetmill companies (the downstream tier), who used key(process) technologies developed by an upstream tierof technology providers (such as DuPont). These car-pet mill companies were competing on both price andend-product performance with relatively low mar-gins. During this time, DuPont Chemicals, a lead-ing innovator in chemical and bio-based process

technologies, introduced a novel process technology,named Stainmaster. Stainmaster allowed the carpetmills to significantly enhance the stain resistance oftheir carpets, an important performance dimensionfor the large volume carpet market. DuPont intro-duced the Stainmaster technology as a highly inte-grated offering, where much of the adopter’s processsteps were specified in the licensing agreement. Eventhe end-product branding had to be uniform and uti-lize the title Stainmaster (Miller 2005).

The performance improvement in the Stainmasterproducts was unquestionable (Mello et al. 2006,p. 192). Together with an intense advertising cam-paign, Stainmaster became a “must have” featureof any carpet. DuPont aggressively pushed Stain-master, which was ultimately adopted by nearlyall large downstream integrators. Although the full-fledged solution of Stainmaster reduced the cost anduncertainty of an adopter’s integration effort, therestrictions to customize the offerings also resulted inlimited ability to differentiate end products. Indeed,according to a senior manager at DuPont, their large-scale rollout combined with the highly integratednature of the technology sparked more intense pricecompetition, reducing the already thin profit mar-gins of carpet mills (Miller 2005). This period also

659

Erat, Kavadias, and Gaimon: The Pitfalls of Subsystem Integration: When Less Is More660 Management Science 59(3), pp. 659–676, © 2013 INFORMS

recorded a rapid and intense consolidation of the car-pet mills industry. Several mills exited in bankruptcy,the downstream industry became more concentrated,and DuPont’s ability to appropriate value (henceforth,value appropriation power) diminished. The lesson fromthe Stainmaster experience was not lost at DuPont.Indeed, one of the members of DuPont’s senior man-agement team framed a key goal for future technol-ogy introductions as follows: “enough value [shouldbe] left on the table for licensees” and “[make] doublysure that the licensees could transform the technologyinto differentiated products” (Miller 2005).

The preceding example is an instance that moti-vates the core trade-off explored in this paper:the technology provider may be ill-served by ahighly integrated subsystem even when the subsys-tem reduces the risk/cost of integration, becausehighly integrated subsystems place limitations ondownstream competitive differentiation. This trade-off arises because of a subsystem effect that is dif-ferent from the more traditional view discussed inthe literature, whereby subsystems are assumed todetermine either the end-product marginal cost ortheir performance (Kauffman et al. 2000). Instead, theStainmaster example suggests an alternative designdimension that may play an equally important andcomplementary role: the amount of end-product func-tions (integration) offered by subsystems.

Our argument about the potential negative impli-cations from subsystem integration, although likelyto be qualitatively true in general, has greater eco-nomic significance in some industrial contexts thanin others. For example, in some industrial markets,such as electronics, dominant technology providerslike Intel have shown an increasing tendency towardoffering highly integrated, full-fledged subsystems(compared to piecemeal solutions). In these contexts,the level of integration not only alters the integra-tion costs/risks (our key concern in this paper) butmay also play a role in determining the end-productperformance, and therefore, the downstream user’sprofitability (Kamien and Tauman 1984, 1986). It ispossible, therefore, for a firm to find that the possibleupside of offering superior end-product performance(identified in past literature, for instance, see Kamienand Tauman 1984) through greater subsystem integra-tion is sufficiently large and outweighs the downsidewe identify from reduced integration risk/costs.

Elsewhere, a single large downstream integrator(e.g., IBM Global Services) may dominate the end-product market. After a unique transformation intoa service solutions company, IBM has leveraged thehardware or software subsystems (upstream tech-nology) very carefully to grow and retain marketshare. Indeed, their focus on carefully managing theirupstream providers is cited as a core element of their

approach.1 Although a complete enumeration of allthe issues that determine a provider’s design choiceslies beyond the scope of our study, we attempt tocharacterize how the potential pitfalls or benefits ofsubsystem integration depend on (i) the subsystemcontribution to the end-product performance as wellas (ii) the heterogeneous capabilities of the down-stream integrators. In that regard, our study examinesthe moderating role of subsystem integration choiceson the relative value appropriation power2 between theupstream provider and downstream integrators.

We consider a business setting where a technologyprovider offers a subsystem that adds value to adopt-ing integrator firms in one (or both) of two possibleways: (i) integration benefits such as the reductionof the cost/time/uncertainty of the adopter’s inte-gration process, which we term “ease-of-integration”benefits, or (ii) higher end-product performance. Werefrain from distinguishing between subsystems thatare component technologies or process technologies,and we build on Ulrich’s (1995) definition of function-ality to accommodate the case of process technolo-gies as follows: a subsystem is defined as a collectionof functions and integration activities,3 and the sub-system “functionality” is the metric that summarizesthe fraction of end-product functions and integrationactivities embedded in the subsystem.

Motivated by the earlier DuPont example, we as-sume that the upstream provider enjoys a monopolyposition, but that the downstream industrial settingis competitive. We also extend our base model toallow heterogeneous capabilities among the integra-tors, and thus explore the constraints that the com-petitiveness in the downstream industry impose onthe provider’s value appropriation power. Moreover,our analysis also examines the role of different feestructures—fixed upfront licensing fees and variable(royalties) fees—in determining the provider’s valueappropriation power.

We explore two main research questions:(1) How does the subsystem functionality affect the

relative value appropriation power between the sub-system provider and the downstream industry?

(2) What introduction strategy should the technologyprovider undertake?

1 We thank the associate editor for offering this example.2 Our use of the term “relative value appropriation power” reflectsthe fact that in several industrial settings the value appropriated bythe upstream provider might be smaller, and therefore her power toappropriate diminishes, despite the fact that the ex ante conditionsfor the downstream integrators are the same. Note that this usageis distinct from the economic interpretation of power in terms ofbargaining outcomes resulting from splitting the total value acrossthe different markets.3 Integration activities are the steps associated with integrating theparticular subsystem into the complete product.

Erat, Kavadias, and Gaimon: The Pitfalls of Subsystem Integration: When Less Is MoreManagement Science 59(3), pp. 659–676, © 2013 INFORMS 661

In answering the first question, we find that, ceterisparibus, greater functionality in a subsystem hastwo opposing effects on the upstream provider’svalue appropriation power. On the positive side, theprovider may be able to obtain a larger ease-of-integration premium from the integrators. On thenegative side, greater functionality may curtail thedownstream ability to competitively differentiate,especially when the subsystem is widely adopted.Thus, greater functionality may make a subsystemless attractive for the integrators, and consequentlyreduce upstream provider’s relative value appro-priation power. The result highlights an important,cautionary message for technology providers onthe potential pitfalls of overintegration, especiallywhen the subsystem is targeted at multiple inte-grators. At the same time, the result also revealsthe nonmonotonic effect of functionality on upstreamvalue appropriation power and on the market sharethat the subsystem can garner. Note that suchnonmonotonicity in profits and market share thatdirectly emerges from the negative effect of function-ality identified earlier, is in contrast to most, if not all,previous models of technology introduction, wherea superior technology performance can only serveto enhance a provider’s value appropriation power(Kamien and Tauman 1984, 1986; Kamien 1992; Eratand Kavadias 2006).

Our results also reveal valuable insights aboutthe implications of a subsystem on the downstreamindustry: even when a subsystem diminishes integra-tion costs and risks, or enhances the end-product per-formance substantially, it may not be widely adopted,as a result of the strategic interactions among theintegrators. Interestingly, the subsystem technologymight result in an increase in the downstream indus-try’s power to appropriate value whereby many inte-grators with similar integration capabilities intenselycompete, and are transformed into fewer integratorsthat are more capable, and ultimately dominate theend-product market. We also find that heterogene-ity between the integrators may restrict the numberof adopters who benefit from the subsystem offered,and therefore diminish the provider’s value appropri-ation power. These insights reveal the complex inter-play between design factors and industry level factorsand thereby add to the extant literature on this topic(Baldwin and Clark 1999, Fixson and Park 2008).

At a much more operational level, we show how thesubsystem introduction strategy, i.e., the optimal num-ber of integrators to target, and the fee structure thata provider employs, depend on subsystem function-ality. For very low and very high functionality levels,the provider should optimally offer the subsystem toa limited number of integrators. Furthermore, mixedfee structures, employing both a fixed and a royalty

component, are optimal when the technology providertargets the subsystem to multiple integrators. Theseresults offer normative support to past empirical find-ings on the prevalence of such mixed fee structures(Rostoker 1984) and add to the growing stream of lit-erature that has identified plausible conditions thatfavor the superiority of volume-based fees (see, forinstance, Wang 1998, Sen and Tauman 2007).

2. Literature ReviewTwo main areas in the academic literature explore dif-ferent aspects of our research question: (i) the newproduct development literature examines the effectof product architecture on the firm’s strategic deci-sions, and (ii) the patent-licensing literature from eco-nomics and strategy examines the implications ofmarket (and technology) characteristics to technologyintroduction.

2.1. The Effects of Product Architecture onFirm Strategy

Ulrich (1995) defines the product architecture as “thescheme by which the function of a product isallocated to its physical components” (Ulrich 1995,p. 419). Building on this definition, we conceptualize asubsystem as a collection of functions and integrationactivities, and the subsystem functionality as the met-ric summarizing the fraction of end-product functionsand integration activities embedded in the subsystem.

It has long been recognized that the productarchitecture has important implications to a firm’soperational performance on dimensions includingflexibility, efficiency, and profitability (Baiman et al.2001, Krishnan and Gupta 2001, Dana 2003, Ülkü andSchmidt 2011). Recently, Fixson and Park (2008) devel-oped a conceptual framework based on an in-depthcase study that examines the role of architecturalchoices in the structure of upstream and down-stream tiers across the bicycle (components) valuechain. The majority of these studies focus on howdesign and architectural dimensions affect a firm’sproduct strategy (e.g., intertemporal price discrimi-nation, upgradeability, product variety, etc.) and theupstream industry structure (Fixson and Park 2008).

A related, yet distinct, stream of research instrategy argues that the product and process architec-tures may influence the value of intraorganizationaland interorganizational coordination and governancemechanisms, thereby influencing a firm’s long-term profitability. Within this stream, Sanchez andMahoney (1996) theorize that modularity in archi-tectures facilitate loose coupling, and thus have thepotential to reduce the cost and difficulty of adap-tive coordination, thereby increasing the strategic flex-ibility of firms to respond to environmental change.Thus, modularity in architectures should not be just


related to firm performance (Worren et al. 2002) butshould also serve as a substitute for extensive coordi-nation and control routines, both within an organiza-tion and between organizations in alliances (Tiwana2008). In a recent study, Cabigiosu and Camuffo(2012) enrich this proposition by hypothesizing thatbecause the need to share information is minimizedwith highly modular designs, the positive impact ofinformation sharing on supply chain performance isitself negatively moderated by the modularity of theproduct designs employed in the supply chain.

Although we examine a specific attribute of theproduct architecture, namely, the functionality of thesubsystem technology, our emphasis is complemen-tary to these past studies. Specifically, we seek tounderstand how the functionality of a technologysubsystem offered by one upstream economic entity(the provider) to multiple downstream entities (theintegrators) determines their respective value appro-priation (profits) and more broadly the structureof the downstream tier (i.e., resultant asymmetriesamong the integrators).

2.2. Technology Introduction and Patent LicensingSince the pioneering work of Arrow (1962), a num-ber of studies have analyzed when and how innova-tions (intellectual property rights) are licensed. Thisstream of literature has mainly focused on market sidedrivers, and it has viewed technology as a unidimen-sional factor that improves the end-product perfor-mance or reduces its marginal cost.

Early work in this stream, by Kamien and Tauman(1984, 1986) and Katz and Shapiro (1985), exam-ines the case of an upstream innovator who licensescost-reducing innovations to downstream competingfirms. They find that upstream innovators obtain lim-ited value through royalty licensing mechanisms.4

Kamien (1992) offers a comprehensive survey of thisstream of literature.

There are very few empirical studies of licensingagreements for the case of upstream innovators, pos-sibly because of the difficulty to obtain detailed dataabout licensing agreements. The few available studiesnote that volume-based royalties are a key element ofmost licensing agreements despite the theoretical pre-diction of their suboptimality. For instance, Rostoker(1984) finds that the fixed-price plus volume royaltyis the most frequently used (46%), followed by purevolume-based royalties (39%). These empirical regu-larities have spurred a number of important inquiries,e.g., the role of information asymmetry (Gallini andWright 1990), risk sharing (Bousquet et al. 1998), etc.

4 For instance, in the case of risk-neutral agents with complete infor-mation, Kamien and Tauman (1986) show that volume-based roy-alties are inferior to fixed-price licenses.

These extensions show that volume-based royaltiesmay alleviate two-sided moral hazard and optimallysplit the innovation risk between risk-averse parties.We add to this stream of literature by demonstratingthat royalties may also benefit the innovator by mod-erating the downstream competition.

Starting with Shapiro (1985), research has expandedthe context of licensing to include the case of licens-ing among competitors within an industry, i.e., theinnovator is no longer an outsider and may actuallycompete in the same product market as the licensee.Within this stream, Arora and Fosfuri (2003) examinethe case of an incumbent firm licensing to her com-petitor and identify two drivers for the innovator’sprofits: (i) a revenue effect, resulting from the licensefees, and (ii) a rent dissipation effect, resulting fromthe licensee who is a competitor possessing a supe-rior (or equal) cost structure. Although we examinea different context, through a suitable interpretationof the rent dissipation effect, our results expand thescope of their findings to the case where the innova-tor is an upstream technology provider. In addition,our explicit consideration of the integration processallows us to identify the role of the upstream technol-ogy on the revenue and rent dissipation effects.

Recently, researchers have examined the productdevelopment and marketing implications of (optimal)licensing decisions. They have attempted to opera-tionalize both the demand side (i.e., market potential)and the supply side (i.e., cost of technological devel-opment) of licensing decisions. Erat and Kavadias(2006) examine the effects from offering subsystemperformance enhancements (defined as the subsys-tem contribution to the end-product performance) onan upstream provider’s licensing decisions. They findthat higher current subsystem performance promptsthe provider to at least keep or even expand thenumber of licensees. Furthermore, aligned with theextant literature, they find that the provider revenuesincrease in the subsystem performance. In contrast,in our model, a highly integrated subsystem, throughits potential to intensify downstream competition (bylimiting the downstream competitive differentiation),may result in a decrease in the provider’s revenues.Stated differently, a technology provider may ulti-mately hurt the downstream adopters (and hence her-self) by trying to be helpful and offering a highlyintegrated subsystem. This contrast illustrates the dif-ferent effects of the two design dimensions of thesubsystem, i.e., end-product performance contribu-tion (examined in Erat and Kavadias 2006) versussubsystem functionality examined in our study.

Overall, past studies have concentrated on the end-product performance effects of technologies and haveignored that a subsystem technology may also be val-ued for its integration effects. Our study highlights


this additional dimension of integration-related effects,and outlines the different set of challenges that influ-ence the introduction strategy for such subsystemtechnologies. Thus, we complement past studies byexplicitly accounting for the effect of subsystem func-tionality on the uncertainty/cost of the integrationprocess.

3. Model SetupA monopolist technology firm introduces a subsys-tem C to a downstream market of n integrators. Thesubsystem C offers functionality f , i.e., the fraction ofend-product integration functions/activities prepack-aged into the subsystem. We assume that a part (if notall) of the technologies embedded in the subsystem areprotected by patents, and therefore other firms cannotreplicate the same subsystem. For tractability, we con-sider a duopoly in the downstream market, i.e., n= 2.

If an integrator adopts the subsystem offeredby the provider, his5 integration process and end-product performance may both be impacted. First,consider the effect on the integration process. A down-stream integrator, upon adoption of the subsystem C,may undertake further development and integration.We consider a model where at the end of the “integra-tion period,” the integrators are either (i) unsuccessfulin their integration efforts and revert back to the exist-ing (status quo) subsystem, say ±C, or (ii) successfuland utilize the new subsystem C in their end product.6

Each integrator i (i = 112) can successfully inte-grate the subsystem C into his end product withprobability p4f 5 after incurring cost Ci4f 5. We assumethat the probability of successful integration is iden-tical across the two integrators, to keep the analysisstraightforward. Still, to account for industrial set-tings close to our example of IBM Global Services, wecapture heterogeneous capabilities among the down-stream integrators through the heterogeneous inte-gration costs. We parameterize this heterogeneity bydefining C14f 5 = C4f 5 − and C24f 5 = C4f 5 + ,respectively, ( ≥ 0). Hence, a nonzero allows us torepresent different degrees of heterogeneity and valueappropriation power among the upstream and down-stream industries.

A subsystem that offers lower functionality neces-sitates greater integration/development efforts bythe integrator firm, because the integrator needs

5 We refer to downstream integrators as “he” and the technologyprovider as “she.”6 For ease of exposition, we have discretized the development/integration process and consider a single “integration period.”Still, in Online Appendix 2.3 (available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1381624), we show thatour results are generalizable to settings with continuous integra-tion time.

to employ alternative subsystems and componentsin order to ensure end-product performance. Thus,lower functionality creates a more complex and costlyintegration process, so that the integrator needs toinvest more resources (Clark 1989) and is less certainabout the integration success because of the potentialfor numerous subsystem interactions (Schilling 2000).Hence, it is likely that p4f 5 is a nondecreasing func-tion of f , and that C4f 5 is a nonincreasing functionof f . Specifically, we assume that there is a nonde-creasing nonnegative function 4f 5 (≤ 1) such that7

Assumption A.0.1. p4f 5=4f 5.

Assumption A.0.2. C4f 5 = K41 − 4f 55, where Kand ≤ 1 are nonnegative constants.

Second, the adoption and successful integration ofthe subsystem into the end product may offer anadditional well-understood benefit, namely, enhancedend-product performance. Note that we conceptualizethe end-product performance as a key competitivedimension in an industry, varying from industry toindustry. In some contexts marginal cost may be thebasis for competition, whereas in others it may betechnology performance such as power consumptionor battery life. Let an integrator’s end-product per-formance be P4 ±C5 when he utilizes the existing sub-system, and P4C5 when he uses the newly offeredsubsystem. In general the end-product performancedepends on various design features of the subsystem.Consistent with our research question, we limit ourfocus only on the effect of subsystem functionality onend-product performance. Therefore, we define theperformance of the end product that uses the subsys-tem C as P4C5= P4f 5.

Let ç4Pi1Pj5 be the payoff to integrator i (i = 112)when his end product exhibits performance Pi and thecompetitive offering (that is j’s end product, j = 3 − i)has performance Pj . The structure of the profit func-tion ç4 · 1 · 5 subsumes the market mechanism thatallocates profits to the competing integrators. Insteadof assuming a specific form for the market mechanism(e.g., Cournot competition), we adopt an axiomaticapproach to modeling competition between integra-tors, and we assume an intuitive structural propertythat any reasonable market mechanism would satisfy.

7 Although we only require that 4 · 5 is a nondecreasing nonneg-ative function, there are also alternate plausible conditions underwhich our results hold. For instance, it may be shown that ourresults hold if the point elasticity of the marginal change in costsis never greater than the point elasticity of the marginal change inprobability (i.e., C ′′4f 5/C ′4f 5≤ p′′4f 5/p′4f 5). A particular case wherethis last condition holds is when p4f 5 and C4f 5 are convex. Evenwhen the previous condition is violated, and C4f 5 and p4f 5 arearbitrary nonincreasing and nondecreasing functions, respectively,our main structural results, including Propositions 1 and 2, andCorollaries 1 and 2 hold.


Assumption A.1. ç4P ′i 1Pj5−ç4Pi1Pj5 is nonincreas-ing in Pj for all P

′i >Pi; i.e., ¡

2ç4Pi1Pj5/¡Pi¡Pj < 0.

In economic terms, the performances are “substi-tutes,” i.e., ç4Pi1Pj5 is submodular in 4Pi, Pj5 (Topkis1978). Intuitively, the marginal benefit from a superiorperformance is greater if the competitor lags signifi-cantly in performance.8

Without loss of generality, we normalize the per-formance of end products currently offered by boththe integrators (i.e., before any adoption decisions) toP4 ±C5= 1. For notational convenience, define (i) a4P5=ç4P4C51P4C55, (ii) b4P5 = ç4P4C5115, (iii) c4P5 =ç411P4C55, (iv) d = ç41115. That is, a4P5 is the pay-off to an integrator when both he and his competi-tor employ the subsystem C; b4P5 is the payoff whenonly he employs subsystem C; c4P5 when only hiscompetitor employs subsystem C; and d when nei-ther of the integrators employs the new subsystem C.Let S4P5 = a4P5 − c4P5, and F 4P5 = b4P5 − d. Thus,S4P5 (F 4P5) represents the incremental benefit of theadopter when his competitor has (not) adopted thesame subsystem.

Finally, given the integration-related and end-product performance-related benefits that the subsys-tem offers, the technology provider sets the adoptionfees ±W for the specific subsystem C. In general, the fee±W = 6W1w7may be composed of two parts: a fixed one-time payment, W , paid upon adoption, and a per-unitroyalty, w, paid for each unit of the end product sold.

The sequence of decisions (Figure 1) for theupstream and downstream parties is as follows. In thefirst stage, the technology provider sets the fees ±Wfor a specific subsystem C. Given the fee, the inte-grators decide based on anticipated profits, whetheror not to adopt the subsystem. Next, the integratorscommence additional development efforts to integratethe acquired subsystem into their end product (at costC4f 5 − or C4f 5 + ) and are successful with prob-ability p4f 5. In case of success, the integrator utilizessubsystem C in their end product, whereas in caseof failure he reverts to the older (status quo) subsys-tem ±C. End products are then competitively sold andgenerate revenues as per the ç4Pi1Pj5 mechanism. Thebasic notation and assumptions of our model setupare summarized in Table 1.

8 It is possible that the integrators compete on several dimensions,with end-product performance being only one of them. In OnlineAppendix 2.2 (available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1381624), we discuss how our assumptions maponto quite complex competition models and we demonstrate thatAssumption A.1 is indeed quite general and is satisfied by awide variety of economic models of competition, including tradi-tional quantity-based competition, differentiated Bertrand modelsof price competition, vertical differentiation models, etc. Further-more, although submodularity is a relatively weak assumption, wealso describe how a significantly weaker set of assumptions is ade-quate to obtain our main results.

Figure 1 Timeline of Decisions

Given subsystem �with functionality f

set fees [W, w]

Adopt the subsystem?

Undertake integration

Technologyprovider’sdecision

Integrator’sdecision

YES

NO

Successful?

Manufacture using newlyintegrated subsystem

Manufacture using oldersubsystem

Compete for end-product market

NOYES

4. Subsystem Functionality, theOptimal Introduction Strategy,and Value Appropriation

This section presents our core result: the role ofthe subsystem’s functionality on the provider’s valueappropriation, and on her optimal introduction strat-egy. The effect of enhancing the end-product perfor-mance is fairly well understood (Erat and Kavadias2006). Hence, in the first step in our analysis, we “shutoff” this mechanism by explicitly assuming that theend-product performance is unaffected by the func-tionality of the underlying subsystem C. This allowsus to offer clear contrast between our analysis andthat offered in Erat and Kavadias (2006). Thus, theperformance obtained by integrator i, Pi, althoughpossibly dependent on broader subsystem character-istics, is assumed to be independent of f . We relaxthis assumption in §5.1.

Assumption B.1. P4C5 is independent of f .

We also assume that the value appropriation powerin the downstream market is minimal; the integratorspossess identical capabilities and cost structures, andtheir costs of integration are identical.9 We considermore general settings in §5.2. For now,

Assumption B.2. = 0; i.e., C14f 5=C24f 5=C4f 5.

Finally, we assume that the subsystem licensingfee ±W is composed of only the single fixed one-time

9 It is a straightforward argument that when all downstream inte-grators are symmetric in their integration capabilities, they pos-sess minimal appropriation power, because they face maximumcompetition.


Table 1 List of Variables and Key Assumptions in the Model

Variable Name Assumption

f Functionalitypi = p = p4f 5 Likelihood of integration success A.0.1 p4f 5= 4f 5 is nondecreasingC1 = C4f 5− Integration costs A.0.2 C4f 5= K41 − 4f 55; K , ≥ 0C2 = C4f 5+

P End-product performance A.0.3 P 4f 5 is nondecreasing in fç4Pi 1 Pj 5 Integrator i ’s payoffs A.1 submodular±W = 6W 1w7 Subsystem adoption fees

payment. Our assumption reflects industrial settingswhere providers avoid more complex forms, such asroyalties, perhaps because of significant costs such asmonitoring for compliance, etc. We relax this assump-tion in §5.3. For now,

Assumption B.3. w = 0.

Whenever the meaning is apparent, we suppressthe notations and use p, C, S, and F for p4f 5, C4f 5,S4P5, and F 4P5, respectively. Furthermore, we followErat and Kavadias (2006) and define the differentintroduction strategies that may arise.

Definition 1. An introduction strategy where thetechnology provider sets fees such that

• both integrators adopt the subsystem, termed asaturation strategy;

• only one integrator adopts the subsystem,termed a niche strategy;

• none of the integrators adopt the subsystem,termed a no-sale strategy.

Next, we define a metric that both allows us to cap-ture the probabilistic nature of the integration processand its ex post outcomes and facilitates the explana-tion of the intuition behind our main result.

Definition 2. The degree of differentiation potential isthe probability that the integrators sell end productswhose performances differ.

Degree of differentiation potential (DDP)

=

0 if introduction strategy is no-sale1p if introduction strategy is niche12p41−p5 if introduction strategy is saturation0

When neither integrator adopts the subsystem (no-sale), then ex post both will sell end products thathave status performance 1, thus making the dif-ferentiation 0. When only one adopts the subsys-tem (niche strategy), then the nonadopter always hasex post end-product performance of 1, whereas theadopter has ex post performance P with probabil-ity p, and with remaining probability 1 − p has expost performance 1. Stated differently, under a nichestrategy, there is a likelihood p that the ex post per-formance of the two downstream integrator’s end

product would be different. In a similar manner,even when both integrators adopt the new subsys-tem (saturation strategy), the integration uncertaintyresults in a nonzero likelihood that their end prod-ucts exhibit different ex post performances. Specifi-cally, the end-product performances can be differentonly when one integrator fails his integration processand the other succeeds, which happens with proba-bility 2p41 − p5.

Proposition 1 describes the (optimal) licensing feethat the provider charges contingent on the numberof integrators she targets.

Proposition 1. The provider optimally employs thesaturation strategy by setting the fee at W =Ws −C, where

Ws = pF − p24F − S51

or she may optimally pursue the niche strategy by settingthe fee at W =Wn −C, where

Wn = pF 0

The corresponding provider payoffs are s = 24Ws −C5 andn =Wn −C. The overall optimal introduction strategy issaturation (niche) iff s > max801n9 (n > max801s9).

Note that because the integrators are symmetric(homogeneous) (Assumption B.2), the niche strat-egy (when W = Wn − C) corresponds to two dis-tinct equilibria, one in which integrator 1 licensesand integrator 2 does not, and another in which inte-grator 2 licenses and integrator 1 does not. Admit-tedly, the nonuniqueness of equilibria may in generalraise a concern for the need for potential equilibriumrefinements. However, given that our focus is on theprovider’s optimal decisions, the issue of which equi-librium emerges, albeit theoretically interesting, is lessrelevant; this is because irrespective of the equilib-rium chosen, the technology provider gains the samerevenues, and hence is indifferent between them.

Proposition 1 reveals an interesting insight: Thetechnology provider’s revenue under the saturationstrategy is not monotonically increasing with the like-lihood of successful integration p. Specifically, for veryhigh values of p, the marginal value of p may fail to


be positive. This result is somewhat unusual10 becausefor p > 1/2, the likelihood of integration failure andthe integration uncertainty (measured by the variancep41−p5) decrease with an increase in p. Thus, we showthat the provider’s revenues may in fact decreasewhen the uncertainty in the integration process andeven the likelihood of integration failure are lower.

The impact of p on the provider’s profitability isexplained as follows: the higher p indicates a higherlikelihood that an adopter is successful in her inte-gration effort. However, under a saturation strat-egy, the higher p also implies that her competitorhas the same greater likelihood of being successful.In that case, the potential for a differentiated out-come (Definition 2) decreases as p increases. Thislower potential of a differentiated outcome makes itless likely that any one integrator obtains the high“monopolist” payoffs, and most likely that both inte-grators obtain the duopoly payoffs. As a result, theintegrators have less incentives to adopt the subsys-tem, and thus may be induced to adopt only throughlower fees. Viewed from the perspective of the tech-nology provider, this suggests that although an easierto integrate subsystem (higher p) is valued by anygiven individual integrator (upon exclusive access),the same subsystem could be less valuable to thedownstream market as a whole. Thus, an easier tointegrate subsystem decreases the value appropria-tion power of the upstream provider even when thedownstream industry is fiercely competitive (symmet-ric firms).

There is an additional interesting connection be-tween our finding and the rent dissipation effect, animportant driver identified from past research exam-ining why and when firms license technologies totheir competitors (Arora and Fosfuri 2003). Rent dis-sipation in these contexts refers to the reduction inthe licensor’s profits because the licensee is a competi-tor in the end-product market. Note that in our casethe licensor does not compete with the licensee. Still,the fact that multiple licensees compete for the sameend-product market, combined with the fact that veryhigh p results in an exceedingly low possibility of adifferentiated outcome, results in a situation wherethe integrator’s rents are reduced. Thus, by reinter-preting the rent dissipation as a probabilistic effectthat depends on the actual integration outcome, wemay expand the scope of the Arora and Fosfuri (2003)results to cases where the innovator is an outsider andthe outcome of the integration process is uncertain.

Corollary 1. 1. There exists a threshold (=F /42 ·4F − S55) for the likelihood of successful integration p suchthat above the threshold, the provider revenues under a sat-uration strategy decrease in p.

10 We thank one of the reviewers for pointing this out.

2. There exists a threshold (=−144F +K5/424F −S555)for the subsystem functionality f such that above thethreshold, the provider revenues under the saturation strat-egy decrease in f .

Corollary 1 summarizes an important managerialinsight obtained from Proposition 1. Introduction of asubsystem with a greater likelihood of successful inte-gration (higher p), and/or with a lower integrationcost (lower C), may not be a beneficial value appro-priation strategy. Easier integration has an importantside effect: it reduces the potential for differentiation,and therefore increases the downstream competitionintensity.

The next proposition characterizes the optimalintroduction strategy as a function of the two keyintegration-related benefits of a subsystem: lower like-lihood of integration failure and lower integrationcost. In addition, the proposition states the role of sub-system functionality on the introduction strategy.

Proposition 2. 1. There exist thresholds on the likeli-hood of successful integration p0, p1, and p2 such that forp < p0 no-sale is optimal, for p0 ≤ p < p1 a niche strategyis optimal, for p1 ≤ p ≤ p2 a saturation strategy is optimal,and for p2 ≤ p a niche strategy is optimal.

2. There exist thresholds on integration costs C0 and C1

such that for C < C0 a saturation strategy is optimal, forC0 ≤ C < C1 a niche strategy is optimal, and for C ≥ C1

no-sale is optimal.3. There exist thresholds on the subsystem functionality

0 ≤ f0 ≤ f1 ≤ f2 ≤ 1 such that the provider finds it optimalto (i) undertake the niche strategy if f ∈ 6f01 f17 or f ∈6f2117, (ii) undertake the saturation strategy if f ∈ 6f11 f27,and (iii) not offer the subsystem to any of the integratorsif f ∈ 601 f07.

Figure 2 plots the optimal introduction strategycontingent on the likelihood of successful integration

Figure 2 Regions Corresponding to Different Levels of IntegrationUncertainty and Cost

No-saleNiche

Likelihood of integration success p

Cos

t of

inte

grat

ion

C

Saturation


p and the integration cost C. For very low likelihoodof successful integration, the expected adoption ben-efit to an integrator is also very low. Hence, the onlyway the provider may induce an integrator to adoptthe subsystem is through a negative license fee, i.e., aform of “subsidy.” This would result in negative prof-its, so the provider does not introduce the subsystem.As the likelihood of successful integration takes onmoderate values, the expected benefit from adoptionincreases. Consequently, the provider offers the sub-system to more and more integrators. However, forvery high likelihood of successful integration, everyadopter can successfully integrate the technology intotheir end product, and as a result, the degree of poten-tial differentiation between them decreases and thecompetitive intensity increases. Following the reason-ing of Corollary 1, this effect dilutes the value theprovider appropriates from saturation relative to theniche strategy.

The impact of the integration cost on the optimalintroduction strategy is intuitive. Integration costsdirectly reduce the license fee that a given integra-tor is willing to pay. Under the saturation strategy,the revenue loss to the provider because of high inte-gration costs is greater compared to her revenue lossunder the niche strategy (in the former case, the inte-gration cost affects more integrators compared to thelatter case). Thus, for very high costs, the provideroptimally adopts a niche strategy (or chooses not tointroduce the subsystem). As the costs become lower,it becomes feasible to offer the subsystem to multipleintegrators (saturation strategy).

It is interesting to compare our results with priorfindings on the optimal licensing of technologies thatreduce the adopter’s marginal cost. A key findingemerges from that literature: breakthrough technolo-gies that significantly lower marginal costs resultin fewer licensees compared to moderate improve-ments (Sen and Tauman 2007). If we interpret thereduction of integration costs as a technology effect,then we see that our result suggests the exact oppo-site: drastic reduction in integration costs results inmore licensees. This contrast stems from the differ-ent role that a one-time fixed integration cost playscompared to the role played by the variable man-ufacturing costs. Specifically, a reduction in integra-tion costs directly reduces an adopter’s cost structure,independent of whether or not a competitor adoptsthe subsystem. In that regard, the adoption decisioncarries a constant (competition-independent) value.Instead, a reduction in the adopter’s manufactur-ing costs (examined in past literature) allows theadopter to compete more effectively, but its valuebecomes lesser when her competitor also has the samelower manufacturing costs. Thus, our result indicatesthat taking a more operational perspective on the

Figure 3 Provider’s Profit and Optimal Introduction StrategyContingent on Subsystem Functionality

Functionality f

No. of adopters

Two integrators One integratorOne integratorNo salePro

vide

r’s

prof

its

Notes. The solid (dashed) curve represents the profit when pursuing theniche (saturation) strategy. The maximum of the two curves gives theprovider’s optimal profit.

technology introduction benefits, e.g., manufacturingcost reduction versus integration cost reduction, hasdramatically different managerial implications for theoptimal introduction strategies.

Figure 3 illustrates the technology provider’s prof-its and her optimal introduction strategy contingenton the subsystem functionality. Higher functionalitydiminishes the integration cost and the likelihoodof an integration failure. As a result, the providertargets more integrators. However, beyond a certainthreshold, additional functionality leads to such a lowlikelihood of an integration failure that every adopt-ing integrator would be successful in the integrationprocess. As a result, the lower potential differentia-tion between the integrators dilutes the overall valuethe provider may appropriate through a saturationstrategy. Consequently, the provider can offer a sub-system with very high functionality only to a sub-set of integrators so as to induce some downstreamdifferentiation and to retain her value appropriationpower.

In summary, our core results reveal the subtledynamics across upstream and downstream indus-tries. In industrial settings where the downstreamindustry is relatively more competitive, the subsystemfunctionality becomes an important lever that deter-mines the value appropriation power of the upstreamsubsystem provider. Overintegration might lead to adecline of the value appropriation power, and even-tual loss, a situation that echoes our introductoryStainmaster account. However, alternative strate-gies may allow the provider to avoid the over-integration trap. For example, subsystems that affectthe end-product performance, and subsequently theend-product market potential, might still induce


downstream integrators to adopt.11 We explore someof these alternative strategies in the next section.

5. ExtensionsOur analysis thus far has focused on a core insight:the effect of subsystem functionality on the poten-tial for differentiation among the adopters, and con-sequently the value appropriation power that thesubsystem grants to the downstream and upstreamindustries. However, our initial analysis has assumedaway some of the complex realities in several indus-trial contexts. In this section we enrich the basemodel to examine the robustness of our core result.Moreover, we discuss how the value appropriationpower shifts across the upstream and the downstreamindustries.

We examine three important extensions: (i) an in-dustrial setting where subsystem functionality affectsthe end-product performance and therefore the down-stream competition (Assumption B.1); (ii) the casewhere the downstream market may exhibit highervalue appropriation power because of the possibil-ity of a dominant integrator (Assumption B.2); and(iii) the case where technology might be offeredthrough more involved licensing arrangements suchas the volume-based royalty fees often found inpractice (Assumption B.3). Although our extensionsyield greater confidence in the robustness of our coreinsight, they also help delineate the settings wheresubsystem functionality plays a more important rolerather than less important.12

5.1. Value Appropriation: The Role of theSubsystem Performance Contribution

The subsystem functionality may affect the end-product performance. For instance, in integrated cir-cuits a prepackaged number of functions within asubsystem (IC) often results in less power losses,higher transmission speed, and therefore overall bet-ter performance. Consider modifying the base-casemodel by relaxing Assumption B.1 to now assumethat P4f 5 is nondecreasing in f . The assumption thatperformance is nondecreasing in functionality is rela-tively intuitive. In DuPont’s Stainmaster process tech-nology, the explicitly defined (and rigidly controlled)production process implied a better performance forthe end product of the carpet mills. Note that thestructure of P4f 5 allows us to capture the impor-tance of functionality in determining the end-productperformance. When the subsystem functionality has

11 We thank the associate editor for framing these alternatives in asuccinct way.12 Note that we do not relax all assumptions at once. Section 5.1relaxes Assumption B.1, but retains Assumptions B.2 and B.3. Sim-ilarly, §5.2 (§5.3) relaxes only Assumption B.2 (B.3).

minimal impact on end-product performance (e.g., asubsystem with a peripheral role in the end-productarchitecture), then P4f 5 would be nearly constantin f . In contrast, when the subsystem functionalityenhances the end-product performance, we wouldexpect a steeply increasing functional form.

Given that the end-product performance deter-mines the end-product market mechanism, we needto specify in greater detail the effects of P4f 5 on anintegrator’s rents. We assume that the competitionmechanism ç4 · 1 · 5 satisfies the following additionalproperties:

Assumption A.2. ç4Pi1Pj5 is increasing (¡ç/¡Pi > 0)and concave in Pi (¡2ç/¡P

2i < 0).

Assumption A.3. ç4Pi1Pj5 is decreasing (¡ç/¡Pj < 0)and convex in Pj (¡2ç/¡P

2j > 0).

Assumption A.2 states that higher end-product per-formance leads to larger profit. However, this profitexhibits decreasing marginal returns in the perfor-mance. Assumption A.3 states that an integrator’sown profit decreases when his competitor’s end prod-uct realizes superior end-product performance (postintegration). Furthermore, this decrease in profit islikely to flatten out when the competitor’s superiorityis very high. Both of these assumptions are intuitive.13

Finally, the end-product performance affects the rel-ative benefit from adopting the subsystem, as follows:

Assumption A.4. F 4P5−S4P5 is increasing and convexin P .

Assumption A.1 immediately implies that the mar-ginal benefit one obtains from being the only onewith the subsystem (i.e., F 4P5) is greater than themarginal benefit one obtains from the subsystemwhen the competitor adopts the same subsystem(S4P5). Assumption A.4 states that the differenceF 4P5−S4P5 increases in the end-product performance,and that the rate of increase is increasing as well.14

13 It is straightforward to show that these assumptions are satisfiedby many economic models of competition such as quantity-basedcompetition, differentiated Bertrand models of price competition,market-share attraction models, etc. However, it is interesting tonote that these assumptions fail to support one specific type of com-petition mechanism, namely, vertical differentiation models. Wethank one of the reviewers for pointing this out.14 This technical assumption primarily allows mathematicaltractability. It is relatively unimportant for the validity of our mainresults, as we demonstrate through a numerical example in OnlineAppendix 2.1 (available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1381624). Also, a simple argument shows theinternal consistency in our assumptions: suppose that the con-verse of A.4 is true; i.e., F 4P5 − S4P5 is nonincreasing for all P .By definition that F 415 − S415 = 0. Because we have assumed thatF 4P5−S4P5 is nonincreasing, this implies that F 4P5−S4P5≤ 0. How-ever, by Assumption A.1 we know that F 4P5−S4P5≥ 0. This resultsin a contradiction.


The economic intuition behind A.4 is that as end-product performance gets higher there should bemore incentives for an integrator to seek exclusivityin the subsystem usage, i.e., the subsystem becomes abigger contributor to competitive differentiation (Senand Tauman 2007).

Proposition 3. 1. There exist thresholds on end-product performance P 0, P 1, and P 2 such that for P < P 0

no-sale is optimal, for P 0 ≤ P < P 1 a niche strategy is opti-mal, for P 1 ≤ P ≤ P 2 a saturation strategy is optimal, andfor P 2 ≤ P a niche strategy is optimal.

2. There exist thresholds on the subsystem functionality0 ≤ f̂0 ≤ f̂1 ≤ f̂2 ≤ 1 such that the provider finds it optimalto (i) undertake the niche strategy if f ∈ 6f̂01 f̂17 or f ∈6f̂2117, (ii) undertake the saturation strategy if f ∈ 6f̂11 f̂27,and (iii) not offer to any of the integrators if f ∈ 601 f̂07.

The first part of our proposition, unsurprisingly,echoes past findings in the literature (Sen andTauman 2007) regarding the impact of end-productperformance on the introduction strategy and thevalue appropriation power of the upstream provider.When a subsystem offers high performance, theprovider finds it optimal to target more integra-tors. However, when the performance improvement isextremely large, i.e., P � 1, an integrator with exclu-sive access can become a (near) monopolist and dom-inate the end-product market. This insight explainswhy, for extremely large performance improvements,the provider appropriates higher value from licens-ing exclusive rights (niche strategy), and allows theadopter to become a near monopolist.

It is interesting, though, that high subsystem func-tionality via its effect on end-product performancebecomes an attractive design choice. Recall from ourdiscussion in §4 that when functionality affects onlythe integration uncertainty/costs, a high functional-ity (overintegration) diminishes the provider’s valueappropriation power and her revenues, because itdecreases the potential for downstream differenti-ation. Thus, to mitigate the revenue decrease, theupstream provider pursues a niche strategy. Stateddifferently, the upstream provider could benefit fromdecreasing the subsystem functionality and pursuinga saturation strategy. However, when functionalityaffects the end-product performance, a relatively highsubsystem functionality makes the niche strategymore profitable compared to the saturation strat-egy. Higher subsystem functionality disproportion-ately increases the integrator’s benefit from being anear monopolist under a niche strategy (end-productperformance effect) and decreases the likelihood ofbeing a monopolist under a saturation strategy (differ-entiation potential effect).

Therefore, in industrial settings where the sub-system functionality shapes the end-product perfor-mance, and consequently the end-product market

competition, we find that the upstream providermay reduce her value appropriation power by offer-ing over-integrated subsystems. Our finding offersa potential explanation for the severity of the chal-lenges during the DuPont Stainmaster introduction.DuPont insisted on pursuing a saturation strategywith Stainmaster, a technology that offered both supe-rior end-product performance and high subsystemfunctionality; indeed, these are the exact conditionsunder which our results support the superiority ofpursuing a niche strategy.

5.2. Value Appropriation: The Role ofDownstream Integrator Heterogeneity

A reality in many downstream industries is that theintegrators have heterogeneous capabilities (e.g., inte-gration costs), and that some integrators tend to bemore dominant than the rest. We relax Assump-tion B.2 (i.e., = 0), and we assume a nonzerocost differential parameter that captures the down-stream heterogeneity in integration capabilities. Theintegration costs for the two integrators are C4f 5−and C4f 5 + , respectively. We also discuss a dif-ferent source of heterogeneity, namely, an a prioriend-product advantage for the dominant integrator.The existence of a heterogeneous downstream mar-ket allows us to capture settings where the down-stream industry enjoys a higher (relative) power invalue appropriation (settings close to our earlier ref-erence to the IBM Global Services).

The baseline results about the dependence of intro-duction strategy on the functionality (illustrated inFigure 3) remain robust. The following propositionclarifies how the regions become larger or smaller,and how the core trade-off is moderated by hetero-geneity in the downstream market.

Proposition 4. The subsystem introduction strategyand the ex post downstream industry structure are affectedas follows by greater ex ante heterogeneity among down-stream integrators:

• fewer no-sale settings,• more niche settings,• fewer saturation settings.

Heterogeneity, in our model, enhances the integra-tion ability of one of the integrators while reducingthe other’s integration ability. The provider appro-priates more revenues with a harder to integratesubsystem (i.e., lower functionality), because the inte-grator with the lowest cost can still profitably adoptthe subsystem, whereas the weaker integrator cannotpay significant fees for the subsystem. As such, theno-sale region diminishes, and the saturation strat-egy becomes less attractive. Instead, the existenceof a dominant downstream integrator constrains theupstream provider to “do business” only with him,


and it favors lower ranges for the subsystem function-ality. In that regard, the upstream value appropriationpower is reduced. Indeed, our finding is consistentwith observations in technology markets with dom-inant integrators (e.g., IBM Global Services and itsability to integrate multiple externally sourced infor-mation technology modules in its offerings); thesedominant downstream integrators are able to dictatethe level of integration to some of their technologyproviders.15

The previous effect of downstream heterogeneityon the upstream value appropriation power is inde-pendent of the type of heterogeneity: the provideralways ends up transacting with one of the integra-tors. Yet, it is interesting to note that the upstreamprovider may not always benefit from transactingwith the a priori dominant integrator. If the integra-tors exhibit heterogeneous ex ante end-product per-formances (P14 ±C5 6= P24 ±C5), then the provider licensesto the integrator with the lower a priori end-productperformance.16 This intriguing difference is explainedas follows: Under differing integration costs, the morecapable integrator is willing to pay the most forthe subsystem, because he has a lower integrationcost; with end-product performance heterogeneity,the lower capability integrator is willing to pay themost for the subsystem, because his status quo per-formance is lower (i.e., higher marginal benefit fromadoption). Hence, when the heterogeneity is in theintegration costs, the provider should, under a nichestrategy, target the more capable integrator; whereaswhen the heterogeneity is in the end-product perfor-mances, the provider should, under a niche strategy,target the less capable integrator.

Our finding allows us to shed more light on thedrivers of the upstream value appropriation power.Downstream heterogeneity limits the upstream poweras it imposes a niche strategy with an associatedrange for the subsystem functionality (see Figure 3).At the same time, depending on the type of a prioridownstream heterogeneity, we find that the upstreamprovider may either enhance the level of down-stream heterogeneity, or she may reduce it. In the for-mer case, the upstream provider may lose her valueappropriation power in the long run, whereas in thelatter she may level competition downstream andretain her value appropriation power.

15 Needless to say, our result, although consistent with low levels ofintegration found in certain technology markets with large integra-tors, is not the unique explanation. For instance, as one anonymousreviewer pointed out, the sheer heterogeneity of the end-productmarket and the ability of the integrator to pick and choose frommultiple modules to develop a customized end product may havethe same effect.16 A complete proof is available from the authors, but the prooflogic is identical to that of Proposition 4.

5.3. Value Appropriation: The Role ofLicensing Mechanisms

In §4 we assume the provider licenses the subsys-tem for a fixed one-time payment. The assumptionallowed us to abstract away from the complex tech-nology transfer arrangements often found in practice,and isolate the effect of the subsystem’s functionality.However, in actual technology markets, many subsys-tem technology providers employ volume-based fees(royalties) in addition to fixed fees (Rostoker 1984,Arora et al. 2004). In this section, we examine therobustness of our key insights, in the presence ofricher licensing structures.

Recall that the general fee structure is ±W = 6W1w7,where W is the fixed-price component and w theper-unit fee (royalty). We relax Assumption B.3 andwe assume that w ≥ 0. Unlike the sunk-cost natureof the fixed fee W , a nonzero per unit fee w mayaffect the integrator’s revenues ç4 · 1 · 5, because itdirectly affects the per-unit costs incurred (upon suc-cessful integration). Hence, for our analysis to carrythrough, we adjust the underlying notation for thepayoffs ç4 · 1 · 5. Because these changes are for themost part intuitive, we describe them in OnlineAppendix 2.1, and we report here only the key resultswith respect to the introduction strategies and therespective upstream and downstream value appropri-ation power.

Proposition 5. A royalty fee is suboptimal when thetechnology provider undertakes the niche strategy. How-ever, when the technology provider undertakes the satu-ration strategy, she may benefit from charging a royaltyw> 0.

Proposition 5 makes an important point: Anupstream provider will choose to calibrate hervalue appropriation power through complex licens-ing mechanisms only under certain settings. Specif-ically, the provider will never employ a royalty feeunder a niche strategy. Intuitively, under a niche strat-egy, the familiar phenomenon of double marginaliza-tion makes its appearance, and it drives the adopter’srevenue to decrease as the royalty (cost per unit)increases. Thus, the provider finds it beneficial not touse a volume-based fee under a niche strategy.

Volume-based royalties, on the other hand, aredesirable only when the provider intends to offerthe subsystem to multiple integrators. In a settingwhere technology is assumed to reduce manufactur-ing cost, Kamien and Tauman (1986) find that withcomplete information and risk-neutral firms, purefixed-price contracts always dominate pure volume-based royalties. Sen and Tauman (2007) extend thismodel along one important dimension, namely, theyallow mixed license structure with both fixed feesand royalties. They find that a provider may employ


Figure 4 Regions When Fees May Also Include Volume-Based Royalties

Functionality f

Fixed fee Mixed Fixed fee

One integratorTwo integratorsOne integratorNumber of adopters

Fee structure

positive royalties when she licenses a cost-reducinginnovation to many downstream firms. Our resultextends these findings to the case of subsystem tech-nologies that alter the integration risk/costs. It illus-trates the necessity of a saturation strategy for thevolume-based royalties to be optimal. The economicintuition is as follows: Whereas a royalty results indouble marginalization when coupled with a nichestrategy, the same royalty may be used to enhance theprovider’s relative value appropriation power, whencoupled with a saturation strategy. Royalties inflatethe per-unit cost for the entire industry, and thereforeinduces the downstream integrators to adjust theircompetitive actions to retain/increase revenues. Thelatter increase, which may increase overall industryrevenues, is then appropriated by the provider usingfixed fees. This insight has a significant managerialimplication: licensing mechanisms not only serve aslevers to appropriate value, but they can also regulatethe power an upstream industry has to appropriatethe value. Thus, the two mechanisms, volume-basedroyalties and fixed fees, are not substitutes but servecomplementary roles.

The next proposition extends Propositions 2 andconfirms the qualitative insights about the determi-nants of the provider’s optimal introduction strategy.

Proposition 6. The effects of the likelihood of suc-cessful integration (p), the integration costs (C), and thesubsystem functionality (f ) on the optimal number ofintegrators remain unchanged even when the provider hasthe flexibility to use royalties in addition to fixed fees. Theregion defining the saturation strategy becomes larger.

The robustness of the structure of the thresholdsgoverning the optimal introduction strategy (shownin Figures 2 and 3) validates that the key determi-nant of the provider’s optimal introduction strategyis the subsystem functionality and not the type oflicensing fee used. At the same time, some second-order effects exist. Specifically, our results indicatethat royalties result in a larger saturation regionfor the following reason: Recall that for very highfunctionality f , the provider switches away from asaturation strategy to a niche because of the lowerdifferentiation potential associated with the saturation

strategy (see Proposition 2). However, the problemof lower differentiation potential can be mitigatedby employing royalties to moderate the competitionbetween the integrators, and thus to prevent the cre-ated value from dissipating.

Finally, the next proposition characterizes thedependence of royalty on the functionality f . Notethat Propositions 6 and 7 together allow us to charac-terize the optimal introduction strategy (i.e., the num-ber of integrators to whom the provider offers thesubsystem and the type of fee structure she utilizes)contingent on the subsystem functionality. Figure 4graphically illustrates this result.

Proposition 7. There exists thresholds fv and f ′v suchthat volume-based royalties are optimal if and only if f ′v ≥f ≥ fv.

Proposition 7 states that any royalty fee shall beemployed only for moderate values of functionality.The result follows directly from our previous dis-cussion on the differentiation potential. The firm canemploy royalties to mitigate the rent dissipation effect(Arora and Fosfuri 2003). However, beyond somelevel of subsystem functionality, royalties are inad-equate to mitigate rent dissipation, and hence theprovider resorts to restricting access to the technology(i.e., selling to fewer integrators).

6. Discussion and ConclusionsIn this paper, we develop an analytical model thatexamines the optimal introduction strategies for anew subsystem technology in an industrial business-to-business (B2B) setting. Past studies have primarilyfocused on the performance contribution of such sub-system technologies to the end product. We focus ona different dimension, namely, the subsystem func-tionality that may influence both the end-product per-formance and the subsystem integration process. Wedefine the subsystem functionality as the number ofend-product functions included in the subsystem.

Our analysis suggests a strong effect of function-ality to both the introduction strategy and the rel-ative value appropriation power of the upstreamtechnology provider. High functionality has a dualeffect on the provider’s profits. On the positive side,


the provider extracts larger ease-of-integration rentsfrom the adopters. Subsystems are easier to inte-grate because of lower integration costs/uncertainty.On the negative side, the subsystem curtails theintegrator’s ability to differentiate from competition.Stated differently, when the subsystem is introducedthrough mass adoption (i.e., saturation strategy), theprovider finds it beneficial to offer less function-ality. Less functionality allows all the downstreamadopters to sufficiently differentiate from each other.Our result sends a cautionary message to technol-ogy providers on the potential pitfalls of overinte-gration, and highlights the need to conduct a moreholistic assessment—the direct gains from the obvi-ous positive effect and the indirect losses from over-integration—when managing this important designdimension.

The identified trade-off is robust to a set of potentialcircumstances that may alter the appropriation powerof the upstream technology provider. Thus, evenwhen one downstream firm dominates in terms ofeither performance or integration cost, the upstreamprovider needs to calibrate the subsystem functional-ity jointly with the introduction strategy to retain hisvalue appropriation power. In other words, there isalways a sufficiently high level of functionality thatrenders a mass adoption strategy suboptimal.

In addition, our study brings forward two impor-tant findings: first, the role of licensing structuresin managing the influence of subsystem function-ality on downstream competition. Second, the factthat the type of heterogeneity among downstreamintegrators may enable the use of functionality as ameans to diminish or strengthen the upstream appro-priation power. Specifically, the technology providercan potentially undertake a saturation strategy, evenwhen the subsystem functionality is high, with roy-alties that change the cost structure of adoptingfirms. Royalties manage the downstream competition,whereas fixed component of fees allow her to appro-priate the integrator surplus. This result identifies thecomplementary nature of the two value appropria-tion levers (fixed fee and royalties). It adds to theliterature discussion that has identified plausible con-ditions where royalties are optimal in licensing agree-ments (Wang 1998, Sen and Tauman 2007).

Our results also serve to complement recent stud-ies that examine the degree to which upstream deci-sions affect the downstream competition intensity,and therefore the value appropriation power acrossthe different tiers in a value chain. Ghosh and John(2009), in a study examining B2B contexts where orig-inal equipment manufacturers (OEMs) procure sub-systems from outside vendors, find that an OEM isless likely to procure a branded component whenthe component does not allow significant (ex post)

differentiation of the end product. Thus, it appearsthat OEMs that integrate technology subsystems intotheir end products account not only for the end-product performance implications of the subsystem,but also the differentiation-related competitive advan-tage that the subsystem brings once adopted andintegrated.

Although our findings suggest that the trade-off wepropose, and its normative implications, have somedescriptive power, we hasten to add that the generalquestion of whether or not integrators and providersconsider the influence of design factors on their com-petitive advantage can be answered only through anempirical study. The anecdotal evidence we offeredabout DuPont’s Stainmaster technology allows us toconjecture that the theoretical model we have laidout, and the few propositions it yielded, may offera reasonable starting point to undertake an empir-ical examination of how the adoption patterns andthe fee structures depend on the design characteris-tics. Although further work needs to be undertakento turn these implications into testable hypotheses,we summarize the more interesting (in our opinion)propositions emerging from our model:

P1: Functionality has a positive effect on the provider’sprofit if she is a niche player.

P2: Functionality has a weaker effect on the provider’sprofit if she is a saturation player compared to the casewhere she is a niche player.

P3: Royalties would be associated with wide adoption(saturation) strategy.

P4: Controlling for provider’s introduction strategy,royalties would be associated with moderate functionalitysubsystems.

Throughout this study, we have focused on onedesign characteristic—functionality—and modeled itsinfluence on the “integration process specific” valueprovided to an adopter by the subsystem. Still,it is doubtless that other managerial actions (suchas customization) may allow a technology providerto perhaps even selectively affect the value that asubsystem offers. Our model may provide a usefulframework for studying such additional managerialdecisions. Finally, much of the extant literature intechnology introduction, to our knowledge, has usednoncooperative game theoretic frameworks, with theimplicit assumption that either the provider or theadopter have market power (i.e., Stackelberg leaders).Although this assumption does allow one to gain aninitial understanding, given the repeated nature ofinteractions in such technology markets, we believe itis necessary to view both technology introduction anddevelopment as a collaborative endeavor between theprovider and the adopter, and possibly as an outcomeof a negotiated joint decision. These questions awaitfuture research.


AcknowledgmentsThe authors thank Ray W. Miller from Dupont for his timeand insightful comments throughout this study. They alsothank Kamalini Ramdas, one anonymous associate editor,and three anonymous reviewers for their comments thatimproved the paper. Stylianos Kavadias’s current affiliationis Cambridge Judge Business School, University of Cam-bridge, Cambridge CB2 1AG, United Kingdom.

AppendixRecall that we defined the payoffs to the integrator (who isthe row player) contingent on the integration outcome asgiven in Table A.1. For instance, we defined ç4P4C5115, thepayoffs to the row player when the outcome of his integra-tion is success, and the outcome of his competitor’s integra-tion process is failure, as b (i.e., the cell {SUCCESS, FAIL}).

Proof of Proposition 1. Based on Table A.1, we mayevaluate the expected payoffs to integrators conditional ontheir adoption decisions (i.e., decisions whether or not tolicense the subsystem). Table A.2 gives this net expectedpayoff (to the row player). We illustrate this for one of thequadrants of Table A.2.

Consider the value in the cell 8A1A9, which denotes theexpected payoffs to the row player when both he and hiscompetitor decide to adopt the subsystem. Specifically, ifthe integrator and his competitor choose to adopt the sub-system, there are a total of four different possibilities for theintegration outcome: (i) both are successful, which occurswith probability p2 and yields payoff a; (ii) both are unsuc-cessful, which occurs with probability 41 − p52 and yieldspayoff d; (iii) the integrator is successful while his competi-tor fails, which occurs with probability p41 − p5 and yieldpayoffs b; or (iv) the integrator fails while his competitoris successful, which occurs with probability 41 − p5p andyields payoffs c. Hence, the expected payoff is given byp2a + p41 − p5b + 41 − p5pc + 41 − p52d; after netting out thelicense fee W and the integration cost C4f 5, this yields thevalue given in the cell 8A1A9 in Table A.2.

Therefore, once the provider sets a particular fee W ,the two integrators play the adoption game that has theexpected payoffs given in Table A.2. For ease of exposition,denote the value in the cell 8A1A9 as VAA, 8N1A9 as VNA,8A1N9 as VAN , 8N1N9 as VNN , and with some abuse of nota-tion denote C4f 5 simply as C.

Table A.1 Payoffs (to Row Player) After Integration

SUCCESS FAIL

SUCCESS a bFAIL c d

Table A.2 Net Expected Payoffs (to Row Player)

A N

A p2a+ p41 − p54b+ c5 pb+ 41 − p5d −C4f 5−W+ 41 − p52d −C4f 5−W

N pc+ 41 − p5d d

Note. A and N denote the adopt and not-adopt strategies, respectively.

We characterize how the Nash equilibria of the adoptiongame depends on the license fee W next.

If one’s competitor does not adopt, then one’s bestresponse is to adopt iff VAN ≥ VNN , i.e., iff pb + 41 − p5d −C −W ≥ d, i.e., iff W ≤ p4b− d5−C.

Similarly, if one’s competitor adopts, then one’s bestresponse is to adopt iff VAA ≥ VNA, i.e., iff p2a + p41 − p5 ·4b + c5 + 41 − p52d − C − W ≥ pc + 41 − p5d, i.e., iff W ≤p4b− d5− p244b− d5− 4a− c55−C.

Also note that by Assumption A.1, ç4P1P5 − ç411P5 ≤ç4P115−ç41115; i.e., a− c ≤ b− d.

Hence, when W ≤ p4b−d5− p244b−d5− 4a− c55−C, thenregardless of the competitor’s action, one’s dominant strat-egy is to adopt the subsystem, which implies that whenW ≤ p4b − d5 − p244b − d5 − 4a − c55 − C, the unique Nashequilibrium is that both adopt.

Similarly, if p4b − d5 − p244b − d5 − 4a − c55 − C < W ≤p4b − d5 − C, then if the competitor does not adopt, thenthe integrator would prefer to adopt, but if the competi-tor adopts, then the integrator would prefer not to adopt.Hence, there are two Nash equilibria 8A1N9 and 8N1A9.

Finally, when W > p4b − d5 − C, then irrespective of thecompetitor’s action, one’s dominant strategy is not to adoptthe subsystem, which implies that the only Nash equilib-rium is one where neither adopts.

Thus, a profit maximizing provider will set license feeWn = p4b − d5 − C so as to induce only one integratorto adopt (niche strategy), and will set Ws = p4b − d5 − p2 ·44b − d5 − 4a − c55 − C to induce both integrators to adopt(i.e., the saturation strategy).

Proof of Corollary 1.Claim 1: The provider’s revenue under saturation strat-

egy is

s = 24Ws −C5

= 24pF −p24F −S5−C51 where F =b−d and S=a−c

dsdp

= 24F −2p4F −S55

≥ 0 iff p<F

24F −S50

Hence, the revenue under the saturation strategy s4p5 isdecreasing in the likelihood of integration success p forp > F /424F − S55.

Claim 2: The provider’s revenue under saturation strat-egy is

s = 24Ws −C5

= 24pF − p24F − S5−C51 where F = b− d and S = a− c

= 24pF − p24F − S5−K41 −p55

since C4f 5=K41 −4f 55 and p4f 5=4f 5

= 24p4F +K5− p24F − S5−K5

dsdf

= 24F +K − 2p4F − S551

dp4f 5/df

≤ 0 when p >F +K

24F − S51

i.e., when f >−1(

F +K

24F − S5

)

0


Before proving Proposition 2, we state the following sim-ple auxiliary lemma. The proof is trivial and is given inOnline Appendix 2.2.

Lemma 1. If F 4x5 is convex, then there exist x1, x2 such that8F 4x5≤ 09⇔ 8x ∈ 6x11x279.

Proof of Proposition 2. The profits under the satura-tion and niche strategies are given by s and n, respec-tively, where

n = pF −C1

s = 24pF − p24F − S5−C50

Let

G4p1C1P5 = n −s = 2p24F − S5− pF +C0

Then, saturation is superior to niche iff G4p1C1P5≤ 0.Claim 1: To prove the claim, observe that G4p1C1P5 is

convex in p. Hence, by Lemma 1, there exists p1 and p2

such that 8G4p1 0 0 05 ≤ 09 ⇔ p ∈ 6p11 p27. That is, there existsp1 and p2 such that saturation is superior to niche iff p ∈6p11 p27.

Niche is superior to no-sale iff pF −C ≥ 0. That is, thereexists a threshold p0 = C/F such that niche is superiorto no-sale iff p ≥ p0. Furthermore, since pF − C ≥ pF −2p24F − S5−C, if saturation is superior to niche, then nichemust be superior to no-sale. That is, p0 ≤ p1.

Hence, since p0 ≤ p1 ≤ p2, when p < p0, no-sale is superiorto niche, which is superior to saturation; when p0 ≤ p < p1,niche is superior to no-sale, and niche is superior to satura-tion; when p1 ≤ p ≤ p2, saturation is superior to niche, whichis superior to no-sale; and when p > p2, niche is superior tosaturation, and niche is superior to no-sale.

Claim 2: To prove the claim, observe that saturation issuperior to niche iff C ≤ C0, where C0 = pF − 2p24F − S5.Furthermore, niche is superior to no-sale iff pF −C ≥ 0, i.e.,there is a threshold C1 = pF such that niche is superior tono-sale iff C ≤C1. Lastly, observe that C1 ≥C0.

Hence, when C ≤C0, saturation is superior to niche, andniche is superior to no-sale; i.e., when C ≤ C0 saturation isthe optimal strategy. When C0 C1,niche is superior to saturation, and no-sale is superior toniche; hence, no-sale is the optimal strategy.

Claim 3: Recall that we defined

G4p1C1P5=n −s = 2p24F − S5− pF +C0

Saturation is superior to niche iff G4p1C1P5≤ 0.Furthermore, recall that p4f 5 = 4f 5 and C4f 5 =

K41 − 4f 55, where 4 · 5 is a nondecreasing function.Unlike in Claims 1 and 2, we cannot simply apply Lemma 1to find the appropriate thresholds for f because the functionG4· · · 5 may not be convex in f .

However, a change of variable allows us to apply theLemma. Specifically, C4f 5 = K41 − 4f 55 = K41 − p4f 55.Hence, G4p1C1P5 = 2p24F − S5− pF +K41 − p5. This func-tion is convex in p, which implies that saturation is superiorto niche iff p ∈ 6p11 p27. Furthermore, since p = 4f 5 is non-decreasing in f , this implies that saturation is superior toniche iff f ∈ 6−14p151−14p257.

Niche is superior to no-sale iff pF −C ≥ 0. That is, nicheis superior to no-sale iff pF − K41 − p5 ≥ 0. Thus, thereexists a threshold p0 = K/4F +K5 such that niche is supe-rior to no-sale iff p ≥ p0. Furthermore, since pF − C ≥ pF −2p24F − S5−C, if saturation is superior to niche, then nichemust be superior to no-sale. That is, p0 ≤ p1. Last, since p =4f 5 is nondecreasing in f , this implies that niche is supe-rior to no-sale iff f ≥−14p05.

Now as with Claims 1 and 2, we may easily see with f0 =−14p05, f1 = −14p15, f2 = −14p25, that no-sale is optimalwhen f < f0, niche when f ∈ 6f01 f17∪ 6f2117, and saturationwhen f ∈ 6f11 f27.

Proof of Proposition 3.Claim 1: Consider G4P1 · · · 5= p42p4F −S5− F 5+C. Func-

tion −F 4P5 is convex in P since F 4P5 is concave in P(by Assumption A.2). Also, by Assumption A.4, F 4P5−S4P5is convex in P . Hence, G4P1 · · · 5 is convex in P .

Thus, by Lemma 1, there exist P 1 and P 2 such that8G4P1 · · · 5 < 09 ⇔ 8P 1 < P < P 29. That is, saturation strategyis optimal iff P 1

0, and P > P 2 and n > 0. Recall that n = pSpIF 4P5−C,where F 4P5= b4P5− d. By Assumption A.2, b4P5 is increas-ing in P . Hence, there is a threshold P 0 such that 8n >09 ⇔ 8P > P 09. Furthermore, P 1 ≥ P 0 (because at the pointP 1 where saturation becomes preferable to niche, n > 0).Hence, niche is optimal iff P 0

P 2.

The remaining case is P < P 0, for which no-sale isoptimal.

Claim 2: We prove this proposition in three steps.Step 1. We will treat P as a purely exogenous variable,

allowing us to directly use Claim 3 of Proposition 2 andClaim 1 of Proposition 3 to obtain the three thresholds for Pand f denoted by P0, P1, P2 and f0, f1, f2, respectively, suchthat saturation is optimal iff f ∈ 4f11 f25 and P ∈ 4P11P25, no-sale is optimal iff f < f0 and P < P0, and niche is optimalotherwise.

Step 2. In Lemma 3, we will characterize the sensitivityof these thresholds. This allows us to analytically derive theshape of the different optimal regions in the (P1f ) space.

Step 3. We will endogenize the P variable by explicitlyaccounting for the dependence between P and f , and showthat a different set of three thresholds exist for f .

Step 1:When P is assumed to be an exogenous variable inde-

pendent of f , Claim 3 of Proposition 2 and Claim 1 ofProposition 3 directly imply the existence of the requiredthresholds. For completeness, these are restated as the nextlemma.

Lemma 2. 1. There exist thresholds 0 ≤ f04P5 ≤ f14P5 ≤f24P5 ≤ 1 such that the provider finds it optimal to (i) under-take the niche strategy if f ∈ 6f01 f17 or f ∈ 6f2117, (ii) undertakethe saturation strategy if f ∈ 6f11 f27, and (iii) not to offer thesubsystem if f ∈ 601 f07.

2. There exist thresholds 1 ≤ P04f 5≤ P14f 5≤ P24f 5 such thatthe provider finds it optimal to (i) undertake the niche strategy ifP ∈ 6P01P17 or P ∈ 6P215, (ii) undertake the saturation strategyif P ∈ 6P11P27, and (iii) not to offer the subsystem if P ∈ 611P07.


Step 2:

Lemma 3. 1. The thresholds f04P51 f14P51 f24P5 are nonin-creasing in P .

2. The thresholds P04f 51P14f 51P24f 5 are nonincreasing in f .

Proof of Lemma 3. To show that f04P5 is nonincreas-ing in P , consider P2 > P1 and assume the converse, i.e.,let f04P25 > f04P15. Choose any f ∈ 4f04P15, f04P255. Since f >f04P15, the point 6f 1P17 cannot be in the no-sale region.Since f < f04P25, the point 6f 1P27 must be in the no-saleregion (by Lemma 2). Thus, there exists an f such that asP increases (i.e., we go from P1 to P2, where P2 > P1), weswitch into no-sale region from saturation or niche region.This is obviously false by Lemma 2. Hence, f04P25 ≤ f04P15for all P2 >P1; that is f04P5 is nonincreasing.

The proofs for the sensitivity of all other thresholds aresimilar and are available from the authors.

Step 3:In this step, we offer the proof of the main claim

(Claim 3). In Steps 1 and 2 we proved the existence ofthresholds on P and f , and their sensitivity. This allows usto analytically construct the optimality regions in the 4P1 f 5space. They are shown in Figure A.1.

Consider the case where P depends on f , specifically, P =P4f 5 is nondecreasing in f . Now consider the curve in Fig-ure A.1, which plots the function P4f 5. Since P4f 5 is alwaysnondecreasing and because the thresholds on P , namely,P04f 51P14f 51P24f 5 are always nonincreasing (by Lemma 3),this implies that they intersect at most once. Hence, it mustbe the case that when P is endogenous and determined byP4f 5, as f increases, the optimal strategy switches from no-sale to niche to saturation to niche.

Proof of Proposition 4. Recall that represents thedownstream cost heterogeneity. A nearly identical proof asgiven for Proposition 1 can be given to show that Wn = pF −C+ and that Ws = pF −p24F −S5−C−. Hence, the differ-ence between niche and saturation is given by Wn − 2Ws =3+ 2p24F − S5− pF +C.

Because this is increasing in , this implies that niche issuperior to saturation for more cases. In the same manner,

Figure A.1 Regions in the (f 1 P ) Space and the Functi

Documents

The Pitfalls of Subsystem Integration: When Less Is MoreStylianos Kavadias, Cheryl Gaimon Scheller College of Business, Georgia Institute of Technology, Atlanta, Georgia 30332 {[email protected],