Flexibility Versus Efficiency? A Case Study of Model Changeovers …faculty.marshall.usc.edu/Paul-Adler/research/FlexVsEff-2.pdf · 2019. 5. 3. · broader product line was signiﬁcantly

Flexibility Versus Efficiency? A Case Study of Model Changeovers in the Toyota Production System

by

Paul S. Adler Barbara Goldoftas David I. Levine

Downloaded from http://www-bcf.usc.edu/~padler/

1047-7039/99/1001/0043/$05.00Copyright q 1999, Institute for Operations Researchand the Management Sciences ORGANIZATION SCIENCE/Vol. 10, No. 1, January–February 1999 43

Flexibility Versus Efficiency? A Case Studyof Model Changeovers in the Toyota

Production System

Paul S. Adler • Barbara Goldoftas • David I. LevineSchool of Business Administration, University of Southern California, Los Angeles, California 90089-1421

Program in Writing and Humanistic Studies, Massachusetts Institute of Technology,Cambridge, Massachusetts 02139

Haas School of Business, University of California, Berkeley, California 94720

This is a careful and insightful case study of how the Toyota Production System manages the paradoxof efficiency and flexibility, which arises periodically in connection with model changeovers. The

authors detail the functioning of four organizational mechanisms—metaroutines, partitioning, switching,and ambidexterity. However, of particular interest is the contextual reinforcing role of training and trustin administrative structures, procedures, and rules.

Arie Y. Lewin

AbstractThis article seeks to reconceptualize the relationship betweenflexibility and efficiency. Much organization theory argues thatefficiency requires bureaucracy, that bureaucracy impedes flex-ibility, and that organizations therefore confront a tradeoff be-tween efficiency and flexibility. Some researchers have chal-lenged this line of reasoning, arguing that organizations canshift the efficiency/flexibility tradeoff to attain both superiorefficiency and superior flexibility. Others have pointed out nu-merous obstacles to successfully shifting the tradeoff. Seekingto advance our understanding of these obstacles and how theymight be overcome, we analyze an auto assembly plant thatappears to be far above average industry performance in bothefficiency and flexibility. NUMMI, a Toyota subsidiary locatedin Fremont, California, relied on a highly bureaucratic organi-zation to achieve its high efficiency. Analyzing two recent ma-jor model changes, we find that NUMMI used four mechanismsto support its exceptional flexibility/efficiency combination.First, metaroutines (routines for changing other routines) fa-cilitated the efficient performance of nonroutine tasks. Second,both workers and suppliers contributed to nonroutine taskswhile they worked in routine production. Third, routine andnonroutine tasks were separated temporally, and workersswitched sequentially between them. Finally, novel forms oforganizational partitioning enabled differentiated subunits towork in parallel on routine and nonroutine tasks. NUMMI’ssuccess with these four mechanisms depended on several fea-

tures of the broader organizational context, most notably train-ing, trust, and leadership.(Flexibility; Bureaucracy; Tradeoffs; Routines; Metar-outines; Ambidexterity; Switching; Partitioning; Trust)

IntroductionThe postulate of a tradeoff between efficiency and flexi-bility is one of the more enduring ideas in organizationaltheory. Thompson (1967, p. 15) described it as a central“paradox of administration.” Managers must choose be-tween organization designs suited to routine, repetitivetasks and those suited to nonroutine, innovative tasks.However, as competitive rivalry intensifies, a growingnumber of firms are trying to improve simultaneously inefficiency- and flexibility-related dimensions (de Meyeret al. 1989, Volberda 1996, Organization Science 1996).How can firms shift the terms of the efficiency-flexibilitytradeoff?

To explore how firms can create simultaneously su-perior efficiency and superior flexibility, we examine anexceptional auto assembly plant, NUMMI, a joint ventureof Toyota and GM whose day-to-day operations were un-

PAUL S. ADLER, BARBARA GOLDOFTAS AND DAVID I. LEVINE Flexibility Versus Efficiency?

44 ORGANIZATION SCIENCE/Vol. 10, No. 1, January–February 1999

der Toyota control. Like other Japanese auto transplantsin the U.S., NUMMI far outpaced its Big Three counter-parts simultaneously in efficiency and quality and inmodel change flexibility (Womack et al. 1990, BusinessWeek 1994).

In the next section we set the theoretical stage by re-viewing prior research on the efficiency/flexibility trade-off. Prior research suggests four mechanisms by whichorganizations can shift the tradeoff as well as some po-tentially serious impediments to each mechanism. Wethen describe our research methods and the NUMMI or-ganization. The following sections first outline in sum-mary form the results of this investigation, then providethe supporting evidence in our analysis of two majormodel changeovers at NUMMI and how they differedfrom traditional U.S. Big Three practice. A discussionsection identifies some conditions underlying NUMMI’ssuccess in shifting the tradeoff and in overcoming thepotential impediments to the four trade-off shifting mech-anisms.

Flexibility Versus Efficiency?There are many kinds of flexibility and indeed a sizableliterature devoted to competing typologies of the variouskinds of flexibility (see overview by Sethi and Sethi1990). However, from an organizational point of view,all forms of flexibility present a common challenge: ef-ficiency requires a bureaucratic form of organization withhigh levels of standardization, formalization, specializa-tion, hierarchy, and staffs; but these features of bureau-cracy impede the fluid process of mutual adjustment re-quired for flexibility; and organizations therefore confronta tradeoff between efficiency and flexibility (Knott 1996,Kurke 1988).

Contingency theory argues that organizations will bemore effective if they are designed to fit the nature oftheir primary task. Specifically, organizations shouldadopt a mechanistic form if their task is simple and stableand their goal is efficiency, and they should adopt an or-ganic form if their task is complex and changing and theirgoal is therefore flexibility (Burns and Stalker 1961). Or-ganizational theory presents a string of contrasts reflect-ing this mechanistic/organic polarity: machine bureaucra-cies vs. adhocracies (Mintzberg 1979); adaptive learningbased on formal rules and hierarchical controls versusgenerative learning relying on shared values, teams, andlateral communication (McGill et al. 1992); generalistswho pursue opportunistic r-strategies and rely on excesscapacity to do well in open environments versus special-ists that are more likely to survive in competitive envi-ronments by pursuing k-strategies that trade less flexibil-ity for greater efficiency (Hannan and Freeman 1977,

1989). March (1991) and Levinthal and March (1993)make the parallel argument that organizations mustchoose between structures that facilitate exploration—thesearch for new knowledge—and those that facilitate ex-ploitation—the use of existing knowledge.

Social-psychological theories provide a rationale forthis polarization. Merton (1958) shows how goal dis-placement in bureaucratic organizations generates rigid-ity. Argyris and Schon (1978) show how defensivenessmakes single-loop learning—focused on pursuing givengoals more effectively (read: efficiency)—an impedimentto double-loop learning—focused on defining new taskgoals (read: flexibility). Thus, argues Weick (1969), ad-aptation precludes adaptability.

This tradeoff view has been echoed in other disciplines.Standard economic theory postulates a tradeoff betweenflexibility and average costs (e.g., Stigler 1939, Hart1942). Further extending this line of thought, Klein(1984) contrasts static and dynamic efficiency. Opera-tions management researchers have long argued that pro-ductivity and flexibility or innovation trade off againsteach other in manufacturing plant performance(Abernathy 1978; see reviews by Gerwin 1993, Suarez etal. 1996, Correa 1994). Hayes and Wheelwright’s (1984)product/process matrix postulates a close correspondencebetween product variety and process efficiency (seeSafizadeh et al. 1996).

Strategy researchers such as Ghemawat and Costa(1993) argue that firms must chose between a strategy ofdynamic effectiveness through flexibility and static effi-ciency through more rigid discipline. In support of a keycorollary of the tradeoff postulate articulated in the or-ganization theory literature, they argue that in general theoptimal choice is at one end or the other of the spectrum,since a firm pursuing both goals simultaneously wouldhave to mix organizational elements appropriate to eachstrategy and thus lose the benefit of the complementaritiesthat typically obtain between the various elements of eachtype of organization. They would thus be “stuck in themiddle” (Porter 1980).

Beyond the Tradeoff?Empirical evidence for the tradeoff postulate is, however,remarkably weak. Take, for example, product mix flexi-bility. On the one hand, Hayes and Wheelwright (1984)and Skinner (1985) provide anecdotal evidence that morefocused factories—ones producing a narrower range ofproducts—are more efficient. In their survey of plantsacross a range of manufacturing industries, Safizadeh etal. (1996) confirmed that in general more product varietywas associated with reliance on job-shop rather continu-ous processes.


ORGANIZATION SCIENCE/Vol. 10, No. 1, January–February 1999 45

On the other hand, Kekre and Srinivasan’s (1990) studyof companies selling industrial products found that abroader product line was significantly associated withlower manufacturing costs. MacDuffie et al. (1996) foundthat greater product variety had no discernible affect onauto assembly plant productivity. Suarez et al. (1996)found that product mix flexibility had no discernible re-lationship to costs or quality in printed circuit board as-sembly. Brush and Karnani (1996) found only three outof 19 manufacturing industries showed statistically sig-nificant productivity returns to narrower product lines,while two industries showed significant returns to broaderproduct lines. Research by Fleischman (1996) on em-ployment flexibility revealed a similar pattern: within 2-digit SIC code industries that face relatively homoge-neous levels of expected volatility of employment, theemployment adjustment costs of the least flexible 4-digitindustries were anywhere between 4 and 10 times greaterthan the adjustment costs found in the most flexible 4-digit industries.

Some authors argue that the era of tradeoffs is behindus (Ferdows and de Meyer 1990). Hypercompetitive en-vironments force firms to compete on several dimensionsat once (Organization Science 1996), and flexible tech-nologies enable firms to shift the tradeoff curve just asquickly as they could move to a different point on theexisting tradeoff curve. Echoing this thesis at a moretheoretical level, Nelson and Winter (1982) argue that theconcept of a production possibilities frontier is itself mis-leading because in the real world, production technolo-gies are largely tacit and managerial rationality is essen-tially bounded, and therefore moving up to or beyond ahypothetical frontier represents a challenge that is quali-tatively no different from moving to a different pointalong such a frontier.

In response to these empirical findings and theoreticalarguments, other researchers argue that while the tradeoffcan be shifted, much of what we observe when firmsmake notable improvement in several dimensions at oncerepresents catching up to best practice (Skinner 1996,Hayes and Pisano 1996, Clark 1996). Pushing the bestpractice frontier, they argue, is a far more difficult task,since tradeoffs are inevitable when organizations mustmake difficult-to-reverse commitments in plants, equip-ment, and capabilities in their implementation of a givenstrategy.

We conclude from these statistical results and theo-retical debates that if there is a tradeoff between effi-ciency and flexibility among the average performers inan industry, at any point in time some firms are belowand others above this tradeoff line. It is not difficult tosee how firms might find themselves below this line; but

what do firms need to do to position themselves aboveit? More intriguingly, how can exceptional firms shift be-yond the tradeoff experienced by even their strongest ri-vals?

This question is not new in organizational theory: thechallenge of simultaneously performing both routine andnonroutine tasks has been explored in a number of studiesof the “ambidextrous” organization (Duncan 1976,McDonough and Leifer 1983, Jelinek and Schoonhoven1993, Tushman and O’Reilly 1997; see also overview inDaft 1998). Reviewing these studies and culling indica-tions from the broader literature suggests four kinds oforganizational mechanisms that can help shift the trade-off. Metaroutines systematize the creative process. Jobenrichment enables workers to become more innovativeand flexible even in the course of their routine tasks.Switching differentiates roles for dealing with the twokinds of tasks, thus allowing workers time to focus oneach. And partitioning differentiates structures for deal-ing with each kind of role, and the resulting specializationpermits subunits to refine their capabilities in each activ-ity, and permits routine and nonroutine activities to becarried out simultaneously in parallel. We first describethese mechanisms, then identify in the organizational re-search literature several potentially important limitationsof each.

First, the cost of output flexibility can be reduced if theassociated internal processes can be made more routine.Schumpeter (1976) argued that large corporations had acompetitive advantage over smaller firms because theywere able to routinize the innovation process at least tosome extent. The operations management literature hasargued that flexible computer-based automation(computer-numerically controlled machine tools, com-puter aided manufacturing, flexible manufacturing sys-tems, etc.) reduces the cost of high product variety (Adler1988). Strategies such as mass customization (Pine 1993)that are based on economies of scope (Goldhar andJelinek 1983) and economies of substitution (Garud andKumaraswamy 1995) rely on metaroutinization as ef-fected through greater product modularity (specifyingstandardized interfaces), parts standardization, orsoftware-based process automation. The underlying in-sight here is that organizations can develop metaroutinesboth for changing among established routines and for in-venting new routines (Nelson and Winter 1982, Volberda1996, Grant 1996).

Metaroutines shift the tradeoff by transforming non-routine into more-routine tasks; but organizations canalso become more ambidextrous by developing their in-novativeness in nonroutine tasks without impairing theirefficiency in routine tasks. In the literature to date, we



have identified three mechanisms that facilitate this kindof ambidexterity. They can be arrayed from more micro-scopic to more macroscopic in scale: enrichment, switch-ing, and partitioning.

First, workers’ routine production tasks can be enrichedto include improvement as well as efficiency goals.Langer (1989) and Schon (1983) discuss the differencebetween mindless and mindful performance of routinework (see also Louis and Sutton 1991). Victor et al.(forthcoming) show that in a TQM environment, produc-tion workers doing their regular production work taskscan be attentive simultaneously to the efficient imple-mentation of routine production procedures and to thenonroutine task of identifying improvement opportuni-ties. These workers may not sit down to document a sug-gestion until the shift is over, but much of the requisitediscovery and analysis can be done on the job.

Second, work can be organized so that people switchsequentially between the two types of tasks rather thanattempting to do them both simultaneously. As comparedto enrichment, switching allows greater focus and reducesthe risk of confusion. Such switching can be supportedby “parallel” organizational structures such as quality cir-cles. These structures enable people to move back andforth between a bureaucratic structure for the routinetasks and a more organic structure for the nonroutinetasks (on parallel structures, see Bushe and Shani 1991,Stein and Moss Kanter 1980, Miller 1978; for closelyrelated concepts, see also Zand 1974 on collateral orga-nizations, Goldstein 1985 on dualistic organizations, andNonaka and Takeuchi 1995 on hypertext organizations).

Finally, ambidexterity can be supported on an evenmore macroscopic scale if the organization as a wholepartitions itself to allow some subunits to specialize inroutine tasks while other subunits specialize in nonroutinetasks. Partitioning has some advantages over enrichingand switching, because enriching and switching do notafford as much opportunity to deepen skills by speciali-zation, and do not allow sustained, focused efforts di-rected at the two types of tasks to proceed simultaneously.As the relative importance of nonroutine tasks increases,it becomes more cost-effective for the firm to partition itsbasic structure into separate units that can be staffed byspecialized personnel and structured and managed differ-ently to assure their optimal performance (Lawrence andLorsch 1967). Miles and Snow’s (1978) Analyzer type offirm uses this mechanism to compete on both efficiencyand innovation fronts against the organizationally simplerDefender and Prospector types. If partitioning is to en-hance flexibility without great loss of efficiency, the dif-ferentiated subunits must effectively coordinate and in-tegrate their efforts. Organization theory suggests when

one or both of the subunits’ tasks are nonroutine, thiscoordination and integration cannot rely on the core bu-reaucratic mechanisms of hierarchy and standardization,but it can in principle be achieved using the mechanismof mutual adjustment in lateral relations between the sub-units (Thompson 1967, Mintzberg 1979).1

Potential ImpedimentsPrior research has recognized that these four mecha-nisms—metaroutines, enriching, switching, and partition-ing—face a number of potentially serious intrinsic im-pediments. Taking metaroutines first, we need todistinguish the cases where metaroutines are embodied incomputer software from those where they are embodiedin employees’ work. Research has reached a rather strongconsensus on the changes in organizational form that al-low software-based automation to shift the tradeoff: be-cause such automation reduces the routineness of thetasks left to workers, the appropriate organizational struc-ture shifts towards the organic (Zammuto and O’Connor1992). But research is far less definitive on cases wherefirms seek to shift the tradeoff frontier by organizationalmeans rather than by advanced automation. Our theoreti-cal analysis and our fieldwork reported below thereforefocus on such cases, and here the literature suggests anumber of factors that make shifting the tradeoff prob-lematic.

Insofar as it affects tasks directly rather than being em-bodied in automation, metaroutinization is associatedwith two complementary impediments. On the one hand,powerful psychological forces encourage “goal displace-ment” so that conformance to the standardized proceduresbecomes the over-riding goal, and the remaining nonrou-tine tasks are ignored (Merton 1958). This defensivenesswill be amplified if, as is often the case, the results of theroutine tasks are more easily measured and are the focusof reward systems. When employees cling to existing rou-tines—and metaroutines, while meta, are still routines—they will see the introduction of new routines as a threat.As a result, metaroutines can lead to organizational rigid-ity.

On the other hand, metaroutines, like routines moregenerally, reduce task autonomy and variety compared topure, unfettered innovation. Thus, metaroutines may re-duce the intrinsic motivational quality of the innovationprocess (Hackman and Oldham 1980). For example,skilled trades maintenance workers might resent the stan-dardization of their creative, problem-solving processwhen formalized “total preventive maintenance” proce-dures are introduced. Design engineers might resent de-tailed procedures specifying tasks that must be performedat each stage of the design process. While some employ-ees might welcome routinization as a relief from the



stresses created by the chaotic quality of the nonroutine,others—perhaps those with higher growth needsstrength—might resist it. Assuming that resistance is themore likely response, some theorists have argued thatroutinization typically needs to be imposed on employeesby management via the efforts of staff specialists(Mintzberg 1979). The demotivating effects of such co-ercive routinization might not be too costly to the firm ifthe resulting tasks were entirely stable and routine, sincein such contexts a passively acquiescent workforce mightsuffice (as in Bowen and Lawler’s (1992) assembly-linemodel). However, a firm relying on its employees to con-tribute to ongoing innovation and learning would surelyhesitate before adopting such an approach.

While a long line of researchers have argued for theeffectiveness of job enrichment—primarily on thegrounds that it strengthens motivating job characteris-tics—other researchers have argued that enrichment hasa number of potentially important intrinsic limitations.First, it is costly, since it requires more training of pro-duction workers: when the core tasks are routine, theskills required for nonroutine tasks will go relatively un-used for lengthy periods. Second, enrichment, at least asit is usually interpreted, is typically paired with job en-largement—enrichment adds new “vertical” tasks whileenlargement adds new “horizontal” tasks to jobs in theroutine production domain. Such job redesign often mil-itates against precise conformance with standardized pro-duction procedures, since cycle times become longer andworkers have more autonomy in deciding work methodsand pacing. Conformance, however, may be necessary forefficiency and indeed for organizational learning in rou-tine task environments (Adler and Cole 1993). Third, byintroducing nonroutine tasks into the workers’ job de-scription, enrichment reduces the programmability andobservability of tasks, creating information asymmetriesand increasing opportunities for opportunistic behavior;this increases agency costs (Eisenhardt 1989). These threelimitations imply that enrichment may simply shift theorganization along the existing tradeoff curve—towardsmore flexibility at the cost of reduced efficiency—ratherthan shifting the tradeoff curve itself.

Switching too encounters potentially serious impedi-ments. Systematic investigations of the effectiveness ofswitching structures such as quality circles have revealedmixed results (see Ledford et al. 1988). A disturbing pro-portion of such structures do not yield the expected per-formance or attitudinal gains, and their mortality rate ishigh (Lawler and Mohrman 1985). The underlying inhib-iting factors reflect a basic tension between the two roles.The efficient performance of routine tasks requires thesupport of a mechanistic organizational form, one where

workers are often assumed to experience work as a dis-utility and managers must therefore exercise close super-vision of detailed prescriptive procedures (Bowen andLawler 1992). The creative performance of nonroutinetasks requires the support of an organic form, where em-ployees are assumed to be intrinsically motivated andmanagers are urged to function as facilitators andcoaches. The compliance-oriented, Theory X style ofmanagement required in routine tasks conflicts with thehigh commitment, Theory Y style required in innovativetasks (Walton 1985, McGregor 1960). Duncan argues thatemployees who have been involved in the nonroutinetasks associated with innovation “are likely to initiallyresist more centralization in rules and procedures and de-cision making”—organizational changes necessary forhigh performance in the more routine, implementationstage (Duncan 1976, p. 180).

Finally, several strands of research suggest that the sub-unit partitioning approach may not be able to shift thetradeoff, since the coordination and integration of differ-entiated subunits requires extensive and expensive man-agement effort. The coordination of tasks across sepa-rately managed subunits requires more planning andmanagement attention. It is difficult and costly to recon-cile the different “thought worlds” that arise in the dif-ferentiated subunits (Dougherty 1992). The maintenanceof different organization structures each with its own pol-icies and practices leads to additional organizational over-head (Bowen and Lawler 1995, p. 78). Partitioning maylead to parochial, self-interested subunit behavior, mul-tiplying the overhead required to reconcile intraorgani-zational conflict (Pfeffer 1978, Duncan 1976). New man-agement positions have to be added to coordinate andintegrate the differentiated subunits and resolve their con-flicts (Lawrence and Lorsch 1967). For all these reasons,Mitzberg (1979, pp. 340–342) argues that mutual adap-tation between differentiated units often fails in machine-bureaucracies.

Contextual FactorsOur presentation of the prior research as summarized inthe literature review above is framed in terms of mecha-nisms and their intrinsic limitations; other strands of re-search can be interpreted as suggesting that the organi-zational context will determine the relative balance of theforces mobilized by the mechanisms and those associatedwith their intrinsic limitations. Organizational theory haslong argued that formal structures and processes are al-ways embedded in—and their effectiveness conditionedby—a broader organizational context of culture and lead-ership (see e.g., Scott 1992).



Prior research on ambidextrous, tradeoff-shifting or-ganizations has identified some of these contextual fac-tors. Tushman and O’Reilly (1997) analyze several am-bidextrous organizations that are able simultaneously topursue evolutionary and revolutionary innovation. Theysee the key sources of ambidexterity in: (a) a decentral-ized structure (read: partitioning) in which headquartersfunctions as a facilitator rather than as a “checker andcontroller,” (b) a common, underlying layer of strong cul-ture and vision which is complemented by another layerof culture that is differentiated between evolutionary andrevolutionary parts of the organization (read: partitioning)or between the corresponding phases of activity (read:switching), and (c) supportive leaders and flexible man-agers. They thus show that distinctive values, culture, andleadership are essential contextual conditions for ambi-dexterity. Jelinek and Schoonhoven’s (1993) study ofhigh-tech firms that excelled at both efficient productionand product technology innovation highlights a similarset of contextual factors.

These two studies are rich in insights, but have notgenerated an overarching theory. In the absence of sucha theory, and with the goal of moving closer towards one,this study therefore adopts an inductive, theory-buildingapproach rather than a deductive, theory-testing ap-proach.

Research Context and MethodsA case study of an organization that excels in both effi-ciency and flexibility dimensions can advance our under-standing of these hypothesized mechanisms and impedi-ments and of how the organizational context influencestheir relative effects. NUMMI was one such organization.In analyzing NUMMI’s flexibility, we focus on its agilityin major model changes.

Research Context: Model Changeovers in the AutoIndustryBeginning in the 1970s, the auto industry “dematured”(Abernathy et al. 1983). Whereas the bases of competitionin the U.S. during the prior period were price and cos-metic styling, the new epoch brought ferocious compe-tition from Japanese manufacturers who shifted consumerexpectations concerning price and conformance-typequality while simultaneously differentiating productsthrough design and technology. Products thus changedmore rapidly and the changes were more substantial. Mi-nor cosmetic model changes still occurred each year, butthe frequency of major model changes—and the extentof product and process change associated with them—increased. Whereas the interval between major modelchanges in the U.S. Big Three auto companies had varied

between four and eight years, competitive pressure led toa shift toward the lock-step four-year cycle adopted byJapanese companies and their U.S. subsidiaries.

These major model changes represented a huge chal-lenge to an auto assembly plant. Anywhere between 60%and 90% of the 1500–2000 components that were assem-bled into a vehicle were redesigned, and as a result, mostof both internal and supplier manufacturing processeswere redesigned too. The ability of a manufacturing plantto introduce new models—the time and cost required to“ramp up” production of the new model to targeted qual-ity and efficiency levels—thus represented a form of flex-ibility that had considerable and growing strategic sig-nificance (Gerwin 1993, p. 398).2

NUMMI operated at exceptional levels of productivityand quality. In 1993, for example, NUMMI took around18 person-hours to assemble a vehicle, as compared to anaverage of 22 hours in a large sample of Big Three plants(Pil and MacDuffie 1996: note that their analysis controlsfor numerous factors that can distort interplant compari-sons). And in the J. D. Power and Associates’ InitialQuality Survey that year, NUMMI’s three main productsscored either first or second place in their respective mar-ket segments: the Corolla had 82 problems per 100 ve-hicles, the Prizm had 87, and the Tacoma truck had 77,compared to an auto industry average of 107.

At the same time, NUMMI was significantly more flex-ible than its Big Three counterparts. Whereas Ford,Chrysler, and GM model changes in 1994 involved plantclosures of 60, 75, and 87 working days respectively,NUMMI was closed for only five days for comparablycomplex major model changeovers in 1993 and 1995.Whereas the typical Big Three plant often took sixmonths to resume normal production rates after a majormodel change, NUMMI took less than four months forthe 1993 changeover and less than three months for the1995 changeover. Moreover, quality at Big Three plantstypically degraded considerably at resumption of opera-tions: during the 1987–1995 period, J. D. Power InitialQuality data shows that the average number of problemsper 100 domestic model vehicles went from 135 in theyear prior to model change to 144 problems in the yearof the model change. And the typical Big Three plant onlyreturned to its normal quality level after a period lastinganywhere from three months to over a year. By contrast,NUMMI’s 1993 model introduction took only a fewweeks to recover from a small slip from world-class lev-els: from 87 to 89 for the Prizm, from 82 to 85 for theCorolla. (Note that the J. D. Power and Associates dataare collected on vehicles sold in the September throughDecember period each year. Since most model changesoccur in August, the annual J. D. Power and Associates



scores are thus very sensitive to quality during the ramp-up of new models.)

As we show below, one of the key factors explainingNUMMI’s agility was its use of a Pilot Team composedof production workers. This Pilot Team designed the newproduction process, suggested changes to product designto facilitate production, and trained line workers for theirnew assignments. By contrast, the traditional Big Threechangeover process left the design of the new product andproduction process to engineers and managers. In recentyears, the Big Three have tried to become more agile, andin each company there had been a small number of pro-jects that imitated the Japanese Pilot Team approach.Some of these projects have rivaled NUMMI’s change-over performance. However, the agility shown in theseexceptional projects had not yet become the norm withinany of the Big Three (Business Week 1994).

Research MethodsIn order to understand how NUMMI achieved its excep-tional efficiency and flexibility, we conducted approxi-mately 60 interviews during 1993 and 1994. Our inform-ants came from NUMMI, UAW Local 2244, andCal-OSHA (California Occupational Safety and HealthAgency). Using a snowball approach, we interviewed in-dividuals from all ranks of the company, including pro-duction workers, skilled trades workers, Team Leaders,Group Leaders, Assistant Managers, Managers, and se-nior executives. Interviewees came from a variety offunctions, including assembly, the model change PilotTeam, quality engineering, assembly engineering, laborrelations, safety, and training. We interviewed union of-ficials from both the contending factions within Local2244 (the Administration Caucus and the People’s Cau-cus). As we explain below, the 1993 model changeoveroccasioned citations by Cal-OSHA—citations that wereappealed by NUMMI and subsequently settled—and as aresult, we also interviewed officials at Cal-OSHA. Infor-mation on Toyota’s model change process was alsodrawn from interviews with managers, engineers, andworkers at TMMK, Toyota’s subsidiary in Georgetown,Kentucky.

Interviews were semistructured, and each lasted 30 to60 minutes. Key informants were interviewed at greaterlength and in some cases several times. Most interviewswere conducted by at least two researchers, and they weretaped and transcribed. Unless otherwise noted, quotationsbelow come from these interviews. All informants wereasked about their work history, their roles in changeovers,and their experience of changeovers. Informants werealso asked about specific topics on which they had spe-cialized knowledge, such as facets of organizational his-tory or policies.

Many informants had previously worked for Big Threeauto companies, and their responses formed the basis ofour characterization of the traditional Big Three change-over practices. We found no systematic studies of BigThree changeover practices outside the short descriptionsin Clark and Fujimoto (1991) and Clark et al. (1992).Concerned that our informants may have been biased, wesubmitted drafts of our characterization of these practicesfor review by several managers currently with Big Threecompanies: no bias surfaced.

To supplement these interviews, we examined a broadrange of company and union documents, union newspa-pers, minutes from union-management meetings, andtraining materials for workers and managers. We alsostudied the materials used to support and contest the Cal-OSHA citations filed at the state Department of IndustrialRelations in Oakland, California. We relied on previouslypublished studies and our own earlier rounds of inter-views for descriptions of NUMMI’s earlier years.

Shifting the Tradeoff: Key FindingsIn accordance with the canons of inductive research, wepresent our results first, in this section, then present thesupporting evidence in the following three sections. Ourresults fall under two broad headings: mechanisms andcontext.

A higher-order result should, however, be noted first.The literature reviewed above focuses on the internal or-ganization of the firm; but we found that NUMMI’s am-bidexterity was very dependent on the nature of its sup-plier relations. The same four generic mechanisms wereused by NUMMI in its supplier relations, and their effec-tiveness was conditioned by the broader context formedby the fabric of NUMMI’s supplier relations. We there-fore weave our discussion of the role of supplier relationsinto the more general storyline below.

Tradeoff-Shifting MechanismsOur first set of results are summarized in Exhibit 1. Thefirst and third columns of Exhibit 1 summarize the dis-cussion above of tradeoff-shifting mechanisms and theirpossible impediments. The second column lists the vari-ous mechanisms that allowed NUMMI to shift the termsof the tradeoff relative to traditional Big Three practice.

First, NUMMI had many more metaroutines than tra-ditional Big Three plants to guide the performance andincrease the efficiency of nonroutine activities. Standard-ized problem-solving procedures facilitated continuousimprovement efforts in regular production. Accumulateddocumentation of changeover experiences facilitated the



Exh

ibit

1K

eyF

ind

ing

s

Trad

e-of

fS

hifti

ngM

echa

nism

NU

MM

ICha

ngeo

verM

echa

nism

sP

ossi

ble

Imp

edim

ents

Fact

ors

atN

UM

MIM

itig

atin

gP

ossi

ble

Imp

edim

ents

ME

TAR

OU

TIN

ES

:S

tand

ard

ized

pro

ced

ures

for

chan

gin

gex

istin

gro

utin

esan

dfo

rcr

eatin

gne

won

es

•P

rob

lem

-sol

ving

pro

cess

isst

and

ard

ized

insi

x-st

epp

roce

dur

e•

Pilo

tTea

mre

lies

onex

tens

ive

doc

umen

tatio

n•

Cha

ngeo

verp

roce

ssha

sd

ocum

ente

dte

mp

late

for

sup

plie

rs•

Han

sei(

refle

ctio

n-re

view

)pro

cess

isst

and

ard

ized

•R

outin

izat

ion

red

uces

task

varie

tyan

dau

tono

my

com

par

edto

unco

nstr

aine

dcr

eativ

ity,a

ndth

eref

ore

red

uces

intr

insi

cm

otiv

atio

n,w

hich

crea

tes

resi

stan

ceor

red

uces

com

mitm

ent

•If

emp

loye

esre

sist

it,ro

utin

izat

ion

mus

tb

eim

pos

edb

ym

anag

emen

tand

staf

f,fu

rthe

rre

duc

ing

com

mitm

ent

•W

orke

rsp

artic

ipat

ein

stan

dar

diz

atio

np

roce

sses

•C

ontr

olov

erth

eou

tcom

esof

stan

dar

diz

edp

roce

sses

issh

ared

•W

ell-d

esig

ned

met

arou

tines

pro

vid

est

ruct

ure

and

role

clar

ityth

atar

ese

enas

usef

ulin

per

form

ing

nonr

outin

eta

sks

•Tr

ustb

ysu

bor

din

ates

that

rout

ines

will

be

used

as“t

ools

”no

t“w

eap

ons”

EN

RIC

HM

EN

T:A

dd

nonr

outin

eta

sks

toro

utin

ep

rod

uctio

nta

sks

•K

aize

nis

wor

ker’s

resp

onsi

bili

ty:

•D

urin

gjo

bd

esig

np

roce

ss(s

tand

ard

ized

wor

k)•

Dur

ing

reg

ular

pro

duc

tion

•D

urin

gac

cele

ratio

n•

Kai

zen

issu

pp

liers

’res

pon

sib

ility

•D

urin

gco

ntra

ctp

erio

d•

Bet

wee

nco

ntra

cts

•Tr

aini

ngis

cost

lyan

dsk

illon

lyra

rely

used

,so

effic

ienc

yis

lost

•A

ssoc

iate

dho

rizon

talj

oben

larg

emen

tre

duc

esco

nsis

tenc

y•

Enr

ichm

entg

ives

mor

eau

tono

my

and

ther

efor

ein

crea

ses

the

risk

ofop

por

tuni

sm

•C

omp

lem

enta

ryin

vest

men

tin

sup

por

tfor

wor

ker

kaiz

en(s

pec

ific

trai

ning

,job

rota

tion,

eng

inee

ring

sup

por

t,m

anag

emen

tin

cent

ives

),w

hich

lead

sto

aco

nsid

erab

leflo

wof

usef

ulid

eas

•Th

eco

rew

ork-

cycl

ere

mai

nsve

rysh

orta

ndhi

ghl

yst

and

ard

ized

•N

umer

ous

man

agem

entp

ract

ices

and

hum

anre

sour

cem

anag

emen

tpra

ctic

escr

eate

hig

hm

utua

ltru

stan

dth

ereb

yre

duc

eth

eris

kof

opp

ortu

nism

SW

ITC

HIN

G:

Sep

arat

etim

esfo

rro

utin

ean

dno

nrou

tine

task

san

dsw

itch

emp

loye

esb

etw

een

them

seq

uent

ially

•K

aize

nis

also

cond

ucte

dof

f-lin

e,in

QC

circ

les

•P

rod

uctio

nw

orke

rsp

artic

ipat

ein

kaiz

enac

tiviti

esd

urin

gp

ilotr

uns

•W

orke

rsro

tate

thro

ugh

Pilo

tTea

m•

Sup

plie

rsp

olle

dfo

rim

pro

vem

ent

idea

saf

ter

each

chan

geo

ver

pro

ject

•C

onfli

ctin

gex

pec

tatio

nsin

two

role

s:hi

gh

auto

nom

yan

dth

eref

ore

hig

hco

mm

itmen

tin

nonr

outin

ero

les

vers

uslo

wau

tono

my

and

ther

efor

elo

wco

mm

itmen

tin

rout

ine

role

s

•N

UM

MIe

nsur

esth

atro

utin

ew

ork

isno

talie

natin

g:

•P

artic

ipat

ive

lead

ersh

ipan

dcu

lture

•W

orke

rtr

aini

ng•

Sup

por

tive

team

s•

Em

plo

ymen

tsec

urity

•G

ains

harin

g•

Hig

hm

utua

ltru

stb

etw

een

man

ager

san

dw

orke

rsin

each

othe

r’sco

mp

eten

cean

dco

mm

itmen

t•

Uni

onvo

ice

add

scr

edib

ility

tom

anag

emen

tcom

mitm

ents

PA

RTI

TIO

NIN

G:

Cre

ate

sub

units

that

spec

ializ

ein

rout

ine

orin

nonr

outin

eta

sks

•A

new

par

titio

nis

crea

ted

:Pilo

tTe

am•

An

old

par

titio

nis

elim

inat

ed:

pro

duc

tion

wor

kers

do

met

hod

sen

gin

eerin

g•

Res

pon

sib

ilitie

sar

ere

dis

trib

uted

acro

ssex

istin

gp

artit

ions

•S

upp

liers

do

mor

ed

esig

nw

ork

•Th

ere

iscl

ose

and

early

mut

ual

adju

stm

entb

etw

een:

•D

esig

n•

Eng

inee

ring

•M

anuf

actu

ring

eng

inee

ring

•P

rod

uctio

nm

anag

emen

t•

Sup

plie

rs

•A

dd

ition

alov

erhe

adis

req

uire

dto

sup

por

tdiff

eren

tstr

uctu

res

ind

iffer

entia

ted

sub

units

•N

ewsu

bun

itsne

edto

be

inte

gra

ted

,b

utin

teg

ratio

nm

echa

nism

are

cost

ly•

Sub

units

with

diff

eren

tiate

dg

oals

tend

top

aroc

hial

beh

avio

r•

Mut

uala

dju

stm

entb

etw

een

sub

units

isun

der

min

edb

yth

eb

urea

ucra

ticco

ntex

t

•A

ssig

nmen

tsto

the

Pilo

tTea

mar

ete

mp

orar

yra

ther

than

per

man

ent,

whi

chhe

lps

keep

goa

lsan

dva

lues

alig

ned

acro

sssu

bun

its•

The

Pilo

tTea

mw

orks

incl

ose

dai

lyin

tera

ctio

nw

ithp

rod

uctio

n,an

dd

oes

muc

hof

itsw

ork

onth

esh

opflo

or,w

hich

red

uces

par

ochi

alis

m•

The

Pilo

tTea

mis

mat

rixed

into

bot

hE

ngin

eerin

gan

dM

anuf

actu

ring

man

agem

entt

ob

alan

ceth

eir

resp

ectiv

eco

ncer

ns•

Mut

uala

dju

stm

enti

sfr

ont-

load

ed,m

inim

izin

gne

edfo

rit

late

rin

the

pro

ject

(whe

nit

ism

ore

exp

ensi

ve)

•E

arly

mut

uala

dju

stm

ente

ffort

sst

rive

tocr

eate

stan

dar

ds

that

sim

plif

yd

owns

trea

mco

ord

inat

ion

•M

etar

outin

esfo

rch

ang

eove

rman

agem

entf

acili

tate

mut

ual

adju

stm

ent

•E

nric

hmen

tand

switc

hing

mec

hani

sms

imp

rove

obje

ctiv

ean

dp

erce

ived

qua

lity

ofch

ang

esp

rop

osed

by

Pilo

tTea

m•

Hig

hle

vels

oftr

ustb

etw

een

inte

rdep

end

entu

nits



work of the Pilot Team and guided interactions with sup-pliers. And a structured reflection-review procedure fa-cilitated efforts to improve changeover management fromproject to project.

Second, NUMMI derived considerable tradeoff-shifting benefit from the enrichment of routine productiontasks. Continuous improvement was defined as a key ad-ditional responsibility of production workers, indeed ofall NUMMI personnel. Workers were encouraged to pullthe “andon cord” to signal problems in their work andstop the line if necessary. NUMMI’s managers put a pre-mium on mindfulness in the conduct of routine activities.Workers’ suggestions were particularly important duringthe acceleration of production on the new model: hereworkers were actively mobilized to identify problems andpropose solutions to help the acceleration. Instead of leav-ing job design to a methods engineering department—NUMMI had no such department—workers were activelyinvolved in the process of job design and redesignthrough the “standardized work” process.

Suppliers’ tasks were also enriched. The traditional BigThree approach defined suppliers’ role as simply fulfillingthe terms of their purchase agreement and assuring thattheir products met the specifications provided by the com-pany. By contrast, NUMMI mobilized suppliers’ productdesign capabilities, and expected—and provided supportfor—continuous improvement and innovation in both thesuppliers’ products and their internal processes.

Third, a broad range of policies encouraged workers toswitch easily between production and improvement tasks.Workers improvement ideas were developed not only on-the-job during regular production but also in off-lineQuality Circle meetings. Workers participated in pilotproduction runs, where they helped identify problems andimprovement opportunities. And workers were also giventemporary assignments to the Pilot Team.

Fourth, NUMMI shifted the tradeoff by creating newpartitions, reallocating tasks across partitions, eliminatingdysfunctional partitions, and improving coordination andintegration between partitions. The Pilot Team was anovel specialized unit, working alongside an engineeringchangeover team, responsible for designing the work pro-cess for the new model and for training workers for theirnew assignments. Responsibilities were reallocatedacross existing partitions, in particular through job en-richment for workers and more active involvement ofsuppliers. The traditional specialized methods engineer-ing department was eliminated, and work methods weredetermined by workers on the line. Various partitionedunits—within the plant, within other parts of the corpo-ration, and suppliers—interacted intensively to assure ef-fective mutual adjustment.

These mechanisms had distinct but mutually reinforc-ing tradeoff-shifting benefits. Metaroutines increased theefficiency of a given level of flexibility. Metaroutines alsoindirectly encouraged greater flexibility by facilitating theidentification of anomalies whose resolution representedopportunities to further increase flexibility. The direct ef-fect of the remaining three mechanisms was to increasethe organization’s innovation capabilities and thereby itsflexibility. They also indirectly encouraged greater effi-ciency when these innovation capabilities were directedat improving ongoing operations.

Contextual FactorsThe results summarized in the final column of Exhibit 1suggest that the four tradeoff-shifting mechanisms func-tioned far more effectively at NUMMI than at the BigThree in part because these mechanisms were embeddedin a very different organizational and inter-organizationalcontext. Our analysis of the NUMMI case highlightedtwo key features of this context: training and trust.

Training was critical. If people lack the knowledge,skills, and abilities required for the effective implemen-tation of the four basic mechanisms, the tradeoff cannotbe shifted. NUMMI invested far more than Big Threeplants in worker training. They also invested more thanthe Big Three in supplier “technical support.”

Trust proved to be a second critical contextual factor.All four of the tradeoff-shifting mechanisms are vulner-able to failures of one or another of the three principlekinds of trust: consistency, competence, and congruence(generalizing from Sako’s typology of kinds of inter-firmtrust (1992): contract, competence, and goodwill).

First, lack of consistency trust—i.e., lack of trust thatthe other party will do what they said they would (Sako’s(1992) contract trust, and Mishra’s (1996) reliabilitytrust)—can undermine support for metaroutinization andfor the other three mechanisms. NUMMI’s culture placeda high premium on consistency, on “walking the talk.”Top management commitment to this value was enactedin the use of cross-level forums in which breakdowns ofconsistency could be surfaced and dealt with under normsof “fact-based management” rather hidden by parochialpolitics. The credibility of this commitment was but-tressed by strong union voice.

Second, in many organizations managers and subor-dinates distrust each other’s competence to fulfill theircommitments (the Sako (1992) and Mishra (1996) com-petence trust, the Mayer et al. (1995) ability trust).NUMMI’s extensive training investments assured highlevels of worker competence. NUMMI’s extensive tech-nical support for suppliers motivated high levels of trustin supplier competence. NUMMI thus moved from in-specting incoming parts to certifying the ability of the



suppliers to produce parts that met specification. Andmanagement competence was buttressed by high levelsof investment in training for first-level managers and apolicy of promotion from within.

Third, all four mechanisms can easily be underminedby lack of trust in goal congruence (Sako’s (1992) good-will trust, Mishra’s (1996) openness and concern). Lackof congruence trust is commonly encountered in the con-flict between horizontally differentiated subunits withinthe organization, between vertical layers in the organi-zation, and between suppliers and customers. AtNUMMI, “teamwork” was a core value expressed notonly in the organization of workers into small productionteams, but also in the ethos governing relations betweendepartments and vertical layers, as well as in labor andsupplier relations. Divisive political motives were damp-ened by top management’s commitment to “fact-basedmanagement.” The union’s voice in the governance of theplant strengthened workers’ confidence that managementdecisions would reflect common goals and not only cor-porate goals.

All three kinds of trust appeared in both interpersonaland system forms. Alongside the three types of interper-sonal trust, NUMMI performance was also predicated onstakeholders’ trust in the consistency, competence, andcongruence of NUMMI’s management system, its sup-plier relations systems, and its labor relations system.

Finally, alongside trust proper, we also found a rangeof mechanisms that induced trust-like behavior throughwhat has been called “calculative trust.” For example:NUMMI’s work organization made its success dependenton front-line employees’ efforts. By making its own suc-cess a hostage to employees’ goodwill, NUMMI man-agement committed itself to not taking advantage of op-portunities to hurt the workforce. (Management’sdifficulties in living up to its implicit promises to act asif it internalized the workforce’s goals are discussed be-low and analyzed further in Adler et al. 1997).

The following sections provide supporting evidence forthese findings. We begin with describing in more detailthe relevant company background.

Company BackgroundThe GM plant in Fremont, California, closed its doors in1982, idling 5,700 workers laid off over the course of itsslow demise since the late 1970s. In a troubled company,this was one of the most troubled plants. Unexcused ab-senteeism often ran over 20 percent. Quality levels andproductivity were far below the GM norm, which itselfwas falling ever further behind the world-class standardthen being set in Japan. Labor relations were highly an-tagonistic.

In December 1984, a new company, New United Mo-tors Manufacturing, Inc. (NUMMI), took over the oldplant. NUMMI was a joint venture between GM and Toy-ota. GM provided the factory and would market half theplant’s output, and Toyota invested $100 million andwould assure the day-to-day management of the plant.

Like the factory, the workforce and union were largelyinherited from the GM days. The personnel selection pro-cess was done jointly by the union and management, andlaid-off GM-Fremont workers had hiring priority. Veryfew of these applicants were turned away. When produc-tion began in December 1984, 99 percent of the assemblyworkers and 75 percent of the skilled trades workers wereformer GM-Fremont employees and UAW members.NUMMI rehired the entire union hierarchy.

By 1986, NUMMI had achieved productivity levels al-most twice those of GM-Fremont in its best years, 40percent better than the typical GM assembly plant, andvery close to the level of NUMMI’s sister plant in Ta-kaoka, Japan (Krafcik 1989). It was also producing thehighest quality levels of any domestic auto plant. In 1988,the company switched from the Corolla FX and Nova tothe regular Corolla and a new nameplate, the Geo Prizm.Through the mid-1990s, the plant continued to improvein quality and productivity, producing some 200,000 Co-rollas and Prizms a year.

In 1989, Toyota announced that it would invest another$350 million to expand the plant and begin production ofToyota compact pick-up trucks. With a capacity to build125,000 trucks, the new line opened in September 1991.By 1993, NUMMI was producing about 120,000 truckswhich, along with Toyota’s imported trucks, were ratednumber one in initial quality by J. D. Power and Asso-ciates.

When NUMMI started up, the trauma of layoffs andthe influx of an entirely new group of managers contrib-uted to unfreezing old attitudes. However, NUMMI’scontinued improvements in quality and efficiency and thepersistence for over a decade of dramatic differences be-tween the performance of NUMMI and that of compa-rable Big Three plants must be attributed primarily dueto NUMMI’s management approach, particularly its useof the Toyota Production System and supporting man-agement policies (Ohno 1988, Monden 1983). The fol-lowing brief overview of these policies shows the extentto which workers’ tasks were enriched and improvementactivities were subject to metaroutinization.

NUMMI’s Production SystemThe Toyota Production System (TPS) structured the workprocess at NUMMI under several complementary poli-



cies: team organization, just-in-time production, levelproduction, mixed-model production, visual control, stan-dardized work, and continuous improvement. Workers atNUMMI were divided into teams of four to six, each ofwhich had a union member as Team Leader. Jobs werevery modestly enlarged: work cycles remained at the in-dustry norm of about 60 seconds, but Team Membersoften rotated jobs within their team. This not only re-lieved boredom and ergonomic strains, but gave workersa broader understanding of the production system. Jobswere also modestly enriched, by giving production work-ers some responsibility for quality, minor maintenancetasks, and line-side housekeeping. Team Leaders func-tioned primarily as lead hands, with modest administra-tive responsibilities but no managerial authority.

Under standardized work, each job was analyzed andthe optimal method was specified in motion-by-motioninstructions prescribing exactly how each job should beperformed. However, NUMMI’s approach to time-and-motion analysis differed from the traditional Big Three’sversion. At GM-Fremont, 80 industrial engineers de-signed the work process, monitoring and timing workersat specific jobs. At NUMMI, by contrast, there were nomethods engineers: Team Members and Team Leadersthemselves identified the optimal procedures for each job.Moreover, at NUMMI these procedures were subject tocontinuous improvement (kaizen). Workers were encour-aged to kaizen their jobs and suggest improvements to thestandardized work sheets. Workers were all trained inToyota’s six-step problem-solving procedure. Sugges-tions that passed muster became the new prescription, butonly until the cycle was renewed by the next suggestion.

Under mixed-model production, if the month’s deliv-ery schedule called for a production mix of 75 percent ofmodel A and 25 percent of model B for example,NUMMI would produce the same model mix each day,and three model A cars moving down the line would al-ternate with one model B. Big Three plants, by contrast,did not aggressively seek to reduce setup times, and theytherefore batched similar products, producing one modelfor several hours at a time. Mixed model production bothrelied on and forced high levels of mindfulness in routineproduction.

In 1991 NUMMI also introduced quality circles (calledProblem Solving Circles), where volunteers from a workgroup selected and studied a problem for several weeksover a lunch provided by the company. PSCs were notintroduced earlier, since Toyota’s philosophy was thatworkers and managers needed time to accumulate the pre-requisite knowledge, skills, abilities, and mutual trust.

NUMMI’s Management SystemNUMMI’s implementation of the Toyota Production Sys-tem was buttressed by management policies that encour-aged worker commitment and skill formation. One im-portant set of commitment-enhancing policies involvedrelations with the union. NUMMI and the UAW used ajoint problem-solving approach on many issues that wereclosely-guarded management prerogatives at GM-Fre-mont. In addition, the collective bargaining agreementpromised a measure of job security. The company’s suc-cessful efforts to avoid layoffs during a mid-1980s down-turn greatly enhanced employees’ confidence in the com-pany’s no-layoff commitment. Kaizen activity wasencouraged by a gainsharing system rewarding all work-ers for improvements in plant-wide quality and effi-ciency: since its introduction in 1990, it paid out $645 in1992, $733 in 1993, $1,285 in 1994, $1,130 in 1995, and$1,316 in 1996.

NUMMI also pursued many avenues for skill forma-tion. New hires received more than 250 hours of trainingduring their first six months on the job, while the averageBig Three auto worker received 42 hours in their first year(MacDuffie and Kochan 1995). Team Member cross-training was required for job rotation, and this trainingwas carefully tracked by Team Leaders. NUMMI’s policyof promotion from within for skilled trades, Team Lead-ers, and Group Leaders was accompanied by an extensiveset of training opportunities. Not only did promotions en-tail further training, but to be considered for advancementto any of these positions, workers were required to par-ticipate in company-sponsored training on their own time.

A Distinctive Kind of BureaucracyConsistent with the predictions of organization theoryconcerning organizations whose primary tasks are veryroutine, NUMMI was a very bureaucratic organization,in the technical sense of the term. Formalization and stan-dardization were extensive. Standardized work and otherkinds of detailed formal procedures governed daily opera-tions. Even the very modest degree of flexibility requiredin NUMMI’s routine operations relied on extensive for-malization and standardization: mixed-model productionstandardized job sequences and relied on formalized JobInstruction training and detailed standardized workcharts; quality circles used a standardized problem-solving process. Work teams were of the traditional ratherthan the self-directed type. While workers participated indefining work methods, changes in methods were ofteninitiated by management and engineering, and wherechanges were initiated by workers their ideas had to beaccepted by the team on the opposing shift and by therelevant supervisors. Vertical and horizontal differentia-tion were also extensive. NUMMI had a relatively tall



hierarchy with six levels between the worker and theCEO, which was about the same number of levels as atGM-Fremont. While NUMMI had eliminated the meth-ods engineering staff, it had unusually large staff groupsin human resources, quality engineering, and productioncontrol.

This bureaucracy, however, did not function in themanner depicted in much conventional theory. Workerswere actively involved in defining formalized proceduresand in refining them over time. Departures from proce-dure were typically treated as opportunities for learningrather than as threats to authority. Under the influence oftop managers’ values and the union’s power, lower-levelmanagers were encouraged to maintain a participativerather than autocratic style. While lower levels did nothave much autonomy to make decisions without priorconsultation with superiors, this apparent centralizationusually took the form of “fact-based” dialogue based onexpertise rather than command-and-control dominationbased on positional authority (consistent with Lincoln andKalleberg’s (1991) finding that Japanese organizationshad more de jure centralization but also more de factoparticipation than comparable American organizations).An extensive system of formal controls was buttressedand complemented by extensive informal controls, andboth formal and informal controls were to a large measurejoint rather than imposed, insofar as their design andfunctioning were strongly influenced by employees andby the union.

As a result, and contrary to the expectations of a longline of organization theorists, NUMMI’s highly bureau-cratic form was associated with high levels of work mo-tivation and commitment (Adler 1993). Managementconducted confidential employee surveys every twoyears; since these were conducted on work time, partici-pation rates were over 90 percent. In these surveys, theproportion of Team Members reporting themselves sat-isfied with their job increased progressively from 65 per-cent in 1985 to some 90 percent in 1991, 1993, and 1995.Although the absence rate had climbed from about 2.5percent to 3 percent over recent years, this rate was stillless than half the industry average. Participation in thesuggestion program grew progressively to over 90 per-cent. Involvement in problem-solving circles was increas-ing over time. Personnel turnover remained under 6%.And the plant still produced some of the highest qualitysmall cars and light trucks in the United States, with qual-ity and efficiency levels that continued to improve overtime.

The ’93 Model ChangeHow does an organization so finely tuned for the efficientperformance of a repetitive task deal with the challenge

of flexibility? In this section we examine how NUMMIintroduced two new passenger cars that differed substan-tially from each other and from the models they replacedat the same time as it introduced important new processtechnology. We do this through a narrative account ofNUMMI’s introduction of the 1993 model-year Corollaand Prizm, and a comparison with traditional Big Threepractice. In the following section, we describe howNUMMI learned from the weaknesses of this 1993changeover to improve the changeover on the truck lineto the 1995 model Tacoma truck.

NUMMI’s Model Change ProcessIn the traditional American system, the vehicle designdepartment first developed its sketches and clay models,then the engineering department converted these to en-gineering specifications. The engineering department inturn passed these specifications to the manufacturing staffat division headquarters, who specified the basic equip-ment and line configuration. The staff finally passed theseprocess specifications to the plant engineers, who in-stalled equipment on the factory floor and organized theassembly tasks. Workers were laid off for the time neededto retool, then recalled when the new equipment wasready.

The advantage of the Big Three’s sequential approachwas that downstream departments could wait to beginwork until the immediately upstream department wasdone. To the extent that artifacts such as specificationsand drawings provided sufficient guidance for the down-stream departments’ work, this approach reduced coor-dination costs. The success of this approach, however,depended on the upstream departments’ ability to predictthe constraints facing the downstream departments. Gapsin this knowledge would surface when the downstreamdepartment found, for example, that two mating parts didnot fit even though they met specifications. The risk ofthe sequential, “throw the drawings over the wall” ap-proach was thus that misfits would only be discoveredlate in the process, at which point a round of mutual ad-justment would be initiated. The industry rule-of-thumbwas that the cost to rectify a given error increases by afactor of about 10 with each of the four or five majorphases of a new vehicle development project. A problemuncovered after release to manufacturing could—and of-ten did—cost hundreds of thousands of dollars to rectifyand weeks or months of delay in the new model launch.

At NUMMI and Toyota, designers, product engineers,plant engineers, production workers, and suppliers col-laborated from the beginning of the changeover planningcycle. Instead of the traditional Big Three sequentialmodel, NUMMI and Toyota saw the interdependence of



these groups as reciprocal, and as one that therefore re-quired extensive mutual adaptation (see Thompson1967).

Manufacturing’s involvement in the preparations forNUMMI’s 1993 model change began very early, in 1989,as soon as the previous major model change was com-pleted. Toyota design engineers began collecting prob-lems and suggestions from NUMMI as well as from Toy-ota’s other plants in Japan and Canada. By early 1990,over a year prior to their model changeover, NUMMI’ssister plant in Takaoka had posted engineers to work withthe design team and had begun reviewing the emergingproduct designs and identifying production process is-sues. NUMMI’s model change would lag Takaoka’s bya year; but later in 1990, over two years before NUMMI’schangeover, the US factory became involved in the prep-aration process, and a team of engineers from NUMMIand representatives of NUMMI suppliers began visitingJapan to work with the vehicle designers, the corporateProduction Engineering department, and their Takaokacounterparts.

The early interactions between design, engineering,and manufacturing personnel facilitated the timely dis-covery of many misfits. It also allowed for more creativityin improving fit quality, through aggressive efforts invalue engineering. The goal of this collaboration—in-deed, the “secret” to NUMMI’s model changeover agilityaccording to a senior Toyota manager—was rigorous cer-tification prior to the resumption of operations that “man[read: workers], machines, materials, and methods” werecapable of doing their intended jobs. Not only did thiscertification dramatically reduce the number of changesrequired after start of production, but certification alsocreated a set of standards that reduced the need for(costly) mutual adaptation in specifying any later changesthat did need to be made. NUMMI thus derived consid-erable benefit by “front-loading” mutual adaptation, thatis, shifting it to earlier phases than the Big Three.

Organization theory provides us with the conceptsneeded to theorize these effects. Thompson (1967) arguedthat pooled interdependence is managed through stan-dards, sequential interdependence through plans andschedules, and reciprocal interdependence through mu-tual adaptation, and that this sequence constitutes aGuttman scale of increasingly effective but increasinglycostly coordination mechanisms (see Adler 1995 for ap-plication to the different phases of product development).The Big Three’s traditional management model assumedthat the changeover task could be managed sequentially,relying on standards, plans, and schedules to assure in-terdepartmental coordination. NUMMI aimed at a higherdegree of joint product/process optimization, and they

therefore interpreted the changeover task as embodyingmore uncertainty than the Big Three did. NUMMI there-fore used more intensive coordination mechanisms, inparticular mutual adjustment. By front-loading this mu-tual adjustment, NUMMI reduced the number and cost ofdownstream problems. Compare Big Three practice: ac-cording to Chevrolet Engineering Director, Dave Hanson,“One major change in [General Motor’s model change-over] strategy this year is an emphasis on validating sup-plier parts as production-ready during pilot assembly. [..]In the past, GM sometimes forged ahead to regular pro-duction when fewer than half a vehicle’s parts had beenvalidated at pilot” (Automotive News, Oct. 1994).

NUMMI’s tradeoff position was further improved bydistinctive capabilities in the upstream departments. Thecoordination and integration of all the units involved inthe model changeover was greatly facilitated by the factthat Toyota designers and design engineers were far moresensitive to manufacturability concerns than their BigThree counterparts. They all began their employment atToyota working on the assembly line for a period; theirmanagers accorded manufacturability assurance a highpriority throughout the design cycle; and Toyota designengineers had accumulated a considerable body of bothtacit and codified manufacturability knowledge (see alsoClark and Fujimoto 1991). The resulting knowledge gavedesign engineers tacit and explicit manufacturability stan-dards with which to guide their work, and thus reducedthe number of issues requiring mutual adaptation discus-sions with manufacturing personnel. This background as-sured both goal congruence across functions and unusu-ally high levels of competence in this part of the designengineers’ task.

The Pilot TeamOne of the most striking differences between the tradi-tional Big Three approach and NUMMI’s was the rolethat production workers played in the design ofNUMMI’s products and production processes. One TeamLeader from each group in the plant was selected to joina Pilot Team. Their primary responsibility was to draftthe standardized work sheets and to train supervisors andworkers in their new jobs. They also worked hand-in-hand with the engineering changeover team, helping tofine-tune the product design, select equipment, and layout the production process.

The Pilot Team at NUMMI was a permanent unit withrotating membership. Its role and size changed dependingon the schedule of model changes. Workers joined theteam for months at a time, usually returning to the shopfloor when the model they were preparing moved intoproduction. Eight months prior to the start of production,



the Pilot Team consisted of eight members; three of themhad worked on the previous major model change in 1988,while the others had only been through the minor modelchange of 1991. Within two months, the Pilot Team hadadded 16 new members. By the start of production, therewas a total of two Group Leaders and 32 Pilot TeamMembers.

Pilot Team members were given a lot of informal on-the-job training, and they had a considerable body ofdocumentation from prior model change projects, includ-ing Pilot Team activity flow charts. Formal training intheir new task, however, was minimal. Their work in-volved considerable improvisation—a degree of impro-visation that contrasted strongly in their minds with thestructured quality of regular line work. One Team Mem-ber described the relatively organic form of their orga-nization in these terms: “The first day, I thought, ‘Okay,I’m on the Pilot Team now. I’ve got a desk.’ But therewere no instructions. It was a crash course for us. We justkind of figured things out.”

The Pilot Team was divided into groups representingthe different sections of the plant (bodyweld, paint, as-sembly, etc.). Functioning in a matrix structure, eachgroup cooperated closely on the one hand with staff en-gineers under the engineering section’s assistant man-ager, and on the other hand with the Group Leaders re-sponsible for the corresponding section of ongoingproduction operations. Through the first of the two majorpilot runs, each section’s Pilot Team Leader reported pri-marily to the Engineering department manager, and fromthen on they reported primarily to the relevant sectionmanager in Production.

Early in 1992, seven months before the start of pro-duction, the Pilot Team traveled to Japan to study theTakaoka plant. (Recall that the earlier NUMMI visits hadbeen by the engineering team and suppliers.) They tookwith them concerns and suggestions for improvementscollected from Team Members in their areas. The PilotTeam members worked on the Takaoka line to learn howtheir counterparts had designed the specific jobs in thepart of the line for which they were responsible. Whenthey returned from Japan, the Pilot Team brought withthem large binders containing illustrations of individualparts and explanations of how they should be assembled.They then experimented in NUMMI’s pilot area—an areaset aside in an unused part of the plant—modifying thisdocumentation to fit the specifics of NUMMI’s line, andturning it into detailed draft work instructions (standard-ized work sheets) for production Team Members. Theyalso worked with their staff engineering partners to designand source the appropriate equipment for each job.

During their time in Japan as well as during their con-tinuing efforts at NUMMI, the Pilot Team and the engi-neers not only refined the process design, they also pro-posed product design modifications. Although somedesign change proposals were rejected as too expensive,most were incorporated. Design changes were particu-larly numerous for the Prizm. The Prizm was sold by GM,and GM had exercised its right to make design changesto the Toyota vehicle on which it was modeled (a car soldin Japan as the Sprinter). The NUMMI Pilot Team’s inputwas particularly important on the Prizm as they were thefirst to assemble the modified design of the car. Manysuggestions aimed to make the cars easier to manufacture,while others improved its cost and quality.

Relative to traditional U.S. practice, the Pilot Team wasa novel partition, facilitating mutual adjustment both be-tween process design considerations and production re-alities, and between process design and product designconsiderations. By contrast, the traditional Big Three se-quential approach lacked cost-effective ways of dealingwith these issues. First, when the product specificationswere released to Big Three staff industrial engineers, theyworked from handbooks and computer models—ratherthan from real production conditions—to define pro-cesses, layouts, and methods. Pilot production was tra-ditionally conducted on special pilot lines located not inthe final assembly plants that would ultimately producethe car, but in special facilities, such as low-volume lim-ousine plants, where pilot production would be less dis-ruptive to ongoing operations and geographically closerto central engineering staffs. These pilots were not con-trolled by plant personnel, but by staff engineers who didnot normally work in the manufacturing plants. Themanufacturing plant would conduct its own trials onlyafter it had shut down for the changeover. And the newproduct and process design would only be tested in realproduction conditions when workers returned to work atthe resumption of operations.

Moreover, whereas the Pilot Team could rely on col-laborative mutual adjustment with other participatingunits, Big Three design engineers traditionally did notactively participate in manufacturability improvement ef-forts. (We should note again, however, that this patternhas been changing in recent years.) Once the design hadbeen thrown over the wall to the manufacturing staff, itwas very difficult to get Big Three designers and engi-neers to review, let alone approve, manufacturability-motivated Engineering Change requests. Even if a designchange proposal was eventually accepted, the averagetime between submission and release was around sixmonths. Our interviewees at NUMMI described Toyota



design engineers as far more actively involved and re-sponsive right through vehicle launch.

Working with SuppliersUnder the traditional Big Three approach, the auto com-pany’s relationships with outside suppliers were arm’slength and often adversarial. If one supplier had risingcosts or declining quality, the auto company couldquickly switch to another. To maintain the credible threatthat a relationship could be easily terminated, the autocompany employed several suppliers for each part andnegotiated only short-term contracts. While this strategymaximized the auto company’s bargaining power, it alsorequired it to perform the design work in-house, and thuscut it off from suppliers’ ideas about product design andlimited suppliers’ customer-specific investment (Helperand Levine 1992, Helper 1990).

NUMMI, following Toyota, took a very different ap-proach. Prioritizing the ability to harness suppliers’ in-novative capabilities and to fine-tune part designs,NUMMI usually kept only one or two suppliers per part,negotiated long-term contracts with them, challengedthem to make product and process improvements, andworked with them when problems arose. In preparationfor the ’93 model, engineers from major suppliers workedin the NUMMI plant throughout 1992, just as Japanesesuppliers had worked with the Toyota design team theprevious year.

However, the transition from the old adversarial ap-proach was not easy for many suppliers. They were notaccustomed to Toyota’s high standards for quality, time-liness, and cost, nor to the expectation of continuous im-provement, nor to the collaboration NUMMI offered toimprove their performance. In the 1993 model change,the challenge of transforming the supplier base and de-veloping new supplier relations was exacerbated by a sig-nificant jump in the proportion of NUMMI’s parts thatwere made in North America. In 1992, NUMMI’s carsused parts from 88 domestic suppliers and had a domesticcontent of 63 percent (using the Environmental ProtectionAgency’s Corporate Average Fuel Economy metric); the1993 models used parts from 124 North American sup-pliers and had a domestic content of 75 percent.

Some suppliers had participated in both NUMMI andBig Three changeovers, and their experiences were verydifferent. Apart from the differences already noted, thegeneral manager of one of NUMMI’s suppliers identifiedseveral others:

NUMMI and Toyota have a detailed “template” for the pro-cess—a master schedule—and they follow that schedule veryclosely. It’s the same template from project to project, with onlyminor variations depending on the complexity of the product

and what they’ve learned from the previous project. We alsosupply parts to GM, but at GM the process changes from projectto project.

Toyota puts a lot of effort into prototyping and pilots, andthat reduces the number of engineering changes after they issueproduction drawings. In the GM system, many suppliers makelow-ball bids to get the business because they count on a wholelot of engineering changes to make their profit. Toyota focusesmore on the details earlier in the process. They use the pilots todiscover every possible problem and solve them. At GM theyuse pilots to “confirm” what they think they already know, notto uncover what they don’t know.

These differences in supplier relations echo those wehave noted in relations between internal units. NUMMIfront-loaded more mutual adaptation with the aim of min-imizing the need for downstream interaction; NUMMIfacilitated this front-end mutual adaptation by collabo-rative rather than arm’s length relations with zero-sum ornegative-sum bargaining; NUMMI benefited from a re-distribution of tasks across the supplier/assembly parti-tion, relocating some innovation tasks in the supplierfirms where they could be performed more knowledge-ably; and NUMMI reduced the cost of this mutual ad-aptation by the use of the metaroutine of a standard mas-ter schedule. In organization-theoretic terms, we wouldsay that as a result, NUMMI’s supplier relations weremore effective at both buffering (to reduce uncertainty inthe core) and bridging (to manage remaining uncertain-ties) (Scott 1992, pp. 193 ff.; Gerwin 1993; Correa 1994).

Pilot ProductionFive months prior to the start of production, NUMMIworkers built the first set of 25 pilot vehicles. This firstpilot build focused on engineering issues. It was thereforeconducted off-line in the pilot area, and it relied primarilyon parts that were custom-built by suppliers.

In June 1992, two months before start of production,the second major pilot build was conducted. Whereas thefirst pilot focused on engineering issues, this one focusedon production issues. It was conducted on the regular as-sembly line, and primarily used parts from suppliers’ reg-ular production lines.

Because of the mixed-model discipline in its regularoperations, NUMMI could intersperse this second set ofpilot vehicles with regular production vehicles, leavingjust two empty spaces in the production sequence to sig-nal the arrival of the pilot vehicle. This represented aremarkable degree of agility, since all the associated partsand equipment had to be in place along the line, andworkers had to be able to switch between the two sets ofparts and tools in the 120 seconds afforded by the twoempty spaces. The benefits of a pilot run on the regular



production line with regular production workers were im-mense: far more problems could be identified than waspossible in an off-line pilot build.

NUMMI’s management of its pilots offers three salientcontrasts with traditional Big Three practice (see alsoClark et al. 1992). First, Big Three plants typically didnot do on-line pilots in already-functioning plants; theywaited until after the old line had shut down and the newequipment had been put into place. This considerably de-layed identifying and solving problems. Second, U.S.firms were traditionally guided by cost concerns and wereunder little pressure to accelerate changeovers; as a result,they laid off most of their production workers during theprotracted plant shutdown, recalling them progressivelyas production accelerated, and training them only as theycame back into the plant. Third, these returning workerswere trained by the small group of core workers whoconducted the first pilot builds, whereas at NUMMI,Group Leaders and Team Leaders had the primary re-sponsibility for training. The Big Three could not usesupervisors to do training, since unlike NUMMI, theirsupervisors were not usually promoted from the shop-floor; they were usually college graduates on a manage-ment track, hired straight into supervisory jobs and thuslacking the relevant technical knowledge. And whereNUMMI had a Team Leader who functioned as a leadhand for each team, the Big Three generally lacked sucha role.

The Acceleration PeriodAt the cost of some overtime to accommodate trainingand pilot builds, NUMMI produced a full schedule of theold model vehicle right up to the week before productionof the new model began. On August 7, 1992, NUMMIstopped production for one week to prepare for thechangeover. During this week, the Pilot Team workedwith managers, engineers, and maintenance workers toset up the new line. Production workers were required—as they had been warned a year in advance—to use aweek of their vacation time. When operations resumedon August 14, the Team Members returned to new jobs.

To give Team Members time to learn their new jobsand maintain high quality, production volumes were keptvery low at first. Instead of producing some 450 cars ashift, Team Members began with four or five cars a shift,trying to work within a 60-second takt time, but takingtime out after each cycle to identify the sticking points.(“Takt” is German for meter. Toyota defines takt time asthe interval between cars as they leave the line.) As theproduction process began to flow more smoothly, pro-duction volumes slowly increased. Within the first week,the pace quickened from one car an hour to one car every

10 minutes: a single car followed by nine empty slots onthe conveyor line. These early vehicles were used to con-firm product and process quality.

As had been done during the previous major modelchange in 1988, management had suspended job rotationstemporarily. Rotations were due to restart on September4, the date when management planned to be producing140 cars per shift and volume shipments could begin.Management wanted first to ensure high quality at a mod-erately high production rate before restoring full rotation.

In the first weeks, the primary goal was training. TeamMembers walked through their jobs, putting on parts, re-moving them, and putting them on again. The other goalin the first days of production was kaizen. Each time thepace increased, new problems appeared: some jobs thathad appeared easy at low volume were seen to be over-loaded, and new technical issues surfaced. In addition tolearning their jobs, workers were also expected to im-prove them. Nearly every worker we interviewed de-scribed suggestions they had made during the first daysof the acceleration for increasing quality, productivity, orsafety. Most of these suggestions had been tested andimplemented very rapidly.

By contrast, the traditional Big Three accelerationswere significantly handicapped by the gaps in knowledgeand conflicts of interests that marked the relationship be-tween workers on the one hand and industrial engineersand managers on the other. First, going into the resump-tion of operations, the layout, methods, and line balancedefined by industrial engineers were further from the op-timum than at NUMMI: industrial engineers worked fromtheoretical models rather than plant experience, and sotheir estimates were typically less accurate than the PilotTeam’s. Second, the fine-tuning required after resumptionof operations was a kind of purgatory, where workerstried to hide under-burdened jobs and foremen tried tolocate and add tasks to those jobs, while other workershad to sabotage production to get foremen to relieve im-possibly overburdened jobs. And finally, nowhere in allthis conflict was any attention paid to fine-tuning the in-dividual worker’s work methods: industrial engineersspecified the theoretically optimal methods, but in prac-tice foremen left workers to improvise whatever methodswould allow them to meet the time standards.

Problems in the AccelerationThe new line took only 77 days to accelerate to full pro-duction. This was far faster than typically found in BigThree plants, but not as fast as NUMMI’s original planof 60 days. The acceleration had to surmount three setsof unanticipated problems. First, some of the new tech-nology introduced with this model change—notably, a



doors-off conveyor and an instrument panel sub-assemblyline—broke down more often than expected. Second,some parts did not arrive on time. NUMMI had changedsome of its supply logistics, consolidating some deliveriesin Chicago and taking more frequent deliveries; errors inplanning and executing these changes led to delivery de-lays. Third, numerous part-fitting and “workability” prob-lems appeared, partly because of weaknesses in some USsuppliers (see for example, Bowen and Ryckebusch 1996,p. 13), and partly because some US suppliers were work-ing to outdated drawings supplied by their Japanese coun-terparts. Working in close collaboration back in Japan,Toyota engineers and Japanese suppliers had made nu-merous fine-tuning changes to parts designs, but oftenthey failed to update the corresponding drawings sent toUS suppliers.

The plant was able to maintain high quality, but theseworkability problems imposed a considerable cost. Pro-ductivity suffered as parts that should have taken five sec-onds to snap into place took 10 or 15 seconds. That dif-ference, while seemingly small, can require that the linebe stopped repeatedly, and NUMMI lost about $9,000 inrevenue every minute the line stopped.

Workability deficiencies not only hurt productivity;they also hurt workers. When parts did not fit well,NUMMI workers tried to force them into place. Duringthe acceleration, management and union officials alikereported seeing workers pounding parts with the palms oftheir hands. The stresses these actions place on the bodycan lead to soft-tissue disorders such as tendinitis andcarpal tunnel syndrome. According to the OSHA-200 log,NUMMI’s injury rate jumped in the first month of pro-duction to a level some 50 percent above that of the prioryear.

A month after the resumption of operations, in responseto the growing number of ergonomic problems, the uniondemanded that rotation be restarted. Redesigning the rele-vant parts would take time, and the union’s position wasthat in the interim, ergonomic strains could be relievedby resuming rotations for the most stressful jobs.NUMMI management refused, arguing that short ofaborting the launch, it was physically impossible to freeup enough people to allow the cross-training that wouldbe needed for rotation. On September 28, 1992, the UAWLocal filed a formal complaint with Cal-OSHA. A fewweeks later Cal-OSHA began an investigation, and thefollowing January, Cal-OSHA issued one warning andtwo “serious” citations based on a large number of er-gonomic problems that the OSHA inspector observed inthe passenger car assembly area. NUMMI appealed thetwo citations. In January 1994 a settlement was reached

with OSHA committing NUMMI to more substantial ef-forts in the ergonomics arena, and later that year a newergonomics agreement was reached with the UAW cre-ating a full-time union ergonomics representative. (Onthis ergonomics conflict, see Adler et al. 1997).

Results of the 1993 ChangeoverThe 1993 model change was a very smooth process com-pared to the traditional performance of Big Three plants.Within five months of the start of production, assemblyefficiency was up to 96 percent, which was two percentover target, and two percent over the level a year earlier.These levels are all the more impressive given that as-sembly efficiency was measured as the ratio of actualdaily output to scheduled output, and the scheduled out-put implied 100% utilization with no line downtime al-lowance built in. The influential J. D. Powers ratings—based on customers’ evaluations of vehicles produced inthe last three months of 1992, that is, just as NUMMI wasaccelerating production—showed only a three-point slipin the number of defects per 100 NUMMI Corollas anda two-point slip per 100 Geo Prizms. That is, the averagenumber of defects a typical car owner identified in thefirst three months of ownership rose by 0.03 and 0.02respectively. This minor dip left NUMMI’s passengercars still ranked first and second among all small carsproduced in the US.

While these results were impressive, the 1993 modelintroduction was not completely successful. As a resultof the problems encountered in NUMMI’s 1993 modelintroduction, the plant reached full production after 77production days instead of the planned 60 days. The totalshortfall in output over this acceleration period was about3,500 cars. Although this was less than a week’s produc-tion, it took many weeks of overtime to make it up.

Moreover, scores of workers were injured, some per-haps permanently. The conflict over ergonomics also con-tributed to a degradation of labor-management relations.In the 1994 union Local elections, the more confronta-tional People’s Caucus won an almost clean sweep, tak-ing all but the President’s position from the more coop-erative Administrative Caucus. The reasons for this shiftwithin the Local were complex, but according to manyinterviewees it reflected a need felt by the rank and filefor a more assertive union voice. Later that year, NUMMIexperienced its first work stoppage, a two-hour walkoutduring the contract negotiations. (We should note, how-ever, that the broader pattern of union-management co-operation was not much affected by this change and thatthe Administration Caucus took back leadership of theLocal in the 1997 elections.)



The 1995 Truck LaunchTwo years after the 1993 passenger car line changeover,NUMMI made its next major model change, this time theintroduction of the Tacoma compact pickup truck. Thecontrast between the two changeovers was striking:whereas the 1993 changeover had taken 77 days to reachfull production, the comparably complex changeover onthe truck line took only 48 days.

Moreover, the health and safety conditions of thischangeover were far superior. The first three months of anew model launch usually have above-average injuryrates; nevertheless, in the first three months of the Tacomalaunch, the truck area reported nearly 30 percent fewerinjuries than during the same period in the prior year, andfewer than on the passenger line at the same time. Thiswas a particularly impressive accomplishment, since inprior years, the truck line had a worse health and safetyrecord than the passenger line. In this section, we analyzethe reasons for these contrasting outcomes.

The “Reflection-Review”Following Toyota policy, NUMMI conducted a“reflection-review” (hansei) soon after the 1993 modelpassenger line reached full production. Before memorieshad time to fade, top management, section managers, As-sistant Managers, Group Leaders, and Pilot Team mem-bers documented the lessons learned from the launch.

Changeover project teams at NUMMI relied on a con-siderable body of documentation in voluminous bindersrepresenting the accumulated lessons learned from priorexperience. The hansei process allowed NUMMI to pro-gressively refine these procedures. One interviewee ex-plained the process in these terms:

The binders give us best-practice procedures for managingmodel changes—just like standardized work sheets give theworker best-practice procedures in regular production. And thelearning process is the same. In manufacturing, anomalies showup as differences between takt time and the worker’s actualcycle time, and these anomalies lead to problem-solving, whichthen leads to defining counter-measures, which in turn leads tonew standardized work procedures. Anomalies in the change-over process are the differences between our target changeovertime and our actual time. The hansei process is simply theproblem-solving procedure we use to improve our model changeprocess.

As in previous reflection-reviews, vice president ofManufacturing and Engineering, Gary Convis, gave thesection managers six weeks to prepare their analyses, thenconducted a plant-wide meeting with all the GeneralManagers, Assistant General Managers, Managers, andsome Assistant Managers to review each section’s sum-mary report. Over the following few days, Convis met

with each section—Managers, Assistant Managers, andPilot Team members—to review the binders and sum-maries. In these multi-level forums, there was little op-portunity to hide deficiencies, and senior management en-couraged a norm of fact-based, critical scrutiny with thegoal of identifying improvement opportunities for thenext project.

The review of the 1993 changeover identified scores ofsuch opportunities in every facet of changeover manage-ment. In particular, it highlighted the need for greater fo-cus on ergonomic issues. This led plant management toaccord a high priority to health and safety in both thecompany’s strategic plan and the planning for the Tacomatruck launch. The goal NUMMI set for 1995 was to cutthe overall plant injury rate by 30 percent.

The Pilot Team and WorkabilityOne of the lessons learned from the 1993 changeover wasthe need for greater staffing continuity and depth of ex-pertise in the Pilot Team. Organizational memory neededto be deployed not only through the formal mechanismof documentation but also through informal mechanismof team composition. As a result, most of the workersselected for the Tacoma Pilot Team had worked on oneor more major model changes.

Unlike the 1993 changeover, the Tacoma Pilot Teammembers were told that ergonomics was a key objectiveand they were given extensive ergonomics training. Thetruck line’s poor health and safety record was in consid-erable measure due to inadequate attention to ease of as-sembly in the earlier trucks’ designs. As a result of thecombined efforts of Toyota and NUMMI engineers andthe Pilot Team, the Tacoma was far easier to assemblethan its predecessors.

NUMMI’s Quality engineers also put more emphasison workability issues than in the 1993 case. They spentmore time analyzing parts before and after each pilot. TheQuality Engineering department also established bettercommunications with suppliers, and suppliers in turnwere able to respond more rapidly to design changes.NUMMI and the Toyota Supplier Support Center (basedin Lexington, Kentucky) provided technical assistance tosuppliers whose parts had caused problems in the 1993changeover.

Training and RotationUnlike NUMMI’s earlier changeovers, the policy in 1995was to ensure that all Team Members rotated between atleast two jobs from the very first vehicle. The explicitgoal was to make full rotation the standard policy for theentire truck and passenger lines.

In the Tacoma changeover, Team Member training wastherefore accorded a very high priority. Training on paid



overtime was made mandatory, whereas for the 1993launch overtime for training was at the Team Members’discretion. Although freeing up training time was difficultbecause of the high demand for the outgoing model truck,within a month of the start of production, virtually everyteam was able to rotate workers between four jobs.

Resumption of OperationsResponding to the experience of the 1993 launch and an-ticipating unforeseen problems and the weeks and monthsthat would be needed before kaizen activity would bringheadcount back down to steady-state levels, NUMMIhired 20 extra employees. Whereas in 1992, Team Mem-bers took a week of vacation time during the changeover,NUMMI asked truck line Team Members to work overthe Christmas/New Year break just before production be-gan to facilitate kaizen, training, and preparation.

Within two months of the resumption of operations, theplant was well head of its production goals. Partly as aresult of some training shortfalls, quality had sufferedsomewhat during the acceleration period; but within fourmonths it had reached a world-class level comparable tothat achieved by the truck line’s sister plant in Japan(Hino).

Safety AwarenessThe higher priority accorded ergonomic issues during theTacoma launch reflected changes introduced in the wakeof the Cal-OSHA citations. Management had made safetyimprovement a strategic priority, and had tied managers’evaluations and rewards to their departments’ safety rec-ord. Some of the new policies may not yet have born alltheir fruit, but as of mid-1995, both the truck line and theplant as a whole were on target to achieving the strategicgoal of a 30 percent reduction in the injury rate.

DiscussionOrganization theory suggests that an organization pro-ducing hundreds of thousands of nearly identical productseach year should be a machine-like bureaucracy with ex-tensive formalization, standardization, hierarchy, spe-cialization, and staffs. NUMMI’s regular operations ap-peared to fit these descriptors. Yet NUMMI was alsoremarkably agile in responding to the challenge of majormodel changes, a challenge that calls for a more organicform of organization. The organization of NUMMI’schangeovers appears to fit this description too. NUMMIappears to have been able to have it both ways. Insteadof being stuck in the middle, weighed down by the con-flicting requirements of efficiency and flexibility,NUMMI seemed able to excel simultaneously in both di-mensions. Moreover, NUMMI was able to further shift

the tradeoff frontier by improvements both in efficiencyand changeover agility between 1993 and 1995.

Mechanisms and ImpedimentsThe resolution of this apparent paradox lies in NUMMI’ssuccessful deployment of the four mechanisms we iden-tified in the prior literature. First, the metaroutines of stan-dardized problem-solving, changeover process proce-dures, and the reflection-review process allowed NUMMIto significantly routinize otherwise nonroutine tasks as-sociated with changeovers, and thereby to improve effi-ciency without impairing flexibility and vice versa. Sec-ond, NUMMI enriched routine production work,encouraging and training line employees and suppliers tostay alert for improvement opportunities. Third, workersswitched easily between routine production roles andnonroutine kaizen roles in quality circles, in pilot runs,and in temporary assignments to the Pilot Team. Fourth,NUMMI used partitioning more effectively than tradi-tional approaches allowed: it created a new, relativelyorganic partition, the Pilot Team, devoted to the nonrou-tine tasks associated with changeovers; it allocated tasksmore effectively across existing make/buy partitions; iteliminated dysfunctional partitions such as the methodsengineering department; and it greatly improved coordi-nation and integration between partitions. NUMMI couldthus enjoy the benefits of both discipline and creativity.

Organizational theory reviewed above predicts a num-ber of potential impediments to such mechanisms: Howdid NUMMI overcome them? We can summarize the les-sons of the NUMMI case by reviewing the four mecha-nisms in turn and summarizing NUMMI’s approach toeach.

First, some strands of theory predict that the routini-zation of nonroutine activities will lead either to resis-tance to modifying routines by employees anxious to pre-serve the status quo or to resistance to the use of routinesby employees anxious to preserve their task autonomyand variety. NUMMI avoided resistance to the modifi-cation of routines with a strong culture of kaizen, rewardsfor innovation, and strong leadership to reinforce that cul-ture. NUMMI overcame resistance to standardization byinvolving the employees themselves in the developmentand refinement of these routines. Instead of staff methodsengineers imposing formalized standards on core em-ployees, workers participated actively in defining and re-fining standardized work sheets. Instead of staff expertsimposing changeover management procedures, the entireorganization used the reflection-review process to pro-gressively define and refine the procedures.

NUMMI appeared to have largely overcome resistance



to relinquishing autonomy and variety in core tasksthrough two complementary means. First, when workersparticipated in the effort to routinize their core tasks, par-ticipation in this activity increased autonomy and variety,albeit only very modestly and in a noncore task, and thusat least partly addressed workers’ demands for moremeaningful work. Second, this fundamental limitation onmotivating job characteristics was legitimized in the eyesof NUMMI workers. Routinization was not imposed onworkers but presented as the path to competitiveness—and job security—in the world of high volume, mass pro-duction.

Second, some theory argues that enriching workers’jobs in mass production activities such as auto assemblywill require inefficient levels of training, and that the jobenlargement commonly recommended to accompany jobenrichment will weaken the discipline required for effi-ciency in routine core tasks. NUMMI overcame the for-mer impediment by a complementary investment in sup-port for worker kaizen activities: in combination, thesetwo investments generated large returns in the form of avery large number of worthwhile, albeit typically small-scale, improvements. This support was in the form of ka-izen training, job rotation to broaden workers’ under-standing of the production system, engineering supportfor the timely testing and implementation of workers’suggestions, and strong management support for workersuggestions. This suggests that the returns to investmentsin worker training may be low or negative until a certainthreshold is reached and an integrated package of man-agement practices creates structures, incentives, and abil-ity for continuous improvement (Levine 1995). The sec-ond possible impediment to enrichment as atradeoff-shifting mechanism is the loss of discipline inthe implementation of standardized procedures for rou-tine tasks that comes with excessive lengthening of cycletimes. NUMMI avoided this problem by keeping the cy-cle times for core production tasks very short and puttinggreat stress on the value of standardized sequences ofmotions in assuring high productivity and quality. A re-lated potential impediment lies in the additional risk ofopportunism created when enrichment reduces task pro-grammability. NUMMI appears to have avoided this classof problems through the establishment of a high level ofmutual trust between workers and managers: trust in com-petence and goal congruence.

Third, some theory predicts that allowing employeesto switch roles sequentially will not significantly shift theefficiency/flexibility tradeoff since workers can hardlyfunction as respected problem-solvers in organicallystructured quality circles and improvement teams if they

are treated as closely monitored proto-robots in their re-petitive jobs during the rest of the week. NUMMI appearsto have overcome this impediment by an extensive set ofpolicies and practices that encouraged innovation and em-ployee involvement, including a participative leadershipstyle in routine production, worker involvement in defin-ing and refining work procedures, a team-based work de-sign, a commitment to employment security, a union thatensured management kept its commitments, and gain-sharing.

Fourth, some theory is skeptical of the partitioningmechanism’s ability to shift the tradeoff because the crea-tion of new subunits typically creates additional manage-ment overhead for coordinating and resolving conflictsbetween subunits. Several factors reduced the coordina-tion and integration costs associated with the creation ofthe Pilot Team. First, the Pilot Team was drawn fromamong the production Team Leaders and the assignmentwas only temporary. This reduced potential gaps inknowledge, values, and incentives. Second, coordinationcosts were reduced by ensuring that the Pilot Teamworked in close interaction with the line organization allthrough the changeover process. The Pilot Team was ma-trixed into line management and it was located in theplant. Very early in the preparation process, line workersidentified for the Pilot Team problems with the currentprocess that they wanted to see fixed. Later in the process,during off-line training and on-line pilots, line workerscontributed kaizen ideas. And on-line pilot productionmultiplied these interaction opportunities. The same prin-ciples governed the role played by suppliers in thechangeover process: their involvement, cooperation, andextended on-site visits to the plant early in the changeoverprocess enabled NUMMI to simplify the downstreamchangeover task.

Our narrative revealed that NUMMI was not alwayssuccessful in its efforts to shift the tradeoff to combinefaster changeovers with greater efficiency. First, some im-provements to flexibility were bought with slack and theassociated loss of efficiency, most notably the decision tohire an additional 20 workers for the 1995 changeover.This decision appears to have been simply an adjustment,probably a wise one, to NUMMI’s position on the exist-ing tradeoff curve. Second, NUMMI stumbled in the1993 changeover when management failed to provideenough ergonomics expertise and failed to maintain theappropriate balance between production and safety goals.It stumbled too, in the 1993 logistics bottlenecks. How-ever, from the points of view of both workers and busi-ness performance, NUMMI was able to learn from thesefailures, as evidenced by simultaneous improvements in



flexibility, efficiency, and health outcomes in the 1995changeover.

Training and TrustTraining and trust appear to be the critical contextual fac-tors determining the effectiveness of these mechanisms’implementation at NUMMI. We have already commentedon training of production workers, and our narrative alsorevealed the importance of “technical support” forNUMMI’s suppliers, which was mainly in the form oftraining. The traditional Big Three model of supplier re-lations was based on a cost-minimization strategy; itpromised a form of flexibility too: in the ease of switchingbetween suppliers. But functioning in an arm’s lengthmode, this model cut the auto assembler off from poten-tial design and efficiency improvements that supplierscould be induced to make if they saw the assembler as along-term partner.

Trust appears to have been a second critical contextualfactor. Consistency, competence, and congruence trustwere key in assuring support for and effective use of me-taroutinization, enrichment, switching, and partitioningmechanisms. If trust levels between units, layers, and sup-pliers were not high, all four tradeoff-shifting mecha-nisms would have been hobbled. Workers’ trust in man-agement was challenged when the 1993 modelchangeover began creating high levels of ergonomicstrain for many workers, leading to a confrontation withthe union and citations by Cal-OSHA. The basic fabricof trust was, however, largely restored when—under thecombined pressure of top management, Cal-OSHA, andthe union—management changed its approach in the1995 truck-line model changeover. New ergonomic as-sessment procedures were put into place, bolstering con-sistency trust. Ergonomics training was increased, in-creasing competence trust. And health and safety becamea company strategic priority, significantly restoring con-gruence trust.

In our description of NUMMI’s organizational struc-ture, we highlighted the distinctive form of bureaucracythat characterized its routine operations, a form that hasbeen characterized as “enabling” in contrast with themore commonly-encountered “coercive” form (Adler andBorys 1996). Our analysis of ambidexterity at NUMMIreveals a strong link between this enabling form of bu-reaucracy and trust. Trust is often seen as an alternativeto bureaucracy and its rules; but as Sitkin (1995) has ar-gued, this is only true if formalized rules (“legalization”in Sitkin’s study) are “viewed as a ‘people-proofing’ sub-stitute for relational trust,” and in this case bureaucracywill indeed tend to undermine trust. However, if “formalprocedures and standards are used to routinize repetitive

interactions and stabilize expectations,” and if the orga-nization follows the “rule of minimization: keep formal-ized trust simple and narrow in scope,” then high levelsof bureaucracy can buttress rather than undermine trust(Sitkin 1995, pp. 209–210).

NUMMI appears to support Sitkin’s conjecture that or-ganizations can be simultaneously high in trust and highin degree of bureaucracy, and that in such organizations,bureaucracy will take a distinctive form. In a high-trustorganization, formalized and standardized procedures donot have to be defined by staff experts and imposed onreluctant subordinates; they can be defined jointly by em-ployees, specialists, and managers. They do not have toserve primarily as mechanisms for coercing effort andcompliance from recalcitrant employees; they can serveas the organizational memory of best practice. Horizontaland vertical specialization do not have to segment theorganization into rivalrous subunits and antagonistic lay-ers of positional authority; they can facilitate the accu-mulation of expertise and its optimal distribution, and thespecialized units and layers can function collaboratively.And finally, staff units do not have to function in oppo-sition to and domination over employees in the operatingcore; they can function as support specialists in a part-nership and consulting mode. In such organizations, in-terpersonal trust and “system trust”—trust in the consis-tency, competence, and congruence of managementstructures and systems—are complementary rather thansubstitutes (on system trust and related concepts, seeGiddens 1990, Luhman 1979, Zucker 1986, Shapiro1987).

Our case study also shows that system and interper-sonal trust were buttressed by a set of policies that createdincentives for management to act as if they trusted em-ployees and suppliers, and as if they were trustworthypartners. From a psychological point of view, affectivetrust, whether in individuals or systems, and its calcula-tive cousin are very different; indeed, as Williamson(1993) has argued, “calculative trust” may be a superflu-ous construct; but the NUMMI case shows how the af-fective forms of trust, both interpersonal and system, andits calculative forms can function in tandem.

Game theory and common sense suggest that there arelong-term costs to short-term opportunism, because overtime both parties find it valuable to maintain a reputationfor fair dealing. We see this relationship at work inNUMMI’s labor and supplier relations. NUMMI’s poli-cies of high training and employment security bolsteredemployees’ consistency trust, since they were more con-fident that NUMMI was in a situation of repeated play.NUMMI (like its parent Toyota) promoted informationsharing among its suppliers in a supplier association. The



Exhibit 2 A Typology of Organizationsestablishment of a mechanism to promote such commu-nication raised the cost to NUMMI of acting uncooper-atively with any individual supplier. In both cases,NUMMI bound itself in ways that made trusting andtrustworthy behavior more likely to be in its self-interest.

Most subtly, NUMMI bound itself to act in a trust-worthy fashion by relying so much on the ideas and com-mitment of its employees and suppliers. NUMMI knewthat short-sighted actions that would destroy trust wouldalso stop both the flow of production and of kaizen sug-gestions. The trust NUMMI exhibited in its workers andsuppliers served to establish a gift exchange relationshipthat bolstered congruence trust at an affective level. Inaddition, however, this trust also placed NUMMI in aposition of dependency. In game-theoretic terms,NUMMI designed its production process so that its needfor worker commitment and kaizen was a hostage it gaveto workers and suppliers. The presence of this implicithostage made it more likely that workers and supplierswould trust NUMMI to act as if it valued their compe-tence and their goals. Under such circumstances, calcu-lative trust was likely to engender affective trust. In thewords of George Nano, Bargaining Committee chair,

The key to NUMMI’s success is that management gave up someof its power, some of its traditional prerogatives. If managerswant to motivate workers to contribute and to learn, they haveto give up some of their power. If management wants workersto trust them, we need to be 50/50 in making the decision withthem. Don’t just make the decision and then say, “Trust me.”(quoted in Adler 1993, p. 180)

The centrality of trust suggests a typology of organi-zational forms shown in Exhibit 2. The lower part of thediagram focuses within the organization, and shows thatthe range of organization design alternatives is not a one-dimensional spectrum from organic to bureaucratic/mechanistic, but rather a two-dimensional matrix con-trasting high versus low extent of bureaucracy on onedimension and high versus low levels of trust on the otherdimension. The conventional one-dimensional contrastbetween organic and mechanistic forms appears as a di-agonal of this part of the matrix, since it is conventionallyassumed that the former will take a high-trust form andthe latter a low-trust form. The NUMMI case suggests thepossibility of a high-trust, high-bureaucracy configura-tion. The top row of the matrix suggests that in its rela-tions with other firms, the organization can operate ineither high-trust, partnership mode or low-trust, instru-mental mode.

On the more conventional view, attempts to shift theefficiency/flexibility tradeoff by combining organic andmechanistic forms within the organization and by mobi-lizing suppliers risk failure since both approaches try to

mix “oil”—the high-trust, organic, innovation subunitswithin the firm—with “water”—low-trust relations withexternal suppliers outside the firm, between the innova-tive subunits and the bureaucratic core of the firm, andwithin the bureaucratic core. But the NUMMI case sug-gests that it is possible to position all the players in thehigh-trust column of the matrix. This is the heart of the“cultural” transformation required by TQM (Hill 1995).In such a configuration, (partner-) suppliers can be mo-bilized to support the changeover process’s need for mu-tual adaptation; the (enabling-) bureaucratic core withhighly routinized tasks pursuing efficiency can cooperateeffectively with (enabling-) organic subunits pursuing in-novation and flexibility; employees and managers canswitch easily between these two very different kinds oftasks; and metaroutines for accelerating organizationallearning can be pursued collaboratively rather than im-posed.

ConclusionOur analysis of NUMMI has revealed it to be an orga-nization that has repeatedly shifted the efficiency/flexi-bility tradeoff. It had an exceptional capability for both



first-order and second-order learning. Four mechanismsmade this possible. Metaroutines made nonroutine tasksmore routine, with the direct effect of increasing effi-ciency for given levels of flexibility and the indirect effectof creating opportunities to increase flexibility. The othermechanisms—job enrichment, role switching betweenimprovement tasks and production tasks, and partitioningNUMMI’s structure into a changeover team and an op-erating core—had the converse effects: directly, they in-creased the organization’s capacity for flexibility at agiven level of efficiency, and indirectly they created ca-pabilities that served to improve efficiency.

Organization theory suggests important potential im-pediments to each of these mechanisms, but NUMMIovercame them because its bureaucratic core, its nonrou-tine components, and its supplier relations were all man-aged in a high-trust mode. Routines and metaroutineswere thus embraced rather than resisted. Organic and bu-reaucratic structures and roles were integrated rather thanopposed. Suppliers were mobilized rather than fended off.

Our contrast of the 1993 and 1995 model introductionsshows that an organization can continue over time tomove along this tradeoff-shifting vector of development.This is perhaps the greatest challenge for an organization,since there are so many incentives encouraging the or-ganization to veer off this vector and join the ranks ofmore conventional organizations that privilege one orother priority, either flexibility or efficiency.

Leadership would appear to be the key preconditionfor such persistence. Without committed leadership,NUMMI could not have made its huge investments intraining: investments that paid off only in the longer termand often only in indirect ways. Without that leadership,the pressures for short-term production performancewould have become much more salient to lower-levelmanufacturing managers than the need for flexibility andinnovation. Without a leadership that continually reas-serted the simultaneous importance of flexibility and ef-ficiency, lower-level managers would have likely slippedinto a more autocratic style in relations with subordinatesand suppliers, which would have in turn undermined con-gruence trust.

Given the fundamental role of leadership in maintain-ing such a development path, future research on ambi-dexterity might usefully focus on its challenges. InTushman and O’Reilly’s (1997) account, top manage-ment’s leadership looms large; but NUMMI presents acontext rather different from the ones they studied, sinceat NUMMI leadership was not just a prerogative and re-sponsibility of top management. As a high-trust, union-ized firm, the leadership function was somewhat moredistributed, and the union shared some of its burdens. And

to the extent that NUMMI’s ambidexterity relied on high-trust relations with suppliers, leadership in the supplierfirms too played a key role. Future research on the lead-ership challenges of ambidexterity might usefullybroaden its focus to include these more complex institu-tional settings.

Future research should also extend our temporal hori-zon. Our comparison of traditional Big Three practiceswith NUMMI practices masks an important evolutionover the longer time period. The flexibility/efficiencytradeoff has shifted greatly since the days when Ford’schangeover from the Model T to the Model A forced theplant to shut down for six months (Hounshell 1984, pp.266ff). Future research could usefully focus on delineat-ing the sequence of organizational innovations that haveprogressively shifted the tradeoff frontier. Some of theseinnovations lie in the domain of organizational structuresand processes, and some in technology; long-term shiftsin the broader institutional context may also have playeda facilitating role. A better understanding of this historymight give us more insight into the prospects for the fu-ture evolution of tradeoff-shifting organizations.

AcknowledgmentsThis research would not have been possible without the generous helpof UAW Local 2244, and of workers and managers at NUMMI andToyota. Financial support from the Sloan Foundation and the MITInternational Motor Vehicle Program is gratefully acknowledged. J. D.Power and Associates provided some data. We have benefited fromcomments by Jim Harbour, Oscar Hautpman, Don Gerwin, PaulLawrence, Ofer Meilich, George Strauss, Michael Tushman, DaveUpton, Clay Whybart, and the Organization Science reviewers.

Endnotes1Intriguing parallels to enrichment, switching, and partitioning appearin Maybury-Lewis’s (1989) analysis of the three “strategies” by whichsocieties manage incommensurate cultural antimonies: “integration”(read: enrichment), “alternation” (read: temporal segregation, orswitching), and “social or spatial segregation” (read: partitioning). Areviewer identified another parallel: if we postulate a homology be-tween shifting a tradeoff to improve performance and dealing withparadoxes to build better theory, our four tradeoff-shifting mechanismsparallel the four generic ways of using paradox identified by Poole andVan de Ven (1989). First, the paradox can be accepted and used con-structively: this is what firms do when they accept that routine andnonroutine tasks need to be managed differently and then develop me-taroutines that make the latter type of tasks more routine and thusincrease their efficiency. Second, the domains of reference can be “spa-tially” distinguished: this corresponds to partitioning. Third, the do-mains of reference can be “temporally” distinguished: this correspondsto switching. And finally, new terms can be introduced that change thebasic assumptions: this is what enrichment does to the assumption thatroutine work is essentially mindless.2Even though the flexibility/efficiency tradeoff is usually assumed toapply (in varying degrees) to all forms of flexibility, it is instructive to



position major model changes relative to the various theoretical typol-ogies of flexibility. Using Volberda’s (1996) strategy-oriented classi-fication, model changes are a form of operational rather than strategicflexibility, requiring high speed but dealing with only moderate variety.In Sethi and Sethi’s (1990) classification of types of operations systemsflexibility, model changes are a form of “product” and “process” flex-ibility, as distinct from routing, volume, and expansion types. InUpton’s (1994) framework for characterizing manufacturing flexibility,the dimensions of flexibility relevant here are product and process (asdistinct from input flexibility); the frequency of major model changesfor which this flexibility is required is a regular four-yearly cycle; therange of the product and process change is moderate, since even thoughmost parts will change, the overall architectures of the product and theprocess evolve only modestly; the mobility required is very high, giventhat highly compressed schedule that NUMMI targets for modelchangeovers; and the demand for uniformity of product quality duringthe changeover period is very high.

ReferencesAbernathy, W. J. 1978. The Productivity Dilemma: Roadblock to In-

novation in the Automobile Industry. Johns Hopkins UniversityPress, Baltimore, MD.

———, K. B. Clark, A. M. Kantrow. 1983. Industrial Renaissance.Basic Books, New York.

Adler, P. S. 1988. Managing flexible automation. California Manage-ment Review, Spring 34–56.

———, ed. 1992. Technology and the Future of Work. Oxford Uni-versity Press, New York.

———. 1993. The learning bureaucracy: New United Motors Manu-facturing, Inc. Barry M. Staw and Larry L. Cummings, eds. Re-search in Organizational Behavior. JAI Press, Greenwich, CT.15 111–194.

———. 1995. Interdepartmental interdependence and coordination:The case of the design/manufacturing interface. Organization Sci-ence 6 147–167.

———, B. Borys. 1996. Bureaucracy: Coercive versus enabling. Ad-ministrative Science Quarterly 41 1 61–89.

———, R. Cole. 1993. Designed for learning: A tale of two auto plants.Sloan Management Review 34 3 85–94.

———, B. Goldoftas, D. Levine. 1997. Ergonomics, employee in-volvement, and the Toyota production system: A case study ofNUMMI’s 1993 model introduction. Industrial and Labor Rela-tions Review 50 3 416–437.

Argyris, Chris, Donald Schon. 1978. Organizational Learning. Addi-son-Wesley, Reading, MA.

Bowen, David E., Edward E. Lawler, III. 1992. The empowerment ofservice workers: What, why, how and when. Sloan ManagementReview. 33 3 31–40.

———, ———. 1995. Empowering service employees. Sloan Man-agement Review. Summer 73–84.

Bowen, H. Kent, Sylvie Ryckebusch. 1996. Injex Industries. HarvardBusiness School case 9-696-090, Boston, MA.

Brush, Thomas, Aneel Karnani. 1996. Impact of plant size and focuson productivity: An empirical study. Management Science. 42 71065–1081.

Burns, Tom, George M. Stalker. 1961. The Management of Innovation.Tavistock, London, England.

Bushe, Gervase R., A. B. Shani. 1991. Parallel Learning Structures:Increasing Innovation in Bureaucracies. Addison-Wesley, Read-ing, MA.

Business Week. 1994. Motown’s struggle to shift on the fly. July 11,111–112.

Clark, Kim B. 1996. Competing through manufacturing and the newmanufacturing paradigm: Is manufacturing strategy passe? Pro-duction and Operations Management. 5 42–58.

———, Bruce Chew, Takahiro Fujimoto. 1992. Manufacturing for de-sign: Beyond the production/R&D dichotomy. Gerald I. Susman,ed. Integrating Design and Manufacturing for Competitive Ad-vantage. Oxford University Press, New York. 178–206.

———, Takahiro Fujimoto. 1991. Product Development Performance.Harvard Business School Press, Boston, MA.

Correa, Henrique Luiz. 1994. Linking Flexibility, Uncertainty andVariability in Manufacturing Systems. Averbury, Aldershot, En-gland.

Daft, Richard. 1998. Essentials of Organization Theory and Design.South-Western College Publishing, Cincinnati, Ohio.

de Meyer, Arnould, Jinichiro Nakane, Jeffrey G. Miller, KasraFerdows. 1989. “Flexibility: The next competitive battle. Themanufacturing futures survey. Strategic Management Journal. 10135–144.

Dougherty, Deborah. 1992. Interpretive barriers to successful productinnovation in large firms. Organization Science 3 179–202.

Duncan, Robert B. 1976. The ambidextrous organization: Designingdual structures for innovation. in Ralph H. Kilman, Louis R.Pondy, Dennis P. Slevin, eds. The Management of OrganizationDesign. North-Holland, New York, vol. 1, 167–188.

Eisenhardt, Kathleen M. 1989. Agency theory: An assessment and re-view. Academy of Management Review. 14 57–74.

Ferdows, Kasra, Arnoud de Meyer. 1990. Lasting improvements inmanufacturing performance. Journal of Operations Management.9 168–184.

Fleischman, Charles A. 1996. The endogeneity of employment adjust-ment costs: The tradeoff between efficiency and flexibility. Fi-nance and Economics Discussion Series, no. 1996–48, Divisionsof Research & Statistics and Monetary Affairs, Federal ReserveBoard, December.

Garud, Raghu, Arun Kumaraswamy. 1995. Technological and orga-nizational designs for realizing economies of substitution. Stra-tegic Management Journal. 16 93–109.

Gerwin, Donald. 1993. Manufacturing flexibility: A strategic perspec-tive. Management Science. 39 395–408.

Ghemawat, Pankaj, Joan E. Ricart I Costa. 1993. The organizationaltension between static and dynamic efficiency. Strategic Man-agement Journal. 14 59–73.

Giddens, A. 1990. The Consequences of Modernity. Stanford Univer-sity Press, Stanford, CA.

Goldhar, Joel D., Mariann Jelinek. 1983. Plan for economies of scope.Harvard Business Review 61 6 141–148.

Goldstein, S. G. 1985. Organizational dualism and quality circles.Academy of Management Review. 10 504–17.

Grant, Robert M. 1996. Prospering in dynamically-competitive envi-ronments: Organizational capability as knowledge integration.Organization Science 7 375–387.



Hackman, J. Richard, Greg R. Oldham. 1980. Work Redesign. Addi-son-Wesley, Reading, MA.

Hannan, Michael T., John Freeman. 1977. The population ecology oforganizations. American Journal of Sociology 82 929–964.

———, ———. 1989. Organizational Ecology. Harvard UniversityPress, Cambridge, MA.

Hart, A. G. 1942. Risk, uncertainty, and the unprofitability of com-pounding probabilities. H. Schultz, O. Lange, F. McIntyre, eds.Studies in Mathematical Economics and Econometrics. Univer-sity of Chicago Press, Chicago, IL. 110–118.

Hayes, Robert H., Gary P. Pisano. 1996. Manufacturing strategy: Atthe intersection of two paradigm shifts. Production and Opera-tions Management. 5 25–41.

———, S. C. Wheelwright. 1984. Restoring Our Competitive Edge:Competing Through Manufacturing. John Wiley & Sons, NewYork.

Helper, Susan. 1990. Comparative supplier relations in the US andJapanese auto industries. Business and Economic History. SecondSeries. 19 153–162.

———, David I. Levine. 1992. Long-term supplier relations and prod-uct market structure. Journal of Law, Economics and Organiza-tion. 8 561–581.

Hill, Stephen. 1995. From quality circles to total quality management.Adrian Wilkinson, Hugh Willmott, eds. Making Quality Critical:New Perspectives on Organizational Change. Routledge, NewYork and London, England. 33–53.

Hounshell, David A. 1984. From the American System to Mass Pro-duction 1800–1932. Johns Hopkins University Press, Baltimore,MD.

Jelinek, Mariann, Claudia Bird Schoonhoven. 1993. The InnovationMarathon. Jossey-Bass, San Francisco, CA.

Kekre, S., Kannan Srinivasan. 1990. Broader product line: A necessityto achieve success? Management Science 36 1216–1231.

Klein, Burton H. 1984. Prices, Wages and Business Cycles: A DynamicTheory. Pergamon, New York.

Knott, Anne Marie. 1996. Do Managers Matter? Unpublished disser-tation, UCLA.

Krafcik, John. 1989. A new diet for US manufacturing: The auto in-dustry enters the 1990s. Technology Review. January 28–39.

Kurke, Lance B. 1988. Does adaptation preclude adaptability? Strategyand performance. L. G. Zucker, ed. Institutional Patterns and Or-ganizations: Culture and Environment. Ballinger, Cambridge,MA. 199–222.

Langer, E. J. 1989. Minding matters: The consequences of mindless-ness/mindfulness. Advances in Experimental Social Psychology.22 137–173.

Lawler, Edward E., III, Susan A. Mohrman 1985. Quality circles afterthe fad. Harvard Business Review. 85 1 64–71.

Lawrence, Paul R., Jay W. Lorsch. 1967. Organization and Environ-ment: Managing Differentiation and Integration. Harvard Uni-versity Graduate School of Business Administration, Boston, MA.

Ledford, Gerald E., Jr., Edward E. Lawler III, Susan A. Mohrman.1988. The quality circle and its variations. J. P. Campbell, R. J.Campbell, eds. Productivity in Organizations: New Perspectivesfrom Industrial and Organizational Psychology. Jossey-Bass, SanFrancisco, CA. 255–294.

Levine, David I. 1995. Reinventing the Workplace: How Business and

Employees Can Both Win. Brookings Institution, Washington,DC.

Levinthal, Daniel, James March. 1993. Myopia of learning. StrategicManagement Journal. 14 S2 95–112.

Lincoln, James R., Kalleberg, Arne L. 1991. Culture, Control andCommitment. Cambridge University Press, Cambridge, MA.

Louis, Meryl Reis, Robert I. Sutton. 1991. Switching cognitive gears.Human Relations. 44 1 55–74.

Luhmann, N. 1979. Trust and Power. Wiley, Chichester, New York.MacDuffie, John Paul, Kannan Sethuraman, Marshall L. Fisher. 1996.

Product variety and manufacturing performance: Evidence fromthe International Automotive Assembly Plant study. ManagementScience 42 3 350–369.

———, Thomas A. Kochan. 1995. Do US firms invest less in humanresources? Training in the world auto industry. Industrial Rela-tions 34 2 147–168.

March, James. 1991. Exploration and exploitation in organizationallearning. Organization Science 2 71–87.

Maybury-Lewis, David. 1989. The quest for harmony. David Maybury-Lewis, Uri Almagor, eds. The Attraction of Opposites: Thoughtand Society in the Dualistic Mode. University of Michigan Press,Ann Arbor, MI. 1–18.

Mayer, Robert C., James H. Davis, F. David Schoorman. 1995. Anintegrative model of organizational trust. Academy of Manage-ment Review 20 3 709–734.

McDonough, Edward F., III, Richard Leifer. 1983. Using simultaneousstructures to cope with uncertainty. Academy of ManagementJournal. 26 727–735.

McGill, Michael E., John W. Slocum, Jr., David Lei. 1992. Manage-ment practices in learning organizations. Organization Dynamics.Summer, 5–17.

McGregor, Douglas. 1960. The Human Side of Enterprise. McGraw-Hill, New York.

Merton, Robert K. 1958. Bureaucratic structure and personality. SocialTheory and Social Structure. Free Press, New York. 3rd ed., 249–260.

Miles, Raymond E., Charles C. Snow. 1978. Organizational Strategy,Structure, and Process. McGraw-Hill, New York.

Miller, E. C. 1978. The parallel organization structure at General Mo-tors: An interview with Howard C. Carlson. Personnel. 55 4 64–69.

Mintzberg, Henry. 1979. The Structuring of Organizations. Prentice-Hall, Englewood Cliffs, NJ.

Mishra, Aneil K. 1996. Organizational responses to crisis: The cen-trality of trust. Roderick M. Kramer, Tom R. Tyler, eds. Trust inOrganizations. Sage, Thousand Oaks, CA. 261–287.

Monden, Y. 1983. Toyota Production System. Institute of IndustrialEngineers, Atlanta, GA.

Nelson, Richard R., Sidney G. Winter. 1982. An Evolutionary Theoryof Economic Change. Belknap/Harvard University Press, Cam-bridge, MA.

Nonaka, I., H. Takeuchi. 1995. The Knowledge-Creating Company.Oxford University Press, New York.

Ohno, Taichi. 1988. Toyota Production System: Beyond Large-ScaleProduction. Productivity Press, Cambridge, MA.

Organization Science. 1996. Special Issue on Hypercompetition. 7 3and 4.



Pfeffer, Jeffrey. 1978. Organization Design. AHM, Arlington Heights,IL.

Pil, Frits K., John Paul MacDuffie. 1996. Japanese and local influenceson the transfer of work practices at Japanese transplants. PaulaVoos, ed. Proceedings of the Forty-eighth Annual Meeting. In-dustrial Relations Research Association, Madison, WI. 278–297.

Pine, B. Joseph, III. 1993. Mass Customization. Harvard BusinessSchool Press, Boston, MA.

Poole, Marshall Scott, Andrew H. Van de Ven. 1989. Using paradoxto build management and organization theories. Academy of Man-agement Review. 14 4 562–578.

Porter, Michael E. 1980. Competitive Strategy. Free Press, New York.Power, J. D., and Associates. 1995. Japanese and domestic launches.

archival data.Safizadeh, M. Hossein, Larry P. Ritzman, Deven Sharma, Craig Wood.

1996. An empirical analysis of the product-process matrix. Man-agement Science. 42 11 1576–1591.

Sako, Mari. 1992. Prices, Quality and Trust: Inter-firm Relations inBritain and Japan. Cambridge University Press, Cambridge, En-gland.

Schon, Donald A. 1983. The Reflective Practitioner: How Profession-als Think in Action. Basic Books, New York.

Schumpeter, Joseph. 1976. Capitalism, Socialism and Democracy.Harper, New York.

Scott, W. Richard. 1992. Organizations: Rational, Natural and OpenSystems. Prentice Hall, Englewood Cliffs, NJ. 3rd. ed.

Sethi, Andrea K, Suresh P. Sethi. 1990. Flexibility in manufacturing:A survey. International Journal of Flexible Manufacturing Sys-tems. 2 4 289–328.

Shapiro, Susan P. 1987. The social control of impersonal trust. Amer-ican Journal of Sociology. 93 3 623–658.

Sitkin, Sim B. 1995. On the positive effect of legalization of trust.Research on Negotiations in Organizations. 5 185–217.

Skinner, Wickham. 1985. Manufacturing: The Formidable Competi-tive Weapon. John Wiley, New York.

———. 1996. Manufacturing strategy on the ‘s’ curve. Production andOperations Management. 5 1 3–14.

Stein, Barry A., R. Moss Kanter. 1980. Building the parallel organi-zation: Creating mechanisms for permanent quality of work life.Journal of Applied Behavioral Science. 16 3 371–388.

Stigler, George. 1939. Production and distribution in the short run.Journal of Political Economy. 47 305–327.

Suarez, Fernando F., Michael Cusumano, Charles Fine. 1996. An em-pirical study of manufacturing flexibility in printed circuit boardassembly. Operations Research. 44 1 223–240.

Thompson, J. D. 1967. Organizations in Action. McGraw Hill, NewYork.

Tushman, Michael L., Charles A. O’Reilly, III. 1997. Winning ThroughInnovation. Harvard Business School Press, Boston, MA.

Upton, David M. 1994. The management of manufacturing flexibility.California Management Review. 36 2 72–89.

Victor, Bart, Andrew Boynton, Theresa S. Jahng. 1998. The effectiveorganization of work under Total Quality Management. Organi-zation Science. forthcoming.

Volberda, Henk W. 1996. Toward the flexible form: How to remainvital in hypercompetitive environments. Organization Science. 74 359–374.

Walton, Richard E. 1985. Toward a strategy of eliciting employee com-mitment based on policies of mutuality. Richard E. Walton, PaulR. Lawrence, eds. HRM Trends and Challenges. Harvard Busi-ness School Press, Boston, MA. 119–218.

Weick, Karl. 1969. The Social Psychology of Organizing. Addison-Wesley, Reading, MA.

Williamson, Oliver E. 1993. Calculativeness, trust and economic or-ganization. Journal of Law and Economics. April 453–486.

Womack, James, Daniel Jones, Daniel Roos. 1990. The Machine thatChanged the World. Rawson Associates, MacMillan PublishingCompany, New York.

Zammuto, Raymond F., Edward J. O’Connor. 1992. Gaining advancedmanufacturing technologies’ benefits: The roles of organizationdesign and culture. Academy of Management Review. 17 4 701–728.

Zand, D. 1974. Collateral organization: A new change strategy. Journalof Applied Behavioral Science. 10 63–89.

Zucker, Lynne G. 1986. Production of trust: Institutional sources ofeconomic structure, 1840–1920. Research in Organizational Be-havior. 8 53–111.

Accepted by Arie Y. Lewin; received January 1997. This paper has been with the authors for one revision.

Documents

Flexibility Versus Efficiency? A Case Study of Model Changeovers …faculty.marshall.usc.edu/Paul-Adler/research/FlexVsEff-2.pdf · 2019. 5. 3. · broader product line was signiﬁcantly