Deliberate Learning and the Evolution of Dynamic Capabilities

8/6/2019 Deliberate Learning and the Evolution of Dynamic Capabilities

1/14

Deliberate Learning and the Evolution ofDynamic Capabilities

Maurizio Zollo Sidney G. WinterDepartment ofStrategy and Management. INSEAD. 77305 Fontainebleau. France

The Wharton School, Department ofManagement, University of Pennsylvania, Philadelphia, Pennsylvania [email protected] [email protected]

AbstractThis paper investigates the mechanisms through which orga-nizations develop dynamic capabilities, defined as routinizedactivities directed to the development andadaptation of oper-ating routines. It addresses the role of (I) experience accumu-lation. (2) knowledge articulation, and (3) knowledge codifi-cation processes in the evolution of dynamic, as well asoperational, routines. Theargument is made thai dynamic ca-pabilities are shaped by the coevolution of these leaming mech-anisms. At any point in time, firms adopt a mix of learningbehaviors constituted by a semiautomatic accumulation of ex -perience and by deliberate investments in knowledge articula-tion and codification activities. The relative effectiveness ofthese capability-building mechanisms is analyzed here as con-tingent upon selected features of the task to be learned, such asi ts frequency, homogeneity, and degree of causal ambiguity.Testable hypotheses about these effects are derived. Somewhatcounterintuitive implications of the analysis include the rela-tively superior effectiveness of highly deliberate learning pro-cesses such as knowledge codification at lower levels of fre-quency and homogeneity of the organizational task, incontrastwith common managerial practice.{Organizational Learning. Dynamic Capabilities: Organizational Routines:Knowledge Codification: Knowledge Articulation: Learning by Doing; TaskFrequency: Task Heterogeneity. Causal Ambiguity)

IntroductionExplaining the vatiation in the degree of success of busi-ness organizations by reference to different degrees andqualities of organizational knowledge and competencehas been a major focus of recent theorizing in both stra-tegic management and organizational theory. Conceptsand labels coined to characterize the phenomenonabound. From pioneering efforts such as Selznick's(1957) "distinctive competence," to the more recent andrefined notions of organizational routines (Nelson and

Winter 1982), absorptive capacity (Cohen and Levintha1990), architectural knowledge (Henderson and Clar1990), combinative capabilities (Kogut and Zander 1992and, finally, dynamic capabilities (Teece et al. 1997)there aredecades of investment in sorting out the traitan d the boundaries of the phenomena. Recent contributions (Eisenhardt and Martin 2000, Dosi et al. 2000) aimat clarifying distinctions a mon g the various constructs. Athe field p rogresses in the characterization of the ph enomena, however, the need for a better understanding of thorigins of capabilities becomes increasingly apparentWhat ultimately accounts for the fact that one organization exhibits "competence" in some area, while anothedoes not? And how do we explain the growth and decaof that particular competence, other than the simple repetition, or lack thereof, of decisions and behavior?

This paper sets forth a theoretical account of the genesis and evolution of dynamic capabilities that focuses othe role of several leaming mechanisms and their interaction with selected attributes of the organizational tasbeing learned. We begin our discussion by presentinggenera! framework linking learning mechanisms to thevolution of dynamic capabilities, and these to the evolution of operating routines. We then draw on evolutionary economics to develop a more fine-grained representation of how organizational knowledge evolves at thlevel of operating routines as well as of dynamic capabilities. In 3, wediscuss the different learning mechanisms in terms of a spectrum of learning investments anidentify the contextual features that shape their roles. Fnally, in4, we explore indetail how some key attributeof the organizational task under consideration affect threlative effectiveness of these learning mechanisms anddevelop testable hypotheses for future empirical work.1. Organizational Learning andDynamic CapabilitiesOur first objective is to develop a framework describinthe linkages among the processes we intend to study. W


2/14

M AURI ZI O ZOLLO AND SI DNEY G. W I NTER Deliberate Learning and the Evolution of Dynamic Capabilities

describe a set of learning mechanisms encompassing boththe relatively passive experiential processes of learningC'by doing") and more deliberate cognitive processeshaving to do with the articulation and codification of col-lective knowledge. These learning processes are respon-sible for the evolution in time of two sets of organiza-tional activities: one geared towards the operationalfunctioning of the firm (both staff and line activities),which we will refer to as operating routines; the otherdedicated to the modification of operating routines, w hichwe identify with the notion of dynamic capabilities.

Below, we provide a more detailed description of howwe conceive of dynamic capabilities and of the functionof the identified leaming mechanisms.Dynamic CapabilitiesBuilding on Nelson and Winter's view of the organizationas a set of interdependent operational and administrativeroutines which slowly evolve on the basis of performancefeedbacks, Teece et al. (1997) define the concept of "dy-namic capabilities" as "the firm's ability to integrate,build, and reconfigure intemal and external competenciesto address rapidly changing environments" (p. 516).While this suggests something of what dynamic capabil-ities are for and how they work, it leaves open the ques-tion of where they come from. Also, the definition seemsto require the presence of "rapidly changing environ-ments" for the existence of dynamic capabilities, butfirms obviously do integrate, build, and reconfigure theircompetencies even in environments subject to lower ratesof change. We propose the following altemative:

D E F I N I T I O N . A dynamic capability is a leamed and sta-ble pattem of collective activity through which the or-ganization systematically generates and modifies its op-erating routines in pursuit of improved effectiveness.In addition to avoiding the near tautology of definingcapability as ability, this definition has the advantage ofspecifically identifying operating routines, as opposed tothe more generic "competencies," as the object on whichdynamic capabilities operate. Also, it begins to spell outsome of the characteristics of this construct. The words"learned and stable pattem" and "systematically" high-light the point that dynam ic capa bilities are structured andpersistent; an organization that adapts in a creative butdisjointed way to a succession of crises is not exercisinga dynamic capability. Dynamic capability is exemplifiedby an organization that adapts its operating processesthrough a relatively stable activity dedicated to processimprovements. Another example is given by an organi-zation that develops from its initial experiences with ac-quisitions or joint ventures a process to manage such proj-ects in a systematic and relatively predictable fashion.

The ability to plan and effectively execute postacquisitionintegration processes is another example of a dynamiccapability, as it involves the modification of operatingroutines in both the acquired and the acquiring unit.Figure 1, below, offers a generic picture of the orga-

nizational processes under consideration and of the link-ages between them. Dynamic capabilities arise fromlearning; they constitute the firm's systematic methodsfor modifying operating routines. To the extent that thelearning mechanisms are themselves systematic, theycould (following Collis 1994) be regarded as "second-order" dynamic capabilities. Leaming mechanisms shapeoperating routines directly as well as by the intermediatestep of dynamic capabilities.Learning MechanismsWhat mechanisms are involved in the creation and evo-lution of dynamic capabilities? What features distinguishan organization capable of systematically developing newand enhanced understanding of the causal linkages be-tween the actions it takes and the performance outcomesit obtains?

In the spirit of bridging the behavioral and cognitiveapproaches to the organizational leaming phenomenon(Glynn et al. 1994), we attend both to the experience ac-cumulation process and to more deliberate cognitive pro-cesses involving the articulation and codification ofknowledge derived from refiection upon past experi-ences.' The three mechanisms focal to this analysis areintroduced and discussed .separately below.

Organizational Routines and Experience Accum ula-tion. Routines are stable pattems of behavior that char-acterize organizational reactions to variegated, intemal orexternal stimuli. Every time an order is received from a

Figure 1 Learning, Dynamic Capabilities, and Operating Rou-tinesLEARNING MECHANISMSExperience accumulationKnowledge articulationKnowledge codification

DYNAMIC CAPABILITIESProcess R&DRestructuring, re-engineeringPost-acquisition Integration

EVOLUTION OF OPERATING ROUTINES


3/14

MAURIZIO ZOLIX) AND SIDNEY C. WINTER Deliberate Learning an d the Evolitlion of Dynamic Capabilities

customer or a decision is made to upgrade a productionproces s, for instance, a host of predictable and interrelatedactions are initiated which will eventually conclude withthe shipping of the ordered goods (and receipt of corre-sponding payment) or with the launch of the new pro-duction system. In spite of the superficial similarity be-tween these two examples, though, the two patterns ofbehavior present a theoretically relevant distinction. Thefirst type of routine involves the execution of known pro-cedures for the purpose of generating current revenue andprofit, while the second seeks to bring about desirablechang es in the existing set of operating in this case, pro-ductionroutines for the purpose of enhancing profit inthe future. Routines of the second type are traditionallyidentified as search routines in evolutionary economics(Nelson and Winter 1982), and are here regarded as con-stitutive of dynamic capabilities.Routines of the two types have different effects on thegeneration and appropriation of rents, depending on thepace of change in the environment. Of course, effectiveoperating routines are always a necessity, and superioroperating routines are always a source of advantage. In arelatively static environment, a single leaming episodemay suffice to endow an organization with operating rou-tines that are adequate, or even a source of advantage, foran extended period. Incremental improvements can be ac-complished through the tacit accumulation of experienceand sporadic acts of creativity. Dynamic capabilities areunnecessary, and if developed may prove too costly tomaintain. But in a context where technological, regula-tory, and competitive conditions are subject to rapidchange, persistence in the same operating routinesquickly becomes hazardous. Systematic change effortsare needed to track the environmental change; both su-periority and viability will prove transient for an organi-zation that has no dynamic capabilities. Such capabilitiesmust themselves be developed through learning. Ifchange is not only rapid but also unpredictable and vari-able in direction, dynamic capabilities and even thehigher-order learning approaches will themselves need tobe updated repeatedly. Failure to do so turns core com-petencies into core rigidities (Leonard Barton 1992).

To our knowledge at least, the literature does not con-tain any attempt at a straightforward answer to the ques-tion of how routinesmuch less dynamic capabilitiesare generated and evolve. That is hardly surprising, for itseems clear that routines often take shape at the conjunc-tion of causal processes of diverse kindsfor example,engineering design, skill development, habit formation,particularities of context, negotiation, coincidences, ef-forts at imitation, and ambitions for control confronting

aspirations for autonomy. In general, however, the liteature that is relatively explicit on the question suggesthat "routines reflect experiential wisdom in that they arthe outcome of trial and error learning and the selectioand retention of past behaviors" (Gavetti and Levintha2000. p. 113). Tbis view is linked to an emphasis on thimportance of tacit knowledge, since tacitness arisewhen learning is experien tial. Tt is also linked to the notion that organizational routines are stored as procedurmemory (Cohen and Bacdayan 1994) and to the imagof routinized response as quasi-automatic. In addition, is consistent with the traditional view of organizationlearning as skill building based on repeated execution osimilar tasks that is implicit in much of the empirical lierature on leaming curves (e.g., Argote 1999).

We incorporate this view in our discussion here, u.sinthe term "experience accumulation" to refer to the centrleaming process by which operating routines bave tradtionally been thought to develop. In adopting this usagwe seek to provide a clear baseline for our analysis osome of the other leaming processes tbat shape routineand dynamic capabilities. With respect to the latter, seems clear that a theory of their development and evolution must invoke mechanisms that go beyond semautomatic stimulus-response processes and tacit accumulation of experience.

Knowledge Articulation. One of the recognized limtations of the behavioral tradition in the study of organzational leaming consists of the lack of appreciation othe deliberative process througb which individuals angroups figure out what works and what doesn't in thexecution of a certain organizational task (Cangelosi anDill 1965. p 196; Levinthal and M arch 1988, p. 20Narduzzo et ai. 2000). Important collective learning happens when individuals express their opinions and beliefengage in constructive confrontations and challenge eacothe r's viewpo ints (Argyris and Schon 1978, Dunca n anWeiss 1979). Organizational competence improves members of an organization become more aware of tboverall performance implications of their actions, and the direct consequence of a cognitive effort more or leexplicitly directed at enhancing their understanding these causal links. We therefore direct attention to a seond mechanism of development of collective comptence, tbe process through whicb implicit knowledge articulated through collective discussions, debriefing sesions, and performance evaluation processes. By sharintbeir individual experiences and comparing their opinionwitb those of their colleagues, organization members caachieve an improved level of understanding of the causmechanisms intervening between the actions required


4/14

M AURI ZI O ZOLLO AND S I DNEY G. W I NTER Deliberate L earning ant! the Evolution of Dynam ic Capab ilities

execute a certain task and the performance outcomes pro-duced. Organizational processes are often subject to sig-nificant causal ambiguity with respect to their perfor-mance implications (Lippman and Rumelt 1982), andparticularly .so in rapidly changing environmental con-texts. Higher-level cognitive efforts and a more deliberatecollective focus on the learning challenge can help to pen-etrate tbe ambiguityalthough some part of it alwayspersists. It is important to note that only a small fractionof articulable knowiedge is actually articulated, and thatorganizations differ substantially on the degree to whichthey transform potentially articulable knowledge into ar-ticulated .statements (Winter 1987. Kogut and Zander1992, Cowan, et al. 2000). While potentially requiringsignificant efforts and commitment on the part of themem bers of the organization, such articulation efforts canproduce an improved understanding of the new andchanging act ion-pertbrm ance links, and therefore resultin adaptive adjustments to tbe exi.sting sets of routines orin enhanced re cognition of tbe need for more fundamentalchange.

Knowledge Codification. An even higher level of cog-nitive effort is required when individuals codify their un-derstandings of the perfbmiance implications of intemalroutines in written tools, such as manuals, blueprints,spreadsheets, decision support systems, project manage-ment software, etc. Knowledge codification is a step be-yond knowledge articulation. The latter is required in or-der to achieve the former, w hile the opposite is obviouslynot true. The fact that in most cases articulated kno wledgeis never codified bears witness to additional costs incurredwhen stepping up the learning effort from a simple shar-ing of individual experience to developing manuals andother process-specific tools.

Note that while some of tbese tools are deliberatelyaimed at uncovering the linkages between actions andpertbrmance outcomes (such as performance appraisals,postmortem audits, etc.), most of them are intended sim-ply to provide guidelines for the execution of future tasks.Wh atever the intentions motivating the codification ef-fort, the process through which these tools are created andconsistently updated implies an effort to understand thecausal links between tbe decisions to be made and theperformance outcomes to be expected, even though learn-ing might not be the deliberate goal of the codificationeffort. To illustrate, both the readers and the authors ofthis article have certainly experienced tbe significant in-crease in the clarity of their ideas consequent to the actof writing the first draft of a research paper. Through thewriting process, one is forced to expose the logical stepsof one's arguments, to unearth the hidden assumptions.

and to make the causal linkages explicit. Similarly, agroup of individuals who are in the process of writing amanual or a set of written guidelines to improve the ex-ecution of a complex task (think of the development of anew product, or the management of the postacquisitionintegration process) will most likely reach a significantlyhigher degree of understanding of what makes a certainprocess succeed or fail, compared to simply telling "warstories" or discussing it in a debriefing session.

Knowledge codification is, in our view, an importantand relatively underemphasized element in the capability-building picture. The literature has emphasized that cod-ification facilitates tbe diffusion of existing knowledge(Winter 1987, Zander and Kogut 1995, Nonaka 1994), aswell as the coordination and implementation of complexactivities. Having identified and selected the change inthe operating routines or the new routine to be estab-lished, the organization should create a manual or a toolto facilitate its replication and diffusion. The principalbenefits to the codification effort are seen as coming fromthe successful use of the manual or tool.While tbis picture is clearly accurate in many cases, itmay also exaggerate some benefits of codification whileneglecting others. It overlooks, for example, importantobstacles to the success of this type of use of codifiedknowledge, such as the difficulty of assuring that the cod-ified guidance is both adequate and actually implem ented.More importantly, this appraisal overlooks the leaming

benefits of the activities necessary for the creation anddevelopment of the tools. To develop a manual for theexecution of a complex task, the individuals involved inthe process need to form a mental model of what actionsare to be selected under what conditions. By goingthrough that effort, they will most likely emerge with acrisper definition of what works, what doesn't work, andwhy.Codification, therefore, is potentially important as asupporting mechanism for the entire knowledge evolutionprocess, not just the transfer phase. It can, for instance,facilitate the generation of new proposals to change thecurrently available routines, as well as the identificationof tbe strengths and the weaknesses in the proposed var-iations to the current set of routines. The cognitive sim-plification inherent in the act of synthesizing on paper (orin a computer program) the logic behind a set of instruc-tions can therefore represent both an economizing ondata-processing requirements, allowing more effectivedecisionm aking (Boisot 1998. Gavetti and Levinthal2000), and a more-or-less deliberate form of retrospectivesense-making with respect to the performance implica-tions of a given set of activities (We ick 1979 and 1 995).This sense-making typically draws upon a specialized


5/14

M AURI ZI O ZOLLO AND S I DNEY G. W I NTER Deliberate Learning and Ihe Evolution of Dynamic Capahiliiies

language and causal understanding developed for and byother codification efforts of a similar type (Cowan andForay 1997, Cowan et al. 2000).To be sure, these advantages do not come for free.There are specific costs attached to the knowledge codi-fication process. Direct costs include the time, the re-sources, and the m anagerial attention to be invested in thedevelopment and updating of task-specific tools, whileindirect costs include a possible increase in the rate of"misfire" or inappropriate application of the routine(Cohen and Bacdayan 1994) if the codification is poorlyperformed, and the more general increase in organiza-tional inertia consequent to the formalization and struc-turing of the task execution. Concem with these costs islegitimate and familiar. A long debate, dating back toWeber's work, has engaged organizational theorists withrespect to tbe advantages and disadvantages of formali-zation. a kindred phenomenon to knowledge codification.For a long time, the skeptics seemed to have the upperhand. More recently, there seems to be increasing will-ingness to see the formalization of operating routines ina positive light; e.g., as sometimes capable of producingan "enabling" rather than a "coercive" bureaucracy(Adler and Borys 1996).

Tbis contingency approach is consistent with the thrustof the current analysis: Instead of rejecting knowledgecodification processes tout court as producers of inertialforces, our objective is to determine the conditions underwhich the leaming and diffusion advantages attached tocodification could more than offset its costs. We acknow l-edge, of course, that this is not likely to be true in allcasesand that codification, like many other things, islikely to produce bad results when done badly. In theconcluding section, we discuss what seems to be involvedin doing it well.

2. The Cyclical Evolution ofOrganizational KnowledgeHaving offered our view on tbe notion of dynamic ca-pabilities and the leaming processes that might supporttheir evolution, we now provide a more detailed descrip-tion of the way in which dynamic capabilities and oper-ating routines evolve in time.

Figure 2 offers a graphical view of a "knowledge evo-lution cycle." It is a simple account of the developmentof collective understanding regarding the execution of agiven organizational task. We have adapted for our pur-poses the classic evolutionary paradigm of variation-selection-retention."Organizational knowledge is here described as evolv-

ing through a series of stages chained in a recursive cycle.

Figure 2 Activities in the Knowledge Evolution CycleGENERATIVEVARIATIONScanning,Recombination

ExtemalStimuli &Feedback

RETENTIONEnactmentRoutinization

INTERNALSELECTIONEvaluationLegitimization

REPLICATIONKn. Sharing/TransfeAdaptive VariationProblem Solving

The logical point of departure for each cycle lies in thvariation stage, where individuals or groups of them generate a set of ideas on how to approach old problems inovel ways or to tackle relatively new challenges. Thhappens on the basis of a combination of external stimu(competitors" initiatives, normative changes, scientifdiscoveries, etc.) with internally generated informatioderived from the organization's existing routines, anmay involve substantial creativity. These sets of ideainitially in embryonic and partly tacit form, are then suject to internal selection pressures aimed at the evaluatioof their potential for enhancing the effectiveness of existing routines or the opportunity to form new one(Nonaka 1994). These pres sures arise as new ideas aconsidered in relation to a shared understanding of tborganization's prior experience, as well as in the contexof established power structures and existing legitimiztion processes. Hypotheses as to the expected advantagefrom the application of the proposed changes are probethrough a collective investment in articulation, analysiand debate of the merits and risks connected to the intiative.

The third phase of the cycle refers to the set of activitieenacted by an organization for the purpose of diffusinthe newly approved change initiatives to the relevant paties within the firm. This diffusion process requires thspatial replication of the novel solutions in order to leerage the newly found wisdom in additional competitivcontexts (see Winter and Szulanski 2001). This mechnism is a relatively new addition to tbe standard variatioselection-retention triumvirate of evolutionary modelinborrowe d from evolutionary biology (Cam pbell 1969). Irationale lies in the simple observation that organizationdiffer in a crucial way from biological entities in tbat theact simultaneously in spatially diverse contexts, posinthe question of how novelties (ideas or change proposa


6/14

M AURI ZI O ZOLL O AND S I DNEY G. W I NTER Deliberate Learning and the Evolution of Dynamic Capabilities

in our case) that survive intemal selection processes dif-fuse to the spatially dispersed parts of the organizationswhere the novel approach can be put to use."*The replication phase, however, does not simply servethe function of diffusing tbe newly minted knowledge

to the places and times where it is needed. Importantly forthe purposes of this paper, the diffusion of knowledgethrough replication processes also contributes new (raw)information that can provide the diversity needed to starta variation phase of a new knowledge cycle. The appli-cation of tbe routines in diverse contexts generates newinformationas to the pertbrmance implications of the rou-tines employed. The hypotheses constructed through thecognitive efforts of the selection phase can now (in the-ory) be tested with empirical evidence. This evidence, ifproperly collected and processed, can further illuminatethe context-dependent cause-effect linkages between de-cisions and pertbnnance outcomes. It can thereby primethe initiation of a new knowledge cycle.

The qualification concerning collection and processingis important because replication and repetition processestend to make knowledge evolve toward a more tacit formas it becomes highly embedded in the behavior of theindividuals involved in the multiple executions of thetask. Repetition leads to automaticity in the execution ofa given task and to a corresponding reduction in individ-ual awareness and collective understanding of the action-pertbrmance linkages, as well as of the purpose of theexecution criteria followed. The abundance of raw infor-mation po tentially available due to the repeated executionof the new tasks is, in fact, typically unusable withoutsignificant cognitive efforts aimed at evaluating, classi-fying, analyzing, and Hnaliy distilling the results into newinitiatives to modify existing routines or create new ones.

The extemal environment plays two distinguishableroles in the process. It supplies diverse stimuli and sub-stance for intemal reflections on possible applications tothe improvement of existing routines. Of course, it alsofunctions as a selection mechanism in the classic evolu-tionary sense as it provides the feedback on the value andviability of the organization's current behaviors. Whilewe fully recognize tbe fundamental relevance of boththese roles, the focus of this paper is on the set of internalprocesses located, from a temporal standpoint, betweentbe two.

A few more observations can be made by enteringsome additional elements in the basic framework. First,the knowledge cycle proceeds, in March's (1991) terms,from an exploration phase to an exploitation one, poten-tially feeding back into a new exploration phase. Explo-ration activities are primarily carried out through cogni-tive efforts aimed at generating the necessary range of

new intuitions and ideas (variation) as well as selectingthe most appropriate ones through evaluation and legiti-mization processes. By contrast, exploitation activitiesrely more on behavioral mechanisms encompassing thereplication of the new approaches in diverse contexts andtheir absorption into the existing sets of routines for tbeexecution of that particular task. We suggest that exploi-tation can prime exploration, and propose that in additionto the familiar trade-off between exploration and exploi-tation processes, there can be a recursive and co-evolutionary relationship between them. This may indi-cate a way to conceptualize tbe managerial challenge ofhandling both processes simultaneously.

The second observation is that the nature of organiza-tional knowledge changes over the cycle. In the firstphases of generative variation and internal selection, theinitial idea or novel insight needs to be made increasinglyexplicit to allow a debate on its merits, and the knowledgereaches a peak of explicitness as the cycle reaches theselection stage. Through the replication and retentionphases then, knowledge becomes increasingly embeddedin human behavior and likely gains in effectiveness whiledeclining in abstraction (as it is applied to a wider varietyof local situations) and in explicitness (as even the peopleinvolved in its application have difficulty in explainingwhat they are doing and why). The changing nature ofknowledge throughout the evolutionary cycle is an is.sueof primary concern to us as well as to some other scholars(Nonaka 1994. Non aka and Takeuch i 1995, Boisot 1998).We here identify as sources of such variations the differ-ent degrees of intentionality necessary at different stagesof the cycle, and construct a theoretical argument aboutthe conditions that warrant increasing efforts to turn im-plicit into explicit knowledge.

3 . Learning Investments and DynamicCapability BuildingWe now concentrate our analysis on the complex inter-play of the leaming mechanisms with the change pro-cesses that come under the heading of dynamic capabil-ities. The following proposition summarizes our view ofthe dynamic capability-building process:

P R O P O S IT IO N . Dynamic capabilities emerge from thecoevolution of tacit experience accum ulation processeswith explicit knowledge articulation and codification ac-tivities.

The reference to coevolution signals the importance ofviewing the influence of the three mechanisms in termsof continuing interaction and mutual adjustment. Our em-phasis is, howev er, not so much on the causal links amo ng


7/14

M AUR I ZI O ZOLLO AN D S I DNEY G. W I NTER Deliberate Learning and the Evolution of Dynamic Capabilities

the three capability-building mechanisms as on the waytheir interaction affects the firm's ability to improve theexisting set of operating routines. In other words, firmsleam systematic ways to shape their routines by adoptingan opportune mix of behavioral and cognitive processes,by leaming how to articulate and codify knowledge.while at the same time they facilitate the accumulationand absorption of experiential wisdom. The ambition hereis to begin an analytical effort aimed at providing aclearer sense of what an "opportune" mix of learning pro-cesses might mean. In this section, we frame the relation-ships among the mechanisms in terms of varying levelsof investment in deliberate learning activities. We thenattempt to identify the costs connected with these activ-ities. Finally, we identify some of the most important con-tingencies that need to be considered to understand howcosts and benefits vary in different contexts.

Leaming Investments. To study the relationshipsamong the three learning mechanisms, we consider theirpositions on a continuum in the investment of resources(financial, temporal, and cognitive) directed towards thepurpose of improving the collective understanding ofaction-performance linkages.Tbe level of investment in developing dynamic capa-bilities will be the lowest when the firm counts on theexperience accumulation process, as the learning happensin an essentially semiautomatic fashion on the basis of

the adaptations that individuals enact in reaction to un-satisfactory performance. The effectiveness of this leam-ing mechanism requires 'V>/y" the stability of personnelexposed to the experienced events, good performance-monitoring systems, and sufficiently powerful incentivesto ensure that individuals will initiate the search routineswhen performance levels decay. Examples of activitiesthat rely primarily on this type of learning mechanisminclude the creation of a specific function or departmentresponsible for the process to be leamed, and the hiringof "specia lists" in the execution of the task under scrutiny(e.g., creating an M&A team, hiring a TQM expert, de-fining the function of the chief knowled ge officer). Th eseactions can be understood as efforts to provide a specificorganizational locus where experience can accumulate.

The "leaming investment" is likely to be higher whenthe organization (or the relevant unit) relies on know ledgearticulation processes to attempt to master or improve acertain activity. In addition to the actions mentionedabove as necessary to facilitate experience accumulationprocesses, the organization will have to incur costs dueto the time and energy required for people to meet anddiscuss their respective experiences and beliefs (Ocasio1997). A meeting organized to debrief the participants in

a complex project can be expensive in terms of both thdirect costs and, most importantly, the opportunity cosderiving from the sacrifice of time dedicated to workinon active projects. Opportunity cost considerations havthe somewhat paradoxical effect of tending to suppreleaming when it is most valuable and needed: The highthe activity levels in the execution of a certain task, thhigher the opportunity costs for the leaming investmendedicated to that specific task, and therefore the lower thlikelihood that the hyperactive team will afford the timto debrief, despite the obvious advantages from the potential identification of process improvements (Tyre anOrlikowski 1994).

The investment of time, efforts, and resources will bthe highest in the case of knowledge codification processes. Here, the team involved in the execution of thtask not only has to meet and discuss, but also has actually develop a document or a too! aimed at the ditillation of the insights achieved during the discussion(sIf a tool (a manual or a piece of software) already existhe team has to decide whether and how to update it, anthen do it.

It is clear that, given a certain task, organizations migdiffer quite substantially in the degree to which they dcide to invest in these kinds of deliberate, cognitiveintense learning activities. Consider by way of exampthe approaches taken by Hewlett Packard and Coming developing their intemal competence in the managemeof strategic alliances."^ Both companies are considered be among the most experienced and sophisticated playein the strategic alliance arena. The ways in which themanaged their capability development processes, thougwere at the opposite extremes of our leaming investmecontinuum. Whereas Coming made a point of avoidinthe codification of its alliance practices, preferring to reon experience-based learning and apprenticeship systemHP decided to invest in a large variety of collective learing mechanisms to both identify and diffuse the best pratices in this particular task. They started with the creatioof a database of all their past alliances, for each of whicthey requested a written postmortem analysis. They theproceeded to create what became a 400-page binder withe distilled wisdom on the criteria to follow and the pfalls to avoid in each phase of the alliance process. Futher, the expert team in charge of the alliance practiinitiated a series of internal workshops/seminars open any HP manager to ensure the diffusion of their allianmanagement practices. In the language of our frameworHP combined support of behavioral leaming mechanism(the construction of a stable team of alliance experts) wideliberate investments in knowledge articulation (regul


8/14

MAURIZIO ZOLLO AND SIDNEY C . WINTER Deliberate Learning and the Evolution of Dynamic Capabilities

debriefing sessions) and codification (written postmortemanalysis and updating of the alliance manual).In terms of learning investments, HP's commitment ofmanagerial resources is much higher than Coming's. Butwas the investment worthwhile? What kind of returns can

organizations reap from higher levels of commitment todeliberate leaming processes? And, most importantly.what factors influence those returns? What weight shouldbe given to the fact that the Coming approach, relyingheavily on tacit accumulation, is more vulnerable to per-sonnel turnover? Ex post, both approaches seem success-ful, suggestingsomedegreeof equifinality, but which oneis preferable, ex ante, for a firm that is trying to developa similar competence?

To start answering these questions, we need to examinesome of the contingencies under which higher degrees ofcognitive effort are likely to be justifiable in terms of greaterleaming effectiveness. Th ree broad categories of factors areparticularly relevant to our purposes:

Environm ental Conditions (such as the .speed of tech-nological developm ent or the time-to-market lags re-quired by customers). Consider the high-speed end: Areinvestments in knowledge articulation and codificationmore justifiable in the context of high-tech industries? Atrade-off is relevant here. On the one hand, the cognitivesimplification afforded by knowledge codification hasbeen argued to be advantageous and preferable to behav-ioral adaptations in these types of industries (Brown andEisenhardt 1997). On the other, speed requirements inoperating routines raise the opportunity costs of deliber-ate leaming investments, particularly codification. In lessturbulent environments, leaming processes based oncraftsmanship, as in Brown and Duguid's (1991) notionof communities of practice, appear to be both more ef-fective and cheaper than their highly inertia! alternatives.Organizational Features. Organizations differ in theirdynamic capabilities partly because they inhabit environ-ments with differing rates of change, but also partly be-cause they place different bets, imphcitly or explicitly, onthe strategic importance of change in the future. Organi-zations that are culturally disposed toward betting onchange, or whose managements have successfully in-stilled an acceptance of continual change practices, arelikely to obtain higher returns from learning at any givenlevel of learning investment because they are more effec-tive in shifting behavior to exploit the novel understand-

ings. From a design standpoint, the presence of structuralbarriers among the different activities, typical, for ex-amp le, of multidivisional firms, will raise the retums fromdeliberate learning investments, since experiential pro-cesses will tend to remain localized to the source of novel

insights unless mobilized through explicit leaming pro-cesses. Finally, scale economies in the use of infonnationare certainly relevant, as the standard view of codificationemphasizes. Hence, larger, more divisionized and diver-sified organizations, as well as firms with strongerchange-prone cultures, are likely to have greater oppor-tunities to benefit from deliberate leaming investments.

Task Features. Returns to investments in deliberateleaming mechanisms are likely to be infiuenced also bya numher of dimensions characterizing the task that theorganization is trying to master. A task of higher eco-nomic im portance, for exam ple, will clearly justify a rela-tively higher investment in cognitive activities aimed atdeveloping competence and avoiding future failures. An-other relevant dimension is the scope of the task, whichdetermines whether a group, a department, or an entireorganization is responsible for its execution. The largerthe scope, the more compelling the recourse to more de-liberate learning mechanisms, since individuals are ableto experience, and therefore und erstand, only a small frac-tion of the task.

The implications of some other task features are lessobvious, however. In the following section, we examinein detail three determinants of the retums to leaming in-vestments. We see substantial potential in the study ofhow knowledge accumulation, articulation, and codifi-cation processes interact with task features. First, theanalysis is at a level amenable to strategic action on thepart of the firm (while there is relatively little the firmcan do to operate on its own cultural features or changeits environmental context). Second, this emphasis tendsto correct a bias that has arguably distorted the organi-zational leaming literature: The organization is typicallyconsidered to be engaged in highly frequent, relativelyhomogeneous types of tasks, reasonably well defined intheir decisional alternatives and sometimes even in theiraction/peribmiance linkages. This is the case with thevast majority of the leaming studies conducted so far inorganizational settings, where the task typically observedconsists of manufacturing or service operations.

4 . The Moderating Role of TaskFeaturesThe central tenet of the present section is that the relativeeffectiveness of the leaming mechanisms depends on thecharacteristics of the tasks that the organization is at-tempting to learn and of the operating routines that it isinterested in adjusting or radically redesigning.

We will focus our attention on three specific dimen-sions of the task or operating routines at hand. The first


9/14

M AUR I ZI O ZOLLO A ND S IDNEY G. W I NTER Deliberate Learning and the Evolution of Dynam ic Capabilities

has to do with its frequency, or how often it gets triggeredand executed within a specific period of t ime. The secondis its degree of heterogeneity, or how novel the task ap-pears each time to the unit that has to execute it. It is amatter of the dispersion in the defining traits of the taskacross multiple occurrences. The third dimension relatesto the degree of causal ambiguity in the action-performance links, or how easy it is to derive clear indi-cations as to what should or should not be done in theexecution of the task. We will elaborate on each of thesein tum.

Frequency. It is clear that organizational tasks vary im-mensely on this dimension. In principle all of them, fromextremely high frequencies such as check processing ina bank to rare events such as a reorganization or a CEOsearch process, are subject to learning mechanisms. Thequestion of what mechanism(s) work better at differentfrequency levels has been substantially neglected, thoughMarch et al. (1991) have insightfully explored the situa-tion at the very bottom of the spectrum. Based on themodel presented above, we argue that at increasing fre-quency levels, the capability-building mechanism basedon tacit accumulation of experiences in the minds of "ex-pert" personnel becomes increasingly effective as a leam-ing mechanism relative to the more explicit investmentsin knowledge articulation and knowledge codificationprocesses. At lower frequency levels, while all the mech-anisms will suffer significant losses in their capability-building power, we argue that the relative effectivenessranking inverts, and knowiedge codification becomes in-creasingly effective relative to knowledge articulation,which in turn becomes more effective than tacit experi-ence accumulation. The reasons are as follows.

Individual Memory. The experience accumulationmechanism relies on the memory of individuals exposedto previous occurrences. Other things being equal, thissuggests the more frequent the event is, the higher thelikelihood that individuals will have retained their im-pressions as to what worked and what didn't work in theprevious experiences. Indeed, the success of codificationefforts may be limited because the results of tacit leam ingare too entrenched and people might believe the consul-tation and application of task-specific tools to be redun-dant.

Coordination Costs. The knowledge articulation andcodification mechanisms become increasingly complexand costly to coordinate as the frequency of the eventincreases. Individuals need to meet to brainstorm, andtypically need face-to-face contact to coordinate the com -pletion or upgrading of a manual or a decision-supportsystem.

Opportunity Costs. Conducting debriefing sessions anupdating tools after the completion of the task cannot bdone too often w ithout diverting attention away from dayto-day operations. A balance between explicit learninactivities and execution activities, between thinking andoing, is essential (March 1991, Mukherjee et al. 1999Ma rch et al. (199 1) argue that with highly infrequeevents, organizations can learn from quasihistories (i.e"nearly happened" events) or from scenario analysiBoth mechanisms entail a substantial amount of invesment in cognitive efforts and, most likely, rely on thcreation of written output or on the use of electronic support systems to identify and make all the assumptionexplicit.

These arguments can be expressed in a slightly moformal way by advancing the following hypothesis.H Y PO T H E SIS 1. The lower the frequency of experiencethe higher the likelihood that explicit articulation an

codification mechanism s w ill exhibit stronger effectivness in developing dynam ic capabilities, as co mparewith tacit accum ulation of past experiences.

The com parison of the results obtained by a few recestudies on the effectiveness of learning processes in reltively low-frequency tasks seems to lend some support this contingency argument. As task frequency increasemoving from reengineering processes (Walston 1998) acquisitions (Zollo 1998) to alliances (Kale 1999), anfinally to quality improvement projects (Mukherjee et a1999), the relative effectiveness of experience accumlation with respect to knowledge articulation and codifcation processes seems to increase. Whereas in reengneering processes prior experiences do not affect thperformance of the current project, while efforts to creawritten guidelines do (Walston 1998), in the case of reltively more frequent quality improvement efforts, conitive (modeling and simulation processes) and experential (i.e., experiments) learning mechanisms shocom parable (and statistically significant) effectivene(Mukherjee et al. 1999). Acquisitions and alliances tyically occur at intermediate frequency levels com pared reengineering and quality improvement processes. Ka(1999) and Zollo (1998) show consi.stent results pointinto a weak but positive role for the experience accumultion mechanism and a stronger perfomiance effect for thknowledge codification proxy (number of task-speciftools developed by the focal firm).^

Heterogeneity. The variance in the characteristics the task as it presents itself in different occurrences prsents a different, albeit related, type of challenge wirespect to the frequency problem. The issue here is thindividuals have to m ake inferences as to the applicabili


10/14

MAURIZIO ZOLLO AND SIDNEY G. WINTER Deliberate Learning an d the Evolution of Dynamic Capabilities

of lessons leamed in the context of past experiences tothe task presently at hand. As task heterogeneity in-creases, inferences become more difficult to make and,when made, they are more likely to generate inappropri-ate generalizations and poorer performance (Cormier andHagm an 1987, Gick and Holyoak 1987, Holland et al.1986, Holyoak and Thagard 1995). How does the degreeof task heterogeneity affect the relative effectiveness ofthe capability-building mechanisms? We submit that themore explicit mechanisms will be relatively more effec-tive in developing d ynam ic capabilities, compared to tacitexperience accumulation, at higher degrees of taskheterogeneity. The rationale is that the hazards of inap-propriate generalization can only be attenuated via an ex-plicit cognitive effort aimed at uncovering the interde-pendence between the dimension(s) of heterogeneity andthe action-perform ance relationships, For ex am ple, a firmthat has made several acquisitions in a wide variety ofsectors will probably find it more difficult to extrapolaterules of conduct in managing acquisition processes, com-pared to another one that has consistently acquired in itsown domain. The former might find it comparativelymore useful to invest in debriefing sessions and in de-tailed postmortem analyses as opposed to simply relyingon its group of M&A experts. The need to understandwhat works and what doesn't in the different contextsexperienced requires an explicit investment in retrospec-tive sense-making, which fosters the development of spe-cific capabilities to address the different contexts in whichacquisitions might be completed in the future.

Empirical evidence unearthed in some recent studies ofacquisition performa nce seems to support this line of rea-soning. Haleblian and Finkelstein (1999) find a U-shapedleaming curve connecting prior acquisition experiencewith the performance of the focal acquisition, whereasZollo and Reuer (2000) replicate and extend this findingto the impact of prior alliance experience on the perfor-mance of the focal acquisition. The initially counterin-tuitive result represented by the declining portion of theU curve is plausibly interpreted as a consequence of thehigh heterogeneity in this kind of organizational task,which increases the likelihood of erroneous generaliza-tions. Consistent with this interpretation, knowledge cod-ification processes have been shown to be strongly relatedto performance in these conditions, and therefore rela-tively more effective than simple experience accumula-tion processes (Zollo 1998)These arguments suggest the following hypothesis forfuture empirical work.H Y P O T H E S I S 2. The higher the heterogeneity of task

experiences, the higher the likelihood that explicit artic-

ulation and codification mech anisms will exhibit strongereffectiveness in developing dynamic c apabilities, as com-pared with tacit accumulation.

Causal Ambiguity. The third task dimension that wetake into consideration is the level of causal ambiguity,or (conversely) the degree of clarity in the causa! rela-tionships between the decisions or actions taken and theperformance outcomes obtained (Lippman and Rumelt1982). Irrespective of the degree of expertise developedin handling a certain task, there are a number of factorsthat obscure these cause-effect linkages. The number andthe degree of interdependence of subtasks obviously areimportant considerations affecting the uncertainty as tothe pertbrmance implications of specific actions. These,in fact, are the two key parameters of complexity ad-dressed in the "NK" formal modeling approach, and it iscomplexity that poses the obstacle to leaming by local"experiential" search (Gavetti and Levinthai 2000). An-other important factor is the degree of simultaneity am ongthe subtasks. If the subtasks can be managed in a sequen-tial fashion, it will be easier to pinpoint the consequencesof each part for the performance of the entire process.

Again, the costs related to the "leaming investments"describ ed abov e will be justified and justifiable in tbepresence of high causal ambiguity, as the higher degreesof cognitive effort implicit in the articulation and codifi-cation of the lessons leamed in previous experiencesshould help penetrate the veil of ambiguity and facilitatethe adjustment of the routines.

We therefore submit this final hypothesis for testing infuture empirical work:HYPOTHESts 3. The higher the degree of causal ambi-guity between the actions and the performance outcomes

of the task, the higher the likelihood that explicit articu-lation and codification mechanisms will exhibit strongereffectiveness in developing dynamic ca pabilities, as com-pared w ith tacit accumulation of past experiences.

5. ConclusionsThis paper proposes a coherent structure for the study ofthe formation and evolution of dynamic capabilitieswithin organizations. It does so by drawing on argumentsderived from both the behavioral and cognitive traditionsin organizational learning studies. Starting from a char-acterization of dynamic capabilities as systematic pattemsof organizational activity aimed at the generation and ad-aptation of operating rou tines, we have proposed that theydevelop through the coevolution of three mechanisms:tacit accumulation of past experience, knowledge artic-ulation, and knowledge codification processes.


11/14

MAURIZIO ZOLLO AND SIDNEY G. WINTER Deliberate Uarning and the Evolution of Dynamic Capabilities

One implication of the analysis is the perhaps provoc-ative suggestion that knowledge codification (and to alesser extent knowledge articulation) activities becomesuperior mechanisms with respect to the accumulation ofexpertise as the frequency and the homogeneity of thetasks are reduced. This leaming-oriented appraisal mnscounter to the logic of codification that now dominatesboth theory and practice. Ordinarily, a bank would co-piously codify its branch operations (how to open an ac-count, execute a wire transfer, etc.) and a manufacturerwould d o the same with its standard operating p rocedure s,but neither would typically be prepared to do so when itcomes to managing a reengineering process or the acqui-sition of a company. This is the natural result of a habitof thinkin g that the costs of codification activities are jus -tified only by their outputs, and not by the learning bene-fits of the codification process itself. In our perspective,the creation of a manual or a decision support tool aimedat the facilitation of a relatively infrequent and hetero-geneous task may be more valid (or at least as valid) asa capability-building exercise as for the benefit derivedfrom the actual use of the tool. It can affect the level ofperformance on subsequent tasks, even if limited in num-ber, and it can affect performance even if individualminds, and not the finished tool, are the key repository ofthe improved understanding. Codification efforts forcethe drawing of explicit conclusions about the action im-plications of experien ce, .something that articulation alone(much less experience alone) does not do.

While the standard logic of codification correctly di-rects attention to the costs of codification and the scaleof application, the leaming perspective suggests other im-portant desiderata for codification to be "done right." A p-plying the logic of our argument here, and drawing on aconsiderable amount of related empirical work and dis-cussion,'^ we propose four guiding principles. First, cod-ification should aim at developing and transferring "knowwhy" as well as "know how." We have emphasized thatcodification efforts provide an occasion for valuable ef-foris to expose action-performance links. Aiming at pro-cess prescriptions alone forfeits this advantage and in-creases risks of inappropriate application. Second,codification efforts should be emphasized at an appropri-ate time in the course of learning. Attempted prematu rely,codification efforts risk hasty generalization from limitedexperience, with attendant risks of inflexibility and neg-ative transfer of learning. When codification is long de-ferred, attempts at careful examination of the causal re-lations may be fmstrated by the entrenched resu lts of tacitaccumulation, which may have attained high acceptancefor both superstitious (Levitt and March 1988) and ra-

tional (March and Levinthal 1993) reasons. What constitutes the "right time" depends on the overall heterogeneity of the tasks and the representativeness of the earlysample; high heterogeneity and low representativenesurge deferral.

The third principle calls for the codified guidance to btested hy adherence. The implementation issues here arparticularly delicate. It is clear, to begin with, that codification cannot be an instrument of continuing learning ithe guidance developed after Trial 7 is simply ignored aTrial 8 and later. Subsequent application provides the experimental test for the causal understanding that the guidance is supposed to embody; the shortcomings of thaunderstanding cannot be identified if the guidance is toeasily dismissed. Because, however, it is implicit herthat the guidance could be fiawed, the desirability of testing by adherence is obviously qualified by the need tavoid inappropriate application. The need for judgmenis inescapable, but the fourth and final point here is thathere is also a need for some supporting structure. Significant departures from the guidance should not be entirely at the discretion of the task team, but should bsubject to review and approval by a body that can assesthe case in the light of the longer-term interest in capability building. At a minimum, such a process can servthe function of making sure that the subsequent discussion in the selection stage of the knowledge cycle is correctly informed about the path actually taken and can consider, for example, whether the departure should badopted as part of the standard guidance. Such a reviewpolicy brings its own hazards, of course, particularly thrisk that the review process may become a bottleneckThe processing of engineering change orders presentstrong parallels with the organizational issues involvehere.

In our view, a more nuanced assessment of knowledgcodification and, more generally, of deliberate leaminprocesse s, is just on e exam ple of the potential benefitboth for theory building and management, of an inquirinto how competence is generated and evolves within aorganization. The framework introduced in this paperparticulariy the knowledge evolution cycle and the relationships among learning, dynamic capabilities, and operating routinesconstitutes, we believe, a significanclarification of the structure of the phenomena. This inquiry is, however, still in its infancy. We know little, foexample, of how the characteristics of the organizationstructure and culture interact with the features of the tasto be mastered in determining the relative effectivenesof the various learning behaviors. Why is it that certai


12/14

MAURIZIO ZOLLO AND SIDNEY G. WINTER Deliberate Learning and the Evolution of Dynamic Capabilities

firms, with comparable levels of expertise, codify a set ofactivities more than others do'.' And under what condi-tions does that enable, as opposed to inhibit, perfor-mance? To what extent is intentionality necessary to pro-duce adaptive adjustments in existing routines'? Thecomplexity of these questions, compounded by the factthat we have only recently started to converge on a par-simonious vocabulary for these concepts, is only com-parable to the magnitude of the expected retums from theadvancement of our knowledge on these issues.Beyond theory building, we hope that the present paperprovides useful guidance for future empirical inquiry into

the role that articulation and codification processes playin creating dynamic capabilities. That is its principal pur-pose, and although the existing empirical base is thin, weconsider that there is already good reason to believe thatsignificant progress in that direction is quite possible.More speculatively, we hope we have contributed in asmall way to a major research thmst that seems to beemerging, the effort to expand our understanding ofhowcognitive activity of a deliberate kind shapes organiza-tional learning, knowledge, and action. Although therehave long been important voices urging a different course(Weick 1979. 1995), too much of our theoretical under-standing of organizations and competitive processes hasbeen framed either by the unrealistic models of hyper-rationality favored in economics or by the more realisticbut still distorted versions of bounded rationality favoredin the behavioral tradition. People acting in organiza-tional contexts often think impertectly but con.stmctivelyabout what they are doing, and theorists of organizationsurgently need to figure out how to think in a similar fash-ion about how to cope with that fact.AcknowledgmentsThe authors would like lo acknowledge the support received from IN-SHAD's R&D department and the Reginald H, Jones Center ofThe Whar-ttm School. They are also grateful lor the many suggestions, L-omnients.and critiques generously offered by Ron Adner, Max Boisot, Tim Folia.Anna Grandori, Bruce Kogut. Arie Lewin, Bart Nooteboom. GabrielSzulanski. Christoph Zoit. and three anonymous reviewers. Of course, iheauthors are responsible for any remaining errors and omissions, tn ils work-ing paper version, this article was titled ""From Organizational Routines toDynamic Capabilities."Endnotes'Of course, this is not to say ihat firms do not articulate and codifyknowledge obtained from ihe external environment. As we will see in2, environmental scanning activities are nn important antecedent togenerative variation processes. However, we argue that such activitiesare usually better utiderstood as stimuli to the initiation of proposals tomodify existing routines, rather than as mechanisms directly shapingthe development of dynamic capabiiities. To illustrate, a sound under-standing of what competitors do and customers desire represents a cru-cial element of any tinn's competitive strategy, but in and of itself does

not make it any more capable of creating and modifying its own set ofoperating routines. Also, operating routines typically involve tacitknowledge; hence they are highly unlikely to be developed or shapedsimply by the observation of competitors, suppliers, customers, or otherexternal constituencies. Such knowledge has to be developed "inhouse" through a set of activities and cognitive processes focused onthe organization's own routines.This model has significantly benefited from many discussions hadduHtig the "Knowledge and Organization" workshop held at WarwickUniversity. We would like to acknowledge the contributions of ArieLewin. Anna Grandori. and Bart Nooteboom. among many other par-ticipants. The model is also closely related to Max Boisot's notion ofthe social learning cycle (Boisot 1998). which influenced our thinkingand spurred some of the insights thai follow.'The justification for the explicit consideration of replication processesas distinct from the retention ones becomes considerably weaker as thenature and the size of the firm considered changes. Small entrepre-neurial start-ups, for example, might necessitate replication processesto a far lesser (and perhaps negligible) degree, compared to a multi-national, muitidivisional organization.'For those intere.sted in this specific empirical context, please refer toan article entitled "Two grandmasters at the extremes." which appearedin the Alliance Analyst on Nov. 25, 1995.^Kale (1999) is the only empirical work, to our knowledge, Ihat hasmeasured all the three mechanisms theorized about in this paper.Knowledge articulation processes are positively and significantly cor-related with alliance performance, with an overall explanatory powercomparable to the knowledge codification proxy, and larger than ex-perience accumulation.''See especially the discussion and references in Adler and Borys {1996)and empirical work cited in 4,ReferencesAdler, P. S.. B. Borys. 1996. Two types of bureaucracy: Enabling and

coercive. Admin. Sci. Quart. 41 61-89.Argote, L. 1999. Organizational Learning: Creating. Retaining and

Transferring Knowledge. Kluwer Academic Publishers. Boston.MA.

Argyris. C. D. Schon. 1978. Organizational Learning. Addison-Wesley, Reading, MA.

Boisot. M. H. 1998. Knowledge Assets: Securing Competitive Advan-tage in the Information Economy. Oxford University Press. Ox-ford. U.K.

Brown. J. S., P. Duguid. 1991. Organizational learning and commu-nities of practice: Toward a unified view of working, leaming andinnovation. Organ. Sci. 2 40-57.

Brown. S. L.. K. Eisenhardt. 1997. The art of continuous change: Link-ing complexity theory and time-paced evolution in relentlesslyshifting organizations. Admin. Sci. Quart. 42 1-34.

Campbell. D. 1969. Variation and selective retention in socio-culturalevolution. General Systems 16 69-85.

Cangelosi. V. E.. W. R. Dill. 1965. Organizational leaming: Obser-vations toward a theory. Admin. Sci. Quart. 10 175-203.

Cohen. M.. P. Bacdayan. 1994. Organizational routines are stored asprocedural memory: Evidence from a laboratory study. Organ.Sci. 5 554-568.

Cohen. W. M.. D. A. Levinthal. 1990. Absorptive capacity: A new


13/14

M AURI ZI O ZO LLO AND S I DNEY G. W I NTER Detiherale Uaminfi and the Evolution of Dynamic Capabililies

perspective on learninj; and innovation. Admin. Sci. Quart. 35128-152 .

Collis. D. 1994. Research note: How valuable are organizational ca-pabilities? S?ra/fg(c" Afw/io^fmcn;,/. 15 14 3-15 2.

Cormier, S., J. Hagman. 1987. Transfer of l^armug: CotUemporaryResearch and Applications. Academic Press . San Diego. CA.Cowan, R., D. Foray. 1997. The economics of codification and thediffusion of knowledge, indust. Corporate Change 6 595-622 ., P. David, D. Foray. 2000. The explicit economics of knowledgecodification and tacitness. Imlusl. Corporate Change 9 2 1 1 -2 5 3 .

Dosi, G., R. Nelson. S. Winter, eds. 2000. The Nature ami Dynamicsof Orgariizaiiiiiuil Cupahilities. Oxford Universily Press, NewYork .

Duncan R., A. Weiss. 1979. Organizational learning: Implications fororganizational design. Res. Organ. Behavior 1 7 5 -1 2 3 .

Eisenhardt, K., J. Martin. 2(H)0. Dynamic capabilities: What are they?Strategic Management J. 21 I I 0 5 - 1 1 2 1 .

Gav etti, G.. D. Levinthal. 2{X)0. Look ing forward and look backw ard:Cognitive and cxperienlial search. Admin. Sci. Quari. 4 5 1 1 3 -137-

Gick, M., K. Holyoak. 1987. The cognitive basis of knowledge transfer.S. M. Cormier and J . Hagman, eds. Transfer of Learning: Con-temporary Research iind Applications. Academic Press , NewYork.

Glynn, M. A., T. K. Lani, J. M. Miliken. 1994. Mapping learning pro-cess in organizations: A multi-level framework linking leamingand organizing. Advances in Managerial Cognition and Organi-zational Information Proces.sing. vol. 5. JAI Press , Greenwich,C T . 4 3 -8 3 .

Hale blian, J.. S. Finkelstein. 1999. The influence of organization ac -quisition experience on acquisition performance: A behaviorallearning theory perspective. Admin. Sci. Quart. 44 29-56 .

Henderson, R. M.. K. B. Clark. 1990. Architectural innovation: Thereconfiguration of existing product technologies and the failure ofestablished firms. Admin. Sci. Quart. 3 5 9 -3 0 .

Holland. J.. K. Holyoak. R. Nisbelt. P. Thagard. 1986. Induction: Pro-cesses of Inference, [^ earning, and D iscovery. MIT Press , Cam-bridge, MA.

Holyoak. K. J.. P. Thagard. 1995. Mentat Leaps. MIT Press . Cam-bridge, MA.

Kale, P. 1999. Building an alliance capability: A knowledge-based ap-proach. Unpublished dissertation. University of Pennsylvania,Philadelphia. PA.

Kogut, B., U. Zander. 1992, Knowledge of the firm, combinative ca-pabilities, and the replication of technology. Organ. Sci. 3 3 8 3 -397.

Leonard-Barton, D. 1992- Core capabilities and core rigidities: A par-adox in managing new product development. Strategic Manage-ment J. 13 111-125 .

Levinthal, D., J. G. March. 1988. A model of adaptive organizationalsearch. Decisions and Organizations. Basil Blackwell, New Y ork.187-218 (originally published 1981).

Levitt, B-, J- G. Ma rch. 1988. Organ izational learning. Ann. Rev. Soc.14 319-340 .

Accepted hy Arie Lewin.

Lippman. S. A., R. P. Rummelt. 1982. Uncertain imitability: An analysis of interfirm differences in efficiency under compelilion . BeJ. Econom. 13 418-438 .

Ma rch. J. G. 1991. Exploratio n and exploitation in organizatio nal lea ming. Organ. Sci. 2 7 1 -8 7 ., D. Levinthal. 1993. The myopia of learning- Strategic Management J. 14 95-112 .

-- . H. A, Simon. 1958. Organizations. Wiley, New York., L. S. Sproull. M. Tamuz. 1991. Leaming from samples of onor fewer. Organ. Sci. 2 1-13.

Mukherjee, A., M. Lapre. L. Van Wassenhove. 1998. Knowledgdriven quality improvement. Management Sci. 4 4 5 3 5 -5 4 9 .

Narduzzo, A., E. Rocco, M. Warglien. 2000. Talking about routines inihe field: The emergence of organizational capabilities in a newcellular phone network company. G. Dosi. R. Nelson, and SWinter, eds. The Nature and Dynamics of Organizational Capabilities. Oxford University Press , Oxford. U.K. 27-50.

Nelson. R., S. Winter. 1982. An Evolutionary Theory of EconomiChange. Harvard University Press . Cambridge, MA.

Nonaka, L 1994. A dynamic theory of knowledge creation. Organ. Sc5 14-37 ., H. Takeuchi. 1995- The Knowledge-Creating Company. OxforUniversity Press. New York.

Ocasio, W- 1997. Towards an attention-based view of the firm. Strategic Management J. 18 187-206.

Selznick. P. 1957. Leadership in Admin istration: A Sociological Interpretation. Row. Peterson, and Co.. Evanston, IL.

Teece, D. J., G. Pisano. A. Shuen. 1997. Dynamic capabilities anstrategic management. Strategic Management J. 18 509-533 .

Tyre, M. J . , W. J . Orlikowski. 1994. Windows of opportunity: Temporal pattems of technological adaptation in organizations. Organ. Sci. 5 9 8 -1 1 8 .

Walston, S. 1998. Does re-engineering really work? Extracted from aunpublisbed dissertation. University of Pennsylvania. Presenteat the EGOS Conf Maastricht. The Netherlands, July 1999.

Weick, K. 1979. The Social P.sychology of Organizing. 2nd edAddison-Wesley, Reading, MA.. 1995. Sensemaking in Organization. . Sage Publications, Thousand Oaks, CA.

Winter. S. 1987. Knowledge and competence as strategic assets. D. Teece. ed. The Com petitive Ch allenge: Strategies for IndustriaInnovation and Renewal. Ballinger. Cambridge. MA, 159-184., G. Szulanski. 2000. Replication as strategy. Organ. Sci. 1 2 7 3 0743.Zander, U., B. Kogut. 1995. Knowledge and the speed of transfer animitation of organizational capabilities: An empirical test. OrgaSci. 6 76-92 .

Zollo, M., 1998. Knowledge codification, process routinization and thdevelopment of organizational capabilities: Post-acquisition integration in the U.S. banking industry. Unpublished doctoral dissertation. University of Pennsylvania, Philadelphia, PA., J. Reuer. 2000. Experience spillovers across corporated deveopment activities . INSEAD Working paper. FontainebleaFrance.


14/14

Documents

Deliberate Learning and the Evolution of Dynamic Capabilities