OrganizationalImprovisation andLearning:A Field Study

Anne S. MinerUniversity ofWisconsin-MadisonPaula BassoffUniversity ofWisconsin-MadisonChristine MoormanDuke University

©2001 by Cornell University.0001-8392/4602-0304/$3.00.

Direct all correspondence to the firstautlior. National Science Foundation Grant#SBR-9410419, the Marketing ScienceInstitute, and the University of Wiscon-sin-Madison Graduate Fund and Schoolof Business have generously supportedthis research. The authors appreciate thecomments of Ted Baker, DeborahDougherty, Kathleen EJsenhardt, MarthaFeldman, Randi Huntsman, KathyMcCord, Thekla Rura-Polley, Bob Sutton,Fuqua's Management Seminar Group,and the illuminating comments of Associ-ate Editor Reed Nelson on previous ver-sions of this manuscript and the researchsupport of Mary Boeding. StephanieDixon, Bob Drane, Sarah Gates, GaborKemeny, Aric Rindfieisch, and David

An inductive study of improvisation in new productdevelopment activities in two firms uncovered a varietyof improvisational forms and the factors that shapedthem. Embedded in the observations were two importantlinkages between organizational improvisation and learn-ing. First site observations led us to refine prior defini-tions of improvisation and view it as a distinct type ofreal-time, short-term learning. Second, observationrevealed links between improvisation and long-termorganizational learning. Improvisation interfered withsome learning processes; it also sometimes played a rolein long-term trial-and-error learning, and the firms dis-played improvisational competencies. Our findingsextend prior research on organizational improvisationand learning and provide a lens for research on entrepre-neurship, technological innovation, and the fusion ofunplanned change and order.*

Organizational theory reveals a growing interest in extempo-raneous organizational action and its potential value to organi-zations. Outside the organizational context, scholars haveobserved improvisation in fields as diverse as theater andmusic (e.g., Spolin, 1963; Bastien and Hostager, 1992; Weick,1993b; Berliner, 1994), education (e.g., Borko and Livingston,1989; irby, 1992), and psychiatry (e.g., Embrey et al., 1996).Researchers have analyzed improvisation by organizations inespecially fast-moving competitive settings, such as newproduct development (Eisenhardt and Tabriz!, 1995; Moormanand Miner, 1998a, 1998b), a changed political context (Alin-sky, 1969), and during emergencies such as a strike (Preston,1991), a failed navigational system (Hutchins, 1991), and afirestorm (Weick, 1993a). Others have focused on the aes-thetic virtues of improvised action (Weick, 1993c; Hatch,1997b). Work to date thus provides ample evidence that theconstruct of improvisation can generate lively discussion andthat instances of improvisation are found in organizations.

One recurring theme of both research and lay observations isthat stored knowledge and skills shape improvisation inimportant ways. Weick (1993a) noted that experience playeda role in successful and unsuccessful improvisation by fire-fighters. Moorman and Miner (1998a) found that organization-al memory moderates the impact of improvisation on newproduct outcomes (see also Moorman and Miner, 1998b, forrelated theory). Brown and Eisenhardt (1995) theorized thatlearned routines shape improvisation in new product develop-ment (see also Eisenhardt and Tabrizi, 1995). Hatch (1998)observed that skilled improvisers often recombine existingroutines (parts of memory) to create novel action, much as amusician reassembles previously performed bundles of notesinto a novel melody.

At the core of prior work is the argument that the result ofprior learning, organizational memory, shapes the skillful andfruitful improvisation of novel performances. Research is lessclear, however, about whether and how improvisation affectslearning, focusing instead on the outcomes of improvisationitself, such as saved firefighters or firms. We wonderedwhether improvisation would result in a different set ofbehaviors or insights relative to what firms would have expe-

304/Administrative Science Ouarterty, 46 (2001): 304-337

Improvisation and Learning

rienced under normal planning and execution. Becauseimprovised activities often occur outside organized routinesor forma! plans, we also questioned whether they could beaccepted and incorporated into future organizationai activi-ties. At an even higher level of organizational learning, it isnot clear whether an organization can learn to plan such anunplanned event as an improvisation.

In examining these issues, we considered two sets of ideas.First, we focused on the extemporaneous quality of improvi-sation, drawing on a large body of prior work across disci-plines (e.g., Pressing, 1988; Borko and Livingston, 1989; Pre-ston, 1991; Bastien and Hostager, 1992; Weick, 1993a,1993b). Speci-fically, we build on the view of improvisation asthe degree to which composition and execution converge intime (Moorman and Miner, 1998b: 698). In addition, weaccepted the notion that preexisting routines do not consti-tute improvisation; there must be some degree of novelty inthe design. We took "composition" or "design" to mean thatimprovisation refers to deliberate, as opposed to accidental,creation of novel activity.

Second, although we had very little sense of the specificinterconnections we would find between improvisation andlearning, we developed a framework for thinking about orga-nizational learning and how we would recognize it. This, too,was grounded in a considerable body of prior scholarly workthat defines learning as a systematic change in behavior orknowledge informed by experience (e.g., Levitt and March,1988; Huber, 1991; Cyert and March, 1992; Glynn, Lant, andMiliiken, 1994; Argote, 1999). This view of organizationalleaming embraces both behavioral learning models, whichemphasize shifts in the mix of organizational routines andaction patterns, and cognitive learning models, which empha-size shifts in ideas, causal models, and cognition. In purebehavioral learning, an external stimulus could generate anew balance in internal routines without any change in theorganization's shared mental models or causal theories(Miner, 1989; Cyert and March, 1992; Burgeiman, 1994). Thekey learning outcome resides in the new mix of routinesenacted. In cognitive learning, a change outside the organiza-tion could stimulate reflection and revision of shared cogni-tive assumptions and causal models, even with no change inthe organization's current routines (Glynn, Lant, and Miiliken,1994). The key learning outcome resides in the organization'sknowledge base.

Early work on organizational leaming tended to describe it asa single process (e.g., Cyert and March, 1992), but contem-porary research elucidates many types of learning processes,such as an organization's learning from its own experienceversus leaming from others, experimentation, trial-and-errorlearning, refinement versus exploration, forgetting, knowl-edge sharing, and knowledge generation. Common to ailforms, however, is that the learning experience generateschange in some fashion. It may, for example, refine priorknowledge or reduce variation in activity (Argote, 1999). Itmay generate new activities, knowledge, or insight (March,1991). Experience-induced changed behavior may or may notbe retained; we follow other work that emphasizes the reten-

305/ASQ, June 2001

tion of change as a potential, but not inherent, part of learn-ing (Glynn, Lant, and Milliken, 1994; Argote, 1999). Further,change may be short-term or long-term and may involveeither local or higher-level organizational patterns, competen-cies, or knowledge (Argote, 1999).

With these core concepts as a starting point, we firstdescribe patterns we found in improvisation and how theyled us to conclude that improvisation can fruitfully be seen asa special type of short-term, real-time teaming. Specifically, inimprovisational learning, experience and related change occurat the same time. We then report empirical findings thatreveal how improvisation can influence long-term organiza-tional learning and adaptation. Taken together, the findingsimply that improvisation not only draws on prior learning butmay be both a special type of short-term learning and a factorthat influences other, longer-term organizational learningactivities.

METHOD

Prior observations of improvisation in a single organization(e.g., Preston, 1991; Hutchins, 1991; Weick, 1993a) oftenbegan with a specific improvisation and traced its links toother organizational actions and theories. Rather than startingwith a prominent improvisational event and working back-wards, we wanted to observe a consistent flow of regularactivity over time in two different organizations and track theebbs and flows of potential improvisation. We chose productdevelopment projects as the context in which to conduct ourexamination. Product development represents an importantprocess through which organizations adapt to—and. In somecases, help to create—^their own environments(Schoonhoven, Eisenhardt, and Lyman, 1990; Dougherty,1992; Brown and Eisenhardt 1995). Although some previousresearch on new product development has provided sugges-tive evidence that links improvisation and organizational learn-ing, this work has not thoroughly explored their interconnec-tion (Nonaka, 1990; Eisenhardt and Tabrizi, 1995; Moormanand Miner, 1998a). Thus, product development represents afertile but only partially mined territory for examining improvi-sation and its potential links to learning.

Our observation focused on two companies in two differentindustries whose names, but not core features, we havealtered for purposes of confidentiality. The first company,FastTrack, develops and sells advanced technology productsto industrial clients as well as to public and private researchlaboratories. The second company, SeeFoods, develops andsells food products to consumers through supermarkets andinstitutional channels. Both companies are well establishedand have formalized structures but vary in size ($2.4 millionand $2.6 billion in annual sales, respectively).

In each firm, we attended approximately 50 product develop-ment meetings that spanned the product development cycle:concept and prototype development stages at SeeFoods andproduct design through launch at FastTrack. We informedemployees at each site that we were conducting a study ofproduct development. To avoid demand effects, we did notuse the word "improvisation" in our work and consistently

306/ASa June 2001

Improvisation and teaming

described our project as general exploratory research onproduct development. We did not ask participants what theythought about the construct of improvisation or solicit com-ments about the link betv\/een planning and execution, but,instead, we used several data sources and modes of inquiryto explore concrete product development activities, process-es, and outcomes (Denzin and Lincoln, 1998; Strauss andCorbin, 1998).

Our field data collection period lasted for more than one year,v^ith the most intense data collection occurring during a nine-month period. This window of observation provided pro-longed interaction, and the data sources permitted some tri-angulation in our analyses {Frankel, 1999). Findings reflectour direct observations of organizational activities, our repeat-ed reviews of approximately 1,500 pages of meeting andinterview transcripts to challenge our initial assumptions, andour discussions and shared writings about these data.

Data Sources

Observation and transcripts of team meetings. Weobserved and tape-recorded approximately 25-30 team meet-ings for one new product development project at each siteover a nine-month period. Research team members also keptmeeting notes, both as a safety precaution in the event ofequipment failure and as a device for recording impressions.Meetings were an important data source for fine-grainedanalysis because they represented the principal means bywhich members exchanged information about actions thataffected the product development process and its outcomes(Adier and Adier, 1998). To ensure that meetings captured arepresentative set of actions by the team, project team mem-bers were asked to list all the things they had done on behalfof the project during a sample period, early in the field obser-vation phase. When we compared these lists to meetingagenda, field workers' notes, and meeting transcriptions forthe sample period, coverage was deemed adequate. Audio-tapes of the meetings were transcribed and reviewed morethan once by the research team member who attended themeeting. These transcripts provided shared data that wasavailable to all research team members, as well as much ofthe raw data we used for analysis of specific improvisationalactivities we had observed. Although we did not intend touse the tapes for discourse analysis, we reviewed the tran-scripts carefully for accuracy because of their central role asa shared, detailed record of project activities. Transcripts alsoincluded observers' notes of unusual body language or othernonauditory signals of major changes in meeting content ortone.

tnterviews. We interviewed team leaders and executivesthroughout the research process. Initial interviews covered abroad range of topics: company history and structure; currentand future projects; development philosophies and process-es; and competitors, suppliers, and customers. We also con-ducted interviews with team leaders after each of severaldevelopment phases, as well as on-the-spot interviewsbefore and after many team meetings. One goal of theseongoing interviews was to gather additional information

307/ASQ, June 2001

about project-related actions and events; another was toavoid such problems as poor recall, hindsight bias, and haloeffects (Hawkins and Hastie, 1990; Fontana and Frey, 1998).We also wished to elicit more general information on compa-ny history and practices as seen by the informants, such asother ongoing product development projects and their linksto the current projects, previous projects and policies, andperceived strengths and tensions within the firms. Interviewswere semistructured, with some questions informed by ourprior research questions and others suggested by developingthemes within the interview itself (Fetterman, 1989; Crabtreeand Miller, 1999). In total, nine interview schedules weredeveloped to reflect management levels (e.g., senior-officerlevel, team leader) and different timing considerations {e.g.,beginning of the project, weekly meeting, long-term learningfotlow-up).

Archival data. We examined such archival information asorganization charts, company brochures, forma! new productdevelopment procedures, project planning documents, andproject information. This provided information on companyhistory, structure, policies, technical competencies, and prod-uct markets as wel! as data on project dates, formal planningsteps, and decisions related to the projects we observeddirectly (Denzin and Lincoln, 1998).

Procedures

To explore our research questions, we used a process ofrecursive scrutiny (Barley, 1986; Ragin, 1987; Borkin, 1999) ofall data sources. One important form of this occurred duringthe data collection phase, when research team memberswould meet to review the progress of the project and exam-ine new transcripts to consider whether and how we couldidentify improvisation. We tried, for example, to identifyimprovisational actions at the level of whole project phasesbut found we could not do that in a satisfactory way. Thesediscussions represented an open-coding process in whichobservers examined and reexamined the same traces ofactivity in search of regularities in behavior (Strauss andCorbin, 1998; Denzin and Lincoln, 1998). Thus, we focused inthis early period on identifying improvisational actions withinspecific projects.

The second overlapping form of iterative data examinationoccurred through ongoing reviews and discussions of tran-scripts and other data over a period of more than four years.Keeping in mind that events reported in interviews andarchival data cannot be taken as a transparent representationof past or future observable behavior, we tended to focus onpatterns that arose from transcript data or that appeared inone or wore data sources. Research team members repeat-edly and independently examined transcripts of meetings andinterviews and then exchanged illustrations of improvisationin an iterative process for more than a year, with later period-ic repetition of this process. We used repeated examinationof our data, critical exchange among team members, andinterchanges with readers about working papers to seekrobustness and plausibility (Ragin, 1987; Eisenhardt, 1989;Strauss and Corbin, 1998; Borkin, 1999). The long-term itera-

308/ASa June 2001

Improvisation and Learning

tive review was especially important in considering higher-level patterns in the firms, contrasting different types ofactivities, and refining definitions. In this study, we did notuse semiotic, deconstruction, or dramaturgical techniques{Feldman, 1995), although we did pay attention to somearea5 in which team members expressed surprise or frustra-tion.

Our analysis of the data revealed that both organizationsengaged in thorough strategic planning processes to identifyand prioritize broad classes of potential products and mar-kets. The companies typically assessed new product devel-opment projects in the context of longer-term priorities, andboth had written procedures for any new product develop-ment project to which resources would be allocated. In bothfirms, the actual design process did not officially occur untilthese analyses and resulting plans had been formallyapproved and funded, clearly making the normal productdevelopment process one of formal planning rather thanimprovisation.

IMPROVISATION IN ORGANIZATIONS

The first focus of our field observations was simply toobserve carefully the activities that appeared to embodyimprovisation and the patterns that might emerge in theseactivities {Denzin and Lincoln, 1998). This agenda revealedseveral different forms of the "execution" aspect of theimprovisations. It also revealed varied foci of the organiza-tion's attention during the improvisation process, whichshaped the "composition" aspect of improvisation. Throughour analyses we came to see improvisation as a special typeof learning that differed from the experimental learning weobserved in these firms.

Forms of Improvisation

As we reviewed instances of improvisation, we noted thatsomething was produced in each case, but the variety wasinteresting and unexpected. We term these direct outcomes"improvisational productions." Like an improvised song ortheatrical skit, these productions were themselves irrevoca-ble—they were "on-line" events—but they might or mightnot have other consequences or lasting impact on the organi-zation. We do not make any claims here about the effective-ness or retention of these products of improvisation. Rather,here we describe the range of new behaviors (behavioral pro-ductions), physical structures (artifactual productions), andnew interpretive frameworks (interpretive productions) impro-vised by the organizations.

Behavioral productions. Consistent with past research(Dougherty, 1992; Hatch, 1998), we saw teams improvisingprocesses (sequences of behavior) in new product develop-ment. Sometimes these processes were coupled with thecreation of artifacts (see following section), but we alsoobserved pure behavioral improvisations as well. SeeFoods,for example, improvised new-product-development micro-actions. Although development teams were required to visitstores that sold products similar to their own, a SeeFoodsdevelopment team changed its behavior during one such trip

309/ASQ, June 2001

The following notational system will beused throughout the paper: SF =SeeFoods and FT = FastTrack. The datesrepresent the primary transcript or inter-view record that describes the event inquestion. Quoted text is taken directlyfrom the transcripts.

to include stores selling products that were dissimilar in con-tent but similar in packaging. Therefore, although the broadpractice of visiting stores remained the same, the teamdesigned a new activity during completion of that task (SF01/16/95).•" Teams also improvised new-product-developmentprocesses. At SeeFoods, for example, company policy man-dated the use of external focus groups at specific points dur-ing the development process. During the process of planningone such external focus group, the team created an internalfocus group of company employees to call in on an as-needed basis, thus permitting more frequent feedback (SF02/22/95).

Another example of improvised processes occurred as Fast-Track's product development team was developing a soft-ware interface between its product and one of the compa-ny's flagship products. White that interface was beingwritten, the older product's software came under revision.This meant that the new product development team continu-ally had to change its test schedules and sequences to corre-spond to software revisions. As the team struggled to followthe process and timing of the team revising the older prod-uct, frustrated team members improvised an approach andset of routines to develop and test their interface: "It seemsthat they are finding new bugs in the old software and thatthey might issue a new release. This is our chance to slip inour changes" (FT 11/28/95). This "slipping in" of changessurreptitiously reversed roles. The product developmentteam had improvised a method to get its changes tested bythe other team now in charge of compatibility testing.

We also observed several instances of SeeFoods' improvis-ing organization-wide processes. In one case, a breakfastproduct had been put on hold by the project team becausethe leader felt the team was being spread too thinly. When acompetitor introduced a breakfast product, SeeFoods tem-porarily dropped its lunch projects to focus exclusively onseveral very underdeveloped breakfast prototypes. The entireorganization, across functions, created a fast-response strate-gy as it worked through the problem (SF 05/31/95).

Finally, we observed one situation at FastTrack when the firmappeared to be improvising entry into a whole new productmarket, responding to a new material capability. We did nothave sufficient data to substantiate unequivocally that thishighest-level strategic improvisation occurred during our peri-od of observation, although other work has documented theimprovisation of the whole mission or goals of an organiza-tion (Fotlett, 1930; Mintzberg and McHugh, 1985; Preston,1991; Weick, 1993b).

Artifactual productions. In addition to improvising behavioralproductions the teams we observed actually created newphysical structures without prior design. At a micro-level,teams sometimes improvised new product features, as docu-mented by one example from our field notes:

FastTrack's design engineers were conducting a series of unrelatedtests. During this process, it occurred to them that the product'ssafety and performance would be enhanced by adding a cover, andthey quickly improvised a mock-up cover not in the product's plans.

310/ASQ, June 2001

im^nwisation and Learning

They developed the mock-up as they went along, without goingback to the drawing board. No external stimulus, such as informa-tion describing a new type of effective cover material, prompted theaction, nor any exogenous request for this feature. Instead, theirongoing actual interaction with the specific materials and behaviorof the product itself stimulated their design and simultaneous pro-duction of the cover mock-up. The mock-up persisted, received for-mal approval, and was included in the final product. [FT 11/21/94)

In another instance, the FastTrack team received cables thatdid not meet specifications. They needed to move the prod-uct along and did not want to fix the cables, so they impro-vised a new circuit board to accommodate the faulty cables(FT 01/16/95). In a final instance at FastTrack, a softwareupgrade on a FastTrack research instrument caused a movingarm to shift too frequently when taking measurements,which damaged the arm. Because an upcoming major prod-uct redesign would eliminate the entire arm, the team did notwant to do a redesign of the arm itself to resolve the prob-lem. Instead, they improvised a temporary "fix" with a smallnylon washer (FT 6/30/95). This washer had never been partof any product design or specifications. The new feature wasconceived, designed, and implemented as team membersworked together to solve the perceived problem.

Many of the improvisational activities observed at SeeFoodsalso created new product features. The team often workedaround a table or in the firm f̂e kitchen, configuring productsinto various packages and designing prototypes as they wentalong. On one occasion when the team was in the process ofworking with product components for a new breakfast pizza,it developed several new features that had not previouslybeen considered. These included product form (flat, rolled),various toppings (icing, sprinkles), and the temperature of theproduct (whether eaten hot or cold) (SF 12/19/95).

Occasionally, we observed artifactual productions at higherlevels that improvised entire products. At FastTrack, engi-neers sometimes designed and physically generated prod-ucts themselves without the help of marketing or manufac-turing personnel. Such material improvisations were referredto as "scientist specials" and typically unfolded through inter-action with specific customers or, in some cases, as part ofscientists' ongoing activities while working on other projects.For example, one team learned from an established cus-tomer that the customer wished to analyze samples of a radi-cally different nature than FastTrack's regular products couldhandle. Team members simultaneously designed and pro-duced the new device during spare moments in their ownlabs, often using parts on hand while negotiating what theproduct would do. There was no separate product design andproduction phase.

is this really improvisation? We argue that it is, for two rea-sons. First, during our period of observation, FastTrack devel-oped new products principally using highly formalized proce-dures. These procedures required substantial writtenanalyses of both market potential and technical issues thatwere reviewed by senior committees with full budgetaryauthority to authorize the project for development or to "kill"it. FastTrack planned practically every product in advance fol-

3n/ASQ, June 2001

towing these steps and evaluated them against the frame-work of the overall strategic plan. In contrast, creation of sci-entist specials evolved through ongoing interactions with cus-tomers and, in some cases, with materials or elements athand of formally approved products. Some individual actionsmight be preplanned within the improvised projects; scien-tists, for example, might plan to call a potential supplier tosee if a part vyas available and then carry out that plan thenext week. The entire product, however, was improvisedfrom the viewpoint of the top level of the organization's setof global priorities because it v\/as designed at the same timeit was produced. Crucially, the same team that was designingthe product also physically produced it and actually handed itover to the customer. This contrasted with al! other productsthat, once finalized, were physically produced by a separatemanufacturing group.

Interpretive productions. We observed several Instances inwhich teams improvised new interpretations or frameworksduring intermediate stages of the planning process. In thesecases, the production seemed to be the reframing of anevent from one general framework to another. At FastTrack,for example, an engineer reported that correcting a programerror unrelated to speed created an unanticipated outcome:an information search that had previously taken 22 secondscould now be completed in two seconds. While discussingthe "bug fix" at one of the meetings, the team turned it intoa speedy reporting feature that could be emphasized in mar-keting efforts (FT 12/05/94). This shift in interpretationmarked the beginning of a process to work through thedetails of how to present it to customers. Al! of this wasreportedly done without a plan, using only the team's interac-tion with the program as a guide.

Another new interpretation involved a FastTrack part that per-formed extremely well in some instances but was average topoor in others. It was particularly unsettling that there was noway of knowing in advance which parts would be good per-formers. One engineer explained during a team meeting thatvariability was inevitable for this sort of item and that nothingcould be done to prevent it. The team then discussedwhether customers who received a "hot" (high-performing)item in one shipment and an "average" item in the nextwould feel short-changed, regardless of the fact that bothitems performed at or above published specifications, Sug-gestions ensued. One person proposed that they test eachpart as it was received: another suggested that the vendorperform testing in return for a sorting fee. Either way, qualitycontrol would be costly. One of the engineers then proposed,"If you see a hot one, let me know. I can phone the cus-tomer and tell them we have this hot item and do they wantit. Then they think: 'Oh yeah, FastTrack's really good guys.They look out for me"' (FT 11/21/94). At least from theteam's perspective, a problem that might have threatenedthe perceived quality of their products was transformedthrough an interpretive improvisation.

In both cases, the team noticed a deviation from expecta-tions. !t could have been ignored, although the part variabilitywould probably have to be explained to customers. Instead,

312/ASQ, June 2001

Improvisation and Learning

We thank our anonymous reviewers fortheir important contributions to this paper,especially comments that prompted us torevisit our data and clarify improvisationas a special t^pe of learning.

team members did two things: they activeSy reframed themeaning of the unexpected events in a novel way, and theyinfused the prior events with new meaning. The initialmoment of improvisation occurred in the reinterpretation,although the entire improvisationai activity included latersteps—like calling customers about the hot items—that exe-cuted the new meaning. Some of these episodes resembledmusical improvisation in which a musician plays an unintend-ed note but goes on to play additional notes that create a pat-tern in which the previous wrong note now appears meaning-ful and melodic (Berliner, 1994).

It is important to note that if the interpretations had not beenacted upon by the teams, they would not qualify as improvi-sation in our view, because there would be no "on-line"activity or production involved. Including interpretive improvi-sation among the types of improvisation identified to date isconsistent with the literature. Preston (1991) described howa plant management team improvised new activities to dealwith a strike, in part, through reinterpreting what kind of busi-ness they were In, which made different actions possible.Moreover, our observations indicated that these presumablyrare events appeared to be an important part of improvisationin new product development.

Implications for the Improvisation Construct

Although prior reports of improvisation in the arts, therapy,and teaching emphasize behavioral improvisation, both arti-factual and interpretive improvisations have also beendescribed in varied settings. Our observations revealed arather complex and nuanced combination of all three andconvinced us that improvisation represents a special type oflearning. Our close review of actual improvisational activity inmultiple settings and our repeated interchanges about theseobservations caused us to reflect on prior definitions ofimprovisation and to propose two important refinements orclarifications about the definition of improvisation and ourconceptualization of its link to other related constructs.^ Wehad started with prior research's general description of impro-visation as involving "no split between composition and per-formance . . . no split between design and production"(Weick, 1993b: 6). Moorman and Miner (1998b) specified thatone good measure of improvisation would be the narrownessof the time gap between design and execution, but such adefinition would also fit a firm outside our sample that usedto generate one-of-a-kind products for specific customers fol-lowing a three-week design and production process. Thissame firm now uses computer-aided design and computer-aided manufacturing systems that have reduced the processto about one day. The result is substantial temporal conver-gence between design and execution but is not improvisa-tion, in our view, because each stage is still substantively dis-tinct.

In our study, we saw products being designed and created asthe teams enacted them. This means that production had animpact on composition in a way quite different from whatmight be expected if only temporal convergence of designand action occurred. With only temporal convergence, design

313/ASQ, June 2001

could take p!ace just prior to an action, yet action wou!d !itt!einfluence how the design might unfold. Repeated review ofour transcripts eventual!y suggested that the inseparability ofthe two processes could be much more fundamental: wesaw them fused to such an extent that we could not disen-tangle them. Designing the nylon washer to fix the damagedFastTrack research instrument arm, for example, meant thatassembly informed design at the same time that designinformed assembly.

These field obseh/ations underscore an important definingquality of improvisation that has been described only general-ly in prior literature: improvisation requires material or sub-stantive convergence. Substantive convergence implies tem-poral convergence because actions in which design andproduction converge with each other substantively also con-verge in time. Temporal convergence, however, does notguarantee substantive convergence. Using these observa-tions as a guideline, we offer a refined definition: improvisa-tion is the deliberate and substantive fusion of the designand execution of a novel production. This revised definitionhelps to illuminate improvisation's standing relative to otherorganizational constructs. It implies that an organizationcould, at any given moment, do nothing, enact preexistingroutines in their usual patterns, plan an activity, execute aprior plan, or improvise. Table 1, which compares improvisa-

Table 1

Discriminant Validity among Improvisation and Reiatad Constructs (Extending Moorman and Miner, 1998b}

Features of Improvisation

Comparative constructs

Materialconver-gence of

design andexecution

Temporalconver-gence of

design andexecution Novel Deliberate

Comments about comparisonand/or field quotes revealing

differences

improvisation-- Deliberate-ly and materially fusingthe design and executionof a novel production

X Implied bymaterial con-

vergence

X X

Adaptation: Adjustmentof a system to extemalconditions (Campbell,1969; Stein, 1989)

Adaptation does not necessarilyinvolve temporal or action con-vergence. Adaptation can, infact, be achieved through plan-ning or deploying existing rou-tines appropriately (e.g., adjustnumber of fire trucks tc matchsize of fire). Some improvisa-tions may involve adaptations toexternal events, but improvisa-tions can also embody createdopportunities.

Bricolage\ Making dowith the materials athand (Levi-Strauss, 1967;Weick, 1993a)*

Improvisation increases thechances that bricolage will occurbecause there is less time toobtain appropriate resources inadvance. They are not the sameconstruct, hov>/ever, as bricolagecan occur in noninnprovisationalcontexts. Being skillful at brico-lage may help produce valuedimprovisation.

314/ASQ, June 2001

hnproviution and Lawmlng

Tabte 1

(Continued)

Comparative constaicts

Compression: Simplifyingand shortening steps toreduce the time it takesto complete each and thetotal process (Eisenhardtand Tabrizi, 1995)»

Materialoonver-gence of

design andexecution

Features of Improvisation

Temporalconver-gence of

design andexecution Novel

X

Comments about comparisonand/or lietd quotes revealing

Deliberate differences

Improvisation contrasts withcompression in which the stepsmove faster but are still distinct.We observed compression inthe field as this example sug-gests: "This is one where weare not just winging it. This isone where we are trying to fol-low the process and do it quick-!y"(FT 9/27/94),

Creativity. Intentionalnovelty or deviation 1romstandard practice (Ama-bile, 1983)*

Creativity may involve absolutelyno improvisation, as when an"off-line" plan or design is itselfcreative. A creative idea mightnever be executed. Creativitymay, however, represent anunusually valuable competencefor improvising organizations.

Innovation: Deviationfrom existing practices orknowledge (Zaitman,Duncan, & Holbek, 1973;Van de Ven and Polley,I992)t

Innovation may be createdthrough improvisation, but inno-vation can also be created byplanning. Innovation is a neces-sary feature of improvisation,but this does not Imply that allinnovation is improvisation—onlythat improvisation is a specialtype of innovation.

Intuition: Typically exam-ined at the individual leveland conceived of aschoices made withoutformal analysis (CrossanandSorrenti, 1997)*

Intuition is a part of some impro-visation, but Improvisation canoccur without using intuition,such as in the collective improvi-sation described by Hutchins(1991).

Learning: Experienceinforms a systematicchange in behavior orknowledge {Argote,1999)^

X X Improvisation is a special caseof learning, but learning from theorganization's own experiencecan also be achieved throughexperimentation, trial-and-erroriearning. and refinement or vari-ance reduction (March, 1998).

* This process or activity may be a frequent or even universal element of improvisation but is conceptually distinct andmay occur in activities other than improvisation.* Improvisation can be seen as a special case or type of this much broader process.

tion with related constructs, modifies and extends a discus-sion in Moorman and Miner (1998b).

Focusing on the substantive convergence in improvisationalso eventually persuaded us that improvisation, in fact, canbe conceived of as a special form of short-term learning. Wedraw on the well-established view of learning as occurringwhen experience generates a systematic change in behavioror knowledge (Levitt and March, 1988; Argote, 1999). Ourobservation of substantive convergence underscores thatthere is indeed a direct link between experience and changed

315/ASa June 2001

in many cases, the improvisation activityhad other referents as well. This wassometimes a preexisting behavioral rou-tine, product, or interpretive framework.Organizational improvisations could,therefore, also involve dual referents,including real-time problems or opportuni-ties as well as preexisting artifacts andbehavioral and interpretive routines. Thepresence of such dual referents may pro-vide a more constrained action spacewithin which instrumental improvisationoccurs, in contrast to aesthetic improvisa-tion, although this intuition remains to betested.

action (either behavioral or cognitive in nature) during improvi-sation. In improvisation, real-time experience informs novelaction at the same time that the action is being taken {duringthe act). This contemporaneous quality contrasts sharply withseveral other types of learning in which prior experience hasthe key impact on changed behavior or knowledge. This isnot to say that prior experience is not an important factor inthe incidence and effectiveness of improvisation, as othershave observed (Weick, 1993a; Hatch, 1997a; Moorman andMiner, 1998b), but the impact of real-time experience onaction is the defining characteristic of improvisation.

Improvisational Referents

In addition to looking at the different forms of productions,we also reviewed the data with an eye to factors that shapedthe specific designs or composition that unfolded in eachcase. What were the organizations paying attention to asthey improvised? We observed several types of these focalpoints but struggled at considerable length with how best tocharacterize the variety of these foci related to the improvisa-tional activity we observed. Were they best seen as triggers?Results? Goals? After contrasting them with reports of otherimprovisations, we came to see them as improvisational "ref-erents," borrowing a term from prior improvisational theory.Pressing (1984: 346) stated that improvisation typically has areferent "to facilitate the generation and editing of impro-vised behavior." Researchers who have focused on musicalimprovisation point to referents such as a melody, chordstructure, or even a rhythmic structure that provides animplicit starting place and continuing touchstone for theimprovisational activity (Weick, 1993c; Berliner, 1994; Hatch,1997b). As Charlie Mingus is reported to have said, " . . . youcan't improvise on nothing; you've gotta improvise on some-thing" (Kernfeld, 1995: 119). The referent both infuses mean-ing into improvisational action and provides a constraint with-in which the novel activity unfolds. In our firms, unexpectedproblems, temporal gaps, and unanticipated opportunitiesprovided crucial referents that anchored and constrained theimprovisational episodes.^ These referents and their impacton the process helped us probe further into the distinct traitsof improvisational learning and enabled us to compare impro-visational learning with the experimental learning processeswe observed in these firms.

The formal plans for the projects we observed includeddetailed time lines, technical steps, and extensive outlineswith subheadings for different aspects of product features,costs, and issues. The plans anticipated many problems inadvance, such as dealing with the technical limitations of aparticular electrical device or anticipated tensions betweenknown consumer preferences. Despite these exhaustive pro-cedures, unexpected events arose. Improvisation often fol-lowed several types of such surprises.

We observed early on that the teams often used improvisa-tion to solve problems created by surprises in bringing aproduct to market. The FastTrack team faced design prob-lems, materials that did not behave as expected, supplyshortages, and faulty parts from suppliers. Missing package

316/ASQ, June 2001

Improvisation and Learning

prototypes caused SeeFoods to improvise new ways of con-ceptualizing food shapes and sizes without guidance abouthow they might fit into the package (SF 04/25/95). Some-times the surprise was an unexpected problem that created aproblem with timing. For example, improvising the nylonwasher allowed FastTrack to keep its product in the marketwithout massive redesign of a part that was to be replaced inthe next generation of the product (FT 6/30/95). The value ofthis and similar jerry-rigged improvisations is that they permit-ted the organizations to move forward in one area despiteunanticipated problems in another or to help solve the thornyproblem of migration from one product generation to another(Garud, Jain, and Phelps, 2000).

Unanticipated opportunities produced a different type of sur-prise. Improvisation was sometimes used to create valuefrom serendipitous conversations and chance encounterswith customers, suppliers, and other organizational mem-bers. A FastTrack project manager described one such occa-sion, in which he received an unexpected call from a col-league in another division: "[This engineer] called me andsaid: 'I know [you have] this company in Boston and I amgoing IQ be in the area, a ten-minute drive away.' So [he toldme] he could take another day or half-day and make thisother visit" (FT 10/31/94). The project manager had notplanned to get information from the Boston company butseized the moment and generated a list of questions for hiscolleague to ask. In other situations, opportunity-linked impro-visations were triggered by an interaction with materials andpeople, as when FastTrack turned the unexpected outcomeof a bug fix into a marketing feature (FT 12/05/94).

In some instances, team members initially framed their activi-ty as solving a problem, but as they improvised, they gener-ated novel actions or interpretations that transformed theproblem into a perceived opportunity. In the case of Fast-Track's parts-variability problem, for example, participantsconsciously provided a new interpretation and related actionsfor an unchanging fact (high part variance). SeeFoods' miss-ing packaging prototypes initially posed a problem, but teammembers discovered during and after their improvisationthat, from their perspective, the problem had turned Into anunanticipated opportunity. This frame shifting in improvisa-tional referent is interesting and represents an importantpotential role for improvisation because of its possible role inimbuing the same situation with quite contrasting meanings.

The referents we have described were all instrumental in thatthey sprang from efforts to advance organizational goals.Although this is harder to substantiate from the direct tran-script data, we also concluded that some improvisationalactivity involved noninstrumental referents. When FastTrackengineers improvised the safety cover, for example, a keyreferent appeared to be its design value and parsimony, eventhough they also rationalized the production in terms of apractical opportunity.

Improvisational Learning vs. Experimental Learning

The specific nature of the improvisational referents gave thelearning it embodied two distinct features. First, the organiza-

317/ASQ, June 2001

tiona! focus was on specific problems and opportunities andon the specific medium or materials involved in working withthem. Even when improvisational activities involved novelinsights or recombinations of past practices, they were tight-ly linked to the specific local issue and time. Second, knowl-edge creation itself was never the goal or main focus of anyimprovisational episode we observed. Insight or understand-ing, if they arose, were collateral outcomes of the improvisa-tional activity.

Given these characteristics, improvisationai learning contrast-ed sharply with the experimental learning we observed inboth organizations, in which controlled situations weredesigned to produce, and typically did produce, some form ofnew knowledge. FastTrack prototype validation, for example,required prespecified tests under carefully controlled condi-tions (FT 10/5/94). Experiments using fish tanks examinedtemperature (FT 6/19/95) and equilibrium (FT 4/17/95).SeeFoods also deliberately sought information in controlledways before assessing it, as when it researched the proper-ties of packaging materials (SF 1/18/95) and used focusgroups that responded to carefully designed examples andquestions (SF 03/14/95). These activities clearly fit standarddefinitions of experimental learning, in which the learnerdeliberately creates contrasting situations in order to gener-ate systematic experience (Cook and Campbell, 1979). Thedeliberate knowledge-discovery in such experimental learningcontrasted sharply with improvisationa! learning as weobserved it. Table 2 compares this type of organizationallearning with improvisational learning, as well as with trial-and-error learning, which we discuss in more detail in a latersection.

In experimenting, the organizations deliberately varied activi-ties and conditions, such as changing the temperature inwhich a part was tried or giving potential customers differentsizes of a product. The nature and degree of this variationwas typically planned in advance and was designed to elicitgeneral, explicit knowledge about causal factors. Whenimprovising, however, teams typically sought no more varia-tion than was needed to address the immediate problem orpossibility.

This is not to say that improvisational learning never pro-duced new cognitive knowledge. During some improvisation-al troubleshooting in both firms, the teams sometimesgained new insights into technical facts or consumer prefer-ences they had not known in advance. This new knowledgewas a collateral—not an intrinsic or even intentional—out-come of the improvisation and was constrained by the specif-ic material, temporal, and cognitive situation. As shown intable 2, the improvisation we witnessed met the core defini-tion of learning but represented a somewhat unorthodox typeof learning: long-term knowledge creation was only a byprod-uct, if it occurred, and the experience was highly localized intime and context. These unusual features brought us to oursecond major focus, as we turned to our data to explore howimprovisational learning relates, if at all, to long-term organi-zational learning and adaptation.

318/ASQ, June 2001

Improraation and Learning

Table 2

Characteristics of impro^risation vs. Experimentai and Trial-and-error Learning*

Characteristic Improvisational learning Experimental learning Triat-and-error learning

How experience links tochange in behavior orknowledge

"Real-time" experienceinforms the design of per-formances or productionsas they are executed.

Varied "off-line" experi-ences are deliberately cre-ated and examined to seehow actions under differentconditions at one time pro-duce varying outcomes at alater time.

Actions are taken "on-line"and their consequencesoccur. The total experienceof actions and outcomesinforms further action orknowledge.

Typical purpose of learning To solve a surprising prob-lem and/or create valuefrom an unexpected oppor-tunity.

To acquire new informa-tion, knowledge of relation-ships, or causa! laws.

For intentional trial-and-error learning, to carry outregular activities and priorplans, observe outcomes,and then revise futureaction or understanding asneeded. Unplanned trial-and-error learning may alsooccur (Miner, 1989).

Nature and extent of initialvariation in inputs andprocess

Low: No preplanned delib-erate variation in inputs.Organization draws on real-time information to gener-ate a specific new activitypattern focused on localcontext.

High: Inputs are deliberate-ly varied and contexts com-pared so outcomes can beattributed to inputs.

Medium: Organizationsoften repeat activity Uiatappears to produce suc-cessful outcomes andavoid activity if disappoint-ed. Contrasts of before-and-after focal behavior arealso possible {March andOlsen, 1976; Cheng andVan de Ven, 1996).

Post-hoc reflection on linksbetween actions and out-comes

Absent or low: Little auto-matic reflection becausethe focus of the action is toget a problem solved or totake advantage of a specif-ic opportunity.

High: Reflection is highbecause observing out-comes under varied condi-tions is the goal of activity.Likely to notice absence ofdilferences as well as con-trasts with expectations orthe past.

Moderate: Unexpected out-comes of activity may benoticed. Reflection on out-comes of behavior canoccur, but is not necessary.Unlikely to notice thingsexpected or simitar to past.

Ouallty of knowledgegenerated

Behaviors are likely to havelocal value (tailored to avery specific context) andknowledge to be idiosyn-cratic to time or place.

Knowledge or behaviorsgained are more likely to begeneralizable, systematic,and contain informationabout main and interactioneffects.

Knowledge may be generalfrom contrast of before-and-after effects but still belocalized and involve simplecomparisons-

* We focus on these three types of learning because they are all capable of producing novel outcomes, unlike othertypes of learning, such as practice teaming or variance reduction, which do not involve innovation.

IMPROVISATION AND LONG-TERM ORGANIZATIONALLEARNING

Repeated review of the transcripts clearly showed thatimprovisation had a widely varied long-term influence onorganizational activities. In some cases, the outcome wasfleeting, with little or no lasting impact on the firnn or eventhe project, tn these cases, it could be said that there wereno long-term learning outcomes, and no substantial changeoccurred in organizational memory. In other cases, the impro-visational production or collateral insights generated duringimprovisation became a more permanent organizational fea-ture, whether within a particular project or more broadly inthe organization. In those cases. Improvisation played a rolein long-term learning, although we make no assumptionsabout whether such long-term learning is either adaptive oraccurate (Levitt and March, 1988; Levinthal and March,

319/ASQ, June 2001

1993). Some improvisational productions seemed to repre-sent "trials" for long-term organizational trial-and-error learn-ing. In contrast to this positive role, improvisation also some-times had deleterious effects on both long-term learning andadaptation.

No Long-term Impact of improvisation

Some improvisational activities were ephemeral and left notrace in the organizations. At FastTrack, for example, a teamimprovised a computer system, found it did not work asexpected, and quickly forgot it (FT 11/23/95). Both the eventand any potential lessons from it left no trace that we coulddetect. Teams also improvised new procedural steps thatthey did not seem to repeat later or even seem to remember.SeeFoods improvised an unplanned switch to emphasize aline of plain, high-taste, and low-fat products (SF 02/24/95).The improvisation seemed to be critical, but the team did notsustain the approach nor remember it as a rejected alterna-tive. Similarly, SeeFoods' use of internal focus groups did notalter its established product development processes in anypermanent manner.

Improvisation as an Element in Long-ternn Trial-and-errorLearning

in other instances, improvisational episodes seemed to havelong-term impacts. Reviewing multiple improvisationalepisodes revealed that improvisational productions some-times served as grist for higher-level processes of long-termtrial-and-error learning, although they did not always do so.Trial-and-error learning refers to a process in which an organi-zation takes action "on-line," and the consequences of thataction lead to change in action or knowledge base (Marchand Olsen, 1976; Cheng and Van de Ven, 1996; Van de Venand Polley, 1992). In behavioral trial-and-error learning, theorganization may simply repeat apparently successful actions.In more cognitive trial-and-error learning, observers mayreflect on the consequences to develop new causa) modelsor information. To explore how improvisation played a role inthis, we first defined the notion of retention of an improvisa-tional production. Retention in this context refers to thedegree to which the improvised production or collateraEinsights are made a permanent part of the organization's arti-facts, processes, or knowledge (Miner, 1989; Van de Ven andPolley, 1992; Moorman and Miner, 1997; Argote, 1999).

Some improvisation productions were retained by the organi-zations we observed. At FastTrack, the improvised circuitboard design and the product's safety cover (artifactual pro-ductions), as well as the buried software interface (new orga-nizational process), were retained throughout our study peri-od. Some problem-oriented improvisational troubleshootingproduced outcomes that did not work, and the team retainedthese insights for future reference, in other cases, FastTrackcustomers became aware of a scientist special andexpressed interest in obtaining it. In rare cases, the firm thenfroze a product's specifications, named it in formal productlists, and transferred production to the manufacturing group.Thus, an idiosyncratic production linked to a specific local set

320/ASQ, June 2001

Iniprovlsation and Lavning

of constraints became a replicated product embedded in thewhole set of standard practices and products of the firm.After SeeFoods successfully moved its breakfast product intothe market in a quarter of the typical time, the firm retainedseveral of the strategy's features for use, even in situationsnot requiring fast response. This retention involved not onlyrepeating some of the improvised steps but reflecting onhow they worked and developing new mental models of theproduct development process (SF 12/9/98).

These patterns of selective retention and replication of cer-tain improvisational productions and collateral insights showthat improvisation can serve as the first step in iong-ternntrial-and-error learning (March and Olsen, 1976; Van de Venand Polley, 1992). The patterns also help clarify how improvi-sation is distinct from trial-and-error learning itself. Table 2,above, depicts how improvisation compares conceptuallywith trial-and-error iearning. A complete improvisationalepisode occurs when the improvised production is complete.The improviser cannot and does not wait to know the conse-quences of the improvised production while executing it. Incontrast, a complete episode of trial-and-error learning occursonly after outcomes of action have been experienced, andnew actions or inferences arise, specifically based on theconsequences of completed action. When a FastTrack teamnoticed that a new material selected during the planningprocess didn't perform as expected (the trial and conse-quence) and revised plans for future products based on thisexperience, trial-and-error learning—but not improvisation—occurred. As shown in table 2, our data paint improvisationas a special form of short-term, situational learning that can,but does not necessarily, serve as a "trial" in long-term trial-and-error learning.

Table 2 also highlights that both trial-and-error and improvisa-tional learning involve on-line initial activity, in contrast toexperimentation, in which different things are typically triedoff-line. If triai-and-error learning involves deliberate reflectionon the causal processes that produced results, however, thenit represents a hybrid that incorporates both on-line and off-tine learning activities (Gavetti and Levinthal, 2000). Becauseexperimental learning typically involved more variation inactions and context than either improvisational or trial-and-error learning, it could generate more genera! knowledge.

Improvisation's Harmful Impact on Long-term Learningand Adaptation

Although we noticed how improvisation could play a positiverole in long-term trial-and-error learning, our extended reviewof field data also revealed several ways in which the improvi-sational activity impeded other teaming processes. For exam-ple, the SeeFoods team put the development of a salad lineon hold to pursue sandwiches so the team could focus itsenergy on only one project (SF 05/23/95). Although thisimprovised activity temporarily focused the team's action,because the action ran counter to research evidence onecost of this focus was the loss of knowledge gained in priorresearch and prioritization. Improvised activities involvingFastTrack scientist specials were also sometimes seen as

321/ASa June 2001

distracting from the original plan for a particular product or aslikely to create knowledge that would not be integrated wellwith other activities (FT 11/28/94). We also observed, andsenior officers stated, that scientist specials sometimes drewresources from preexisting strategy that was the result ofintensive investigation about market potential, broad cus-tomer preferences, and firm-level objectives (FT 12/2/94).Table 3 notes this destructive potential of improvisationallearning a\or\g with other long-term impacts as comparedwith other types of learning.

In some instances, improvisational activity seemed to replaceor foreclose more formal experimentation, as when the Fast-Track team members debated whether to improvise a proto-type in the absence of drawings or to wait for the drawingsso that they could set up tests and experiments:

Mfg. Rep: We're ready to build but don't have the documents tobuild.

Table 3

The Impact of Improvisation vs. Experimental and Trial-and-error Learning on Long-term OrganizationalLearning*

Impact Improvisational learning Experimental learning Trial-and-error learning

Long-term retention oflearning outcomes

Low: The focus of theactivity is not to generatenew knowledge but to getproblem solved rapidly ortake advantage of specificopportunity.

High: Because long-termlearning is the goal of theactivity and formal proce-dures are therefore likely toexist to capture findings.

Moderate: Learning may ormay not be the goal.Regardless, the organiza-tion may repeat actionswith apparently good out-comes, and actors maynotice and use systematicoutcomes from prior activi-ties.

Impact on other organiza-tional activities

Activity is on-iine andimpact may be difficult toanticipate and mitigate. Ifcontext has high interde-pendence, improvisation ismore likely to have unex-pected harmful impact.May draw attention awayfrom prior learning andrelated priorities.

Action is off-line and exper-iment is completed beforeconsequential action thathas implications for theorganization is taken. Canexamine and managepotential interdependenciesbefore final action. Likely tobe integrated with prior(earning and related priori-ties.

On-line actions and out-comes drive learning, butoff-line reflection and/orchanges to action occuronly after consequences ofprior action are known.May be less likely thanimprovisation to produceunexpected harmfulimpacts associated withinterdependencies, butmore likely than experimen-tation.

Link to higher-level capabili-ties

Organizations can developand maintain capabilities atfruitful improvisation. Addi-tional capabilities may beneeded to link this short-term form of learning tolong-term organizationallearning.

Organization can developand maintain explicit capa-bilities in experimentationin the form of researchskills.

Organization can developand maintain capabilities atintentional trial-and-errorlearning through proce-dures that allow it toobserve the outcomes ofits own actions {i.e., post-mortems) and deriveknowledge and/or differentactions from them.

How improvisational learn-ing interacts with theseother types of learning toaffect long-term learningprocesses

Improvisational learningitself occurs in real-time,with no intrinsic long-termlearning outcomes.

Improvisation can drive outexperimentation.

Improvisations! episodescan serve as trials for long-term trial-and-error learningprocess.

• We focus on these three types of learning because they are all capable of producing novel outcomes, unlike othertypes of learning, such as practice learning or variance reduction, which do not involve innovation.

322/ASQ, June 2001

bnfHvvitation and Learning

Project Leader: I'm afraid that if we wait for all the assembly draw-ings, it will be too late. If, for example, we find a cover that has tobe modified.

Mfg. Rep: Our purpose now is to identify everything that has to bedone in a pilot build. . . . Verifying main bench performance, run-ning through test specs, running to maximum power—any of thesemay give us action items to resolve. Realistically, there should be aformal test plan written. [FT 12/12/94]

In this instance, the manufacturing representative stronglyfavored the pre-planned "formal test plan" to generate sys-tematic knowledge. The project leader felt forward motionhad top priority and favored improvisation. How and whetherimprovisation may drive out experimentation obviously hasimportant implications for improvisational learning's impact onorganizations.

Even if it did not reduce the value of prior learning or driveout experimentation, both firms observed that processes wewould define as improvisation were seductive but dangerousto their long-term strategies because they could fragmentfirm efforts. As one FastTrack respondent told us, "[The com-pany! was jumping the gun on every new opportunity, andanybody who had an idea for something immediately startedpursuing it. We split resources quite a bit, and also we lostfocus somewhat from the core business. . . ." (FT 9/25/94).

Beyond the fragmentation of whole product lines, improvisa-tion could potentially produce actual incompatibilities at vari-ous levels. At one FastTrack team meeting, the group agreedthat a prior team's improvisational activities created problemsthat reduced the current team's ability to fix other problems(FT 12/05/99). In another session, they considered improvis-ing around a prior design but decided to exercise caution: "Itold Dan we want to cut off 3/8 of an inch. That would beideal. But if it looks like we're going to have to start movingthings around . . . which equals time . . . which may meanthat more doesn't work because the layout changed . . .don't do it" (FT 01/01/95). Here, the team worried aboutpotentially incompatible physical features of the product, inaddition, multiple improvisations of individually attractiveproduct features could also produce specification creep,which made the product no longer match the original marketfor which it was targeted.

The FastTrack team seemed more comfortable improvisingtechnical features that were part of relatively independentsubassemblies rather than improvising elements that werelikely to interact with the whole complex product design.Such a pattern is consistent with the notion that improvisa-tional adjustments are more problematic when product fea-tures are not independent of one another (see table 3). Thedanger of improvising whole product features is intuitivelyclear, because they must all eventually work together in aphysical product. Less obviously, improvisation could produceincompatibilities between new product processes underwayand/or ways of thinking about the project, although we didnot observe such incompatibilities directly.

Last, improvisation sometimes generated organizational con-flict. In one team, for example, the manufacturing representa-

323/ASQ, June 2001

tive frequently argued against continuing improvisation thatled to changes in product features and expressed dissatisfac-tion with the lack of final specifications: "There's a lot ofwork that must be done this week which stems from thefact that the design wasn't frozen" (FT 12/12/94).

In some prior work on product development, observers havespeculated that psychological or cultural differences amongscientists and marketing and manufacturing professionalsproduce different preferences for formalization and early lock-in (Dougherty, 1990; Tushman and Anderson, 1997). Closereview of our data suggested that the conflict may also arisefrom the different ways in which improvisation's value anddangers influenced each group's responsibilities. Manufactur-ing was where "the rubber hit the road" in terms of anyproblems with incompatibility and inconsistency. Once thephysical product is assembled and shipped, only a productrecall permits another change. Marketing and scientific pro-fessionals saw improvisationa! activity as tools for flexibilityand adaptability, which sometimes generated conflict at high-er levels with financial and manufacturing managers, whotended to see improvisation as a dangerous source of ineffi-ciency and costly errors.

In spite of these dangers of improvisation, both firms permit-ted and, in some cases, facilitated improvisational activity.This apparent puzzle leads to the final link we observedbetween improvisation and long-term learning: both firmsappeared to have developed several distinctive competenciesrelated to improvisation itself.

Organizational Competencies in Improvisation

Repeated reviews of our data indicated that the organizationshad gained something more fundamental from improvisationthan the solutions to specific problems, exploitation of newopportunities, and the adaptive gains from repeating usefulimprovised productions. We observed that the organizationsappeared to have learned how to limit the special risks ofimprovisational learning and/or to facilitate fruitful higher-levelimprovisationa! learning. In considering these practices, weconcluded that both organizations appeared to have devel-oped organizational competencies related to improvisation indifferent functional areas. We could not directly observe theactual development of these competencies, which appearedto have unfolded over many years in each firm, but weobserved consistent manifestations of these competenciesat several levels of analysis.

Generating improvisational activities. We observed meta-routines in each of the organizations that enab!ed them togenerate or faci!itate improvisation. It appeared that at onetime in FastTrack's history, scientist specials did not have adistinct name or set of norms associated with how theymight unfold. By the time we observed the firm, FastTrackappeared to have several ways in which scientist specialscould unfold, but informal judgment was invo!ved in how totackle them. !f, for examp!e, an idea for a scientist specialapproximated a modification of an existing project, it couldproceed via completion of a special form. Other scientist spe-cials were simply noted as special projects, although by the

324/ASQ, June 2001

Improvisation and Learning

end of our observation period they had received a specificdesignation indicating that engineers would design and per-sonally produce the product. Not surprisingly, the companywas more iikely to approve such whole-product improvisationwhen the proposed scope was small.

SeeFoods also appeared to have institutionalized higher-levelroutines that were an ongoing source of improvisation. Priorto our observation period, the product design group hadretained a consultant who worked with groups to providenew product ideas. Because few of these ideas becameproducts, a vice president developed a model that attemptedto avoid segregating planning and execution in productdesign. We observed several carefully orchestrated interac-tions between a product team and a host of internal andexternal visitors who worked together for a day on a particu-lar strategic-level question regarding a new product. The keyfeature in the context of improvisation was the firm's deliber-ate effort to merge execution issues, such as material andmanufacturing details, with the creation of the entire productline. Therefore, although specific activities were scripted,they were designed to achieve improvisation. Interviewsrevealed that SeeFoods prided itself on maintaining a culturethat condones and even encourages improvisation, withoutusing this term to describe it.

Generating valued improvisations. The firms' practices thatfacilitated improvisation also influenced the nature of whatwas improvised. Although it is very difficult to make system-atic claims without more controlled comparisons, we thoughtthe firms tended to improvise in certain areas with goodresults. FastTrack, for example, displayed a competencyspecifically for technical improvisation at several levels ofactivity that it deliberately sought to retain in a certificationprocess it finalized during our observation period.

The new administrative standards required that all productsfollow a formal, written development process that would beaudited to ensure compliance, with no exceptions whatsoev-er. Although scientist specials did not follow linear productdevelopment planning, check-offs, reviews, or any formalprocedures, the company did not eliminate scientist specialsor even add any structure to govern them. Instead, it created,on paper, a new "Track III" official product developmentprocess that involved no prior planning, market evaluation,design specification, or production planning, A technical teamimprovising a scientist special could then label its activities a"Track HI" product and continue to improvise an idiosyncraticproduct with no change in its usual activities (FT 04/17/95).The organization's decision to legitimize product improvisa-tion seemed to reflect its confidence in and evaluation of itsown technical improvisational competency. FastTrack recordsand behavior did not reflect a parallel confidence or apparentskill in other improvisational domains.

Other observations indicated that the firms had developedcompetencies at generating fruitful improvisation throughmeta-routines that could reduce the chances of harmfulimprovisations. FastTrack fostered two types of complemen-tary safeguards against the dangers of improvisation. In par-

325/ASQ, June 2001

ticular, the teams were aware of potential specification creepin regular projects. In addition, scientist specials were con-strained by the formal structure of the organization and byengineers' own internalization of the firm's goal ofprofitability.

SeeFoods dealt with some potential dangers of improvisationby emphasizing certain windows of time in which improvisa-tion was valued and using it during those time periods. Thisapproach resembles the improvisational patterns that jazzmusicians follow when allocating blocks of time during whichimprovisation occurs. As at FastTrack, the evidence pointednot just to organizational channels permitting improvisationbut also to a tendency for the improvisation to be productive.SeeFoods considered some procedures for stimulatingimprovisation proprietary, attesting both to the firm's aware-ness of its improvisational competencies (although the com-pany would not label them as such) and belief in the potentialvalue of these procedures. Evidence of a complex, routinizedimprovisationai competency was also found in follow-upinterviews one year later, when it was discovered that thefirm had evolved from the use of ideation sessions to pro-duce improvisations to a more complex process in whichteam members used a fused design to prepare prototypesthat they shared with consumers on the following day.SeeFoods management believed that this process could trun-cate the development time by half for a new product devel-opment effort.

Harvesting valued improvisations. The two firms some-times retained individual improvisational productions they cre-ated and replicated them in other settings. In both organiza-tions, we discovered that improvisational productions—whether a procedure or an artifact—were selectively harvest-ed. In such selective retention, apparently unsuccessfulimprovisations were not routinized. Therefore, the presenceof these competencies involved a complex relationshipbetween real-time improvisational learning during the produc-tion of an individual product, feature, or process and latertrial-and-error learning, in which the improvisational episodeserved as a first step. FastTrack's ability to harvest technicalimprovisational productions within a given project seemed toreside primarily in the observational skills and memory ofspecific teams and engineers who paid close attention toactivities and served as an informal organizational memory.At a higher level of analysis, the deliberate harvesting of theentire scientist-special process to reproduce a previously one-of-a-kind product for a wider market also represented a com-plex, organization-wide improvisational competency.

SeeFoods also harvested many valued improvisational pro-ductions. Management encouraged team leaders to reviewprojects and to document deviations from plans—includingany improvisationa! activity—so that they might be evaluatedand, if valuable, possibly be incorporated into future projects.SeeFoods' senior management also reviewed and theorizedabout the micro-dynamics of formal and informal productdevelopment processes, including how the organization couldencourage teams to !earn from one another. Reports on pro-cedural changes in product development over time suggest-

326/ASQ, June 2001

hnpro^sation and LMming

ed that SeeFoods had a skill in recognizing fruitful processimprovisations, including those generated in response to sur-prising events, along with skill in trying to deploy them inother settings. Consistent with industry trends toward delib-erately trying to codify more organizational knowledge, uppermanagement appeared to systemize cross-project learningthrough formalized interteam knowledge transfer procedures.During the period of our observation, however, the informalorganizational competencies developed to detect and retainapparently useful innovations, including improvised actions,seemed to pre-date this more explicit effort.

Our observation of these competencies still leaves ampleroom for additional biases and dangers in generating and har-vesting improvisation {Miner, Moorman, and Bassoff, 1997).Our research strategy revealed evidence of their existence,but we could not assess whether the competencies weresufficient to overcome fragmentation, for example, or tomaximize improvisations' full benefits for these firms.

DISCUSSION AND CONCLUSION

Extensions and Contrasts with Existing ImprovisationLiteratureOrganizational improvisation. Much theory and research onimprovisation has focused on individuals and small groups,but, consistent with work in other fields, our study highlightsorganizationa! improvisation. While our field observations sug-gested that individual improvisation was often part of team-or organization-level improvisation, other processes were col-lective, as when the entire group devised a product changethat no single individual had designed. These findings parallelother research demonstrating that a collective can design anew action pattern without advance planning and even with-out members' awareness of the pattern as they execute it(Hutchins, 1991). Furthermore, the distinct competencies inimprovisation did not appear to reside in specific individuals;rather, they flowed from broader organizational routines, cul-tures, and collective capabilities. These factors also implythat improvisational action can occur and be studied at anylevel of analysis, including strategic improvisation by an entirefirm.

Previous work tends to report on one specific form of impro-visation, but our observations indicated at least three types ofimprovisational productions (behaviors, artifacts, and interpre-tations), as well as varied instrumental referents, such asproblems, temporal gaps, and opportunities that both focusand constrain improvisational action. This variety stimulatedus to consider several basic aspects of improvisation as aconstruct and activity. Future research can fruitfully considerthe implications of different improvisational forms and refer-ents for organizational performance.

The value of improvisation. Hatch (1997a) has suggestedthat organizational researchers may fall prey to romanticizingimprovisation's likely value for organizations or the difficultyol generating high-quality improvisation. Our observationssupported this concem, but they also advance our under-standing because they make explicit several specific, distinct

327/ASa June 2001

ways that improvisation can both bear fruit and raise dangersfor organizations. The observations suggest that futureresearchers should not expect a single benefit from improvi-sation; instead, we should anticipate and examine such dis-parate valued outcomes as temporary action patches, trou-bleshooting, the successful use of serendipitousopportunities, and even aesthetic creations.

Our findings also imply that the improvisational contextshould influence the trade-offs between benefits and dan-gers. They are consistent with the intuition that the degree ofunderlying interaction in the system, sometimes termed epis-tasis (Kauffman, 1996; Levinthal, 1997) or complementarity(Milgrom and Roberts, 1990), wiil moderate improvisation'simpact in two ways. First, in highly interactive systems, indi-vidual improvisational actions have a greater chance of caus-ing inconsistency and coordination problems than they wouldin a modularized system. Although this is easy to see in thecase of product features, the same logic applies to interac-tions among behavioral routines, interpretations, or wholestrategies. Second, improvisational activity often precludessystematic variation in several different causal factors, so it isless likely to produce good knowledge about interactioneffects themselves (Kauffman, 1996). Aggressive investiga-tion of how this and other contextual factors may influenceor bias the value of improvisational learning seems especiallypromising for further work (Miner, Moorman, and Bassoff,1997).

In contrast to Moorman and Miner's (1998b) prior predictions,we did not find that these firms had developed generalimprovisational competencies; rather, they had developedimprovisational competencies in very specific areas. This find-ing may be partially explained by Schon's (1983) claim thatthe medium of interaction (e.g., an architect's sketch pad, therelation between a patient and therapist, or the materialsused by FastTrack engineers) has a crucial impact, such thatcompetencies are rooted in a specific domain. AlthoughSchon had in mind improvisation by individual professionals,our findings suggest that the specific medium of organiza-tional interaction might similarly shape and bind organizationalimprovisational competencies to particular areas of activity.This possibility underscores the potential value of researchon how improvisation may help create and sustain distinctcommtinities of practice In organizations (Brown and Duguid,2001).

Defining improvisation. Our work also refines some priordefinitions of improvisation while encouraging resistance tothe inclusion of unnecessary additional features in its defini-tion. Specifically, our emphasis on substantive rather thantemporal convergence of planning and execution should helpempirical researchers distinguish improvisation from rapidsequencing of distinct action steps, which Eisenhardt andTabrizi (1995: 85) called a "compression strategy" when donedeliberately to achieve rapid product development. As organi-zational memory and organizational processes becomeembedded in rapid electronic processing systems, this defini-tional refinement should help keep improvisational research

328/ASQ, June 2001

Improvisation and Lawning

clearly distinct from research that investigates compressedstages.

Some improvisation scholars have implied that its definitionshould include such constructs as the use of the improviser'srich prior background or memory as part of improvisationitself (e.g., Crossan and Sorrenti, 1997; Cunha, Cunha, andKamoche, 1999). We suspect that several of these suggest-ed additions arise because the writers, in developing theirdefinitions, have focused on examples of successful improvi-sation carried out by skilled improvisers. Our field observa-tions made clear that improvisation can also be unskilled andcan cause harm. This convinces us that it is crucial to avoiddelimiting improvisation in ways that restrict it to success ifwe are to build an empirical literature that can elucidate fac-tors that moderate improvisation's impact. Some organiza-tional learning research, for example, has suffered from atendency to conceptualize learning as occurring only when itis accurate or useful. This has sometimes choked offresearch on the factors that influence when learning is and isnot accurate or useful (Levinthai and March, 1993; March,1998; Miner and Anderson, 1999). In this formative stage ofthe empirical literature on improvisation, we argue for thebasic definition used in this study to leave open for futureresearch questions of how it is correlated with other impor-tant activities.

In many cases, we observed improvisation emerge undertime pressure to solve problems or address opportunitiesquickly. This is consistent with other work indicating thatexternal time pressure, coupled with lack of relevant priorroutines, may well be a common trigger for improvisation(Weick, 1996; Baker, Miner, and Eesley, 2001). We do notinclude it in our definition, however, because inactivity isalways possible, and there were other triggers for improvisa-tion even in our goal-oriented setting, such as the safetycover that FastTrack engineers just made up as they wentalong. The prediction that external time pressure will increasethe odds of improvisation is an empirically testable predic-tion.

Having said that, we note that in this study we sought com-mon patterns in improvisation in action, which leaves consid-erable room for further empirical work on the ways in whichimprovisation may vary. First, considering the novelty compo-nent in our definition, improvisation episodes may vary in twoways: in the overall proportion of activity that is innovativeand in the degree to which the novel activities represent adeparture from preexisting plans or routines, regardless oftheir proportion of the whole. For example, an episode mightinvolve a large proportion of innovative activity, as when mostparts of a new product are improvised, but still exhibit lowradicality, as when the improvised new parts are similar toparts in prior products. Prior categorizations of improvisationalintensity have often merged these qualities, although theymay have different implications for the value and risk ofimprovisational activity.

Measuring improvisation. The emphasis on substantiveconvergence of design and implementation means that

329/ASa June 2001

Moorman and Miner's (1998b) suggestion that organizationalresearchers consider the length of time between plans andimplementation as an indicator of improvisation is incom-plete. Through computer-aided design, for example, a firmmight be able to reduce the time between product designand production but not engage in improvisation, because thetwo processes remain distinct. This emphasis also points tothe value of trying to measure the positive interaction thatmerges design and execution rather than viewing theabsence of formal planning as a measure for improvisation.

Prolonged interaction with our sites also convinced us thatimprovisation can and should be studied using both qualita-tive and quantitative approaches. Our initial observationswere driven by a qualitative analysis of meeting transcripts,interview notes, formal planning documents, and archivalinformation. In a concurrent effort, we also developed formalpaper-and-pencil tools to capture teams' impressions of manyaspects of a very small sample of 100 narrow actions thatoccurred early in our observations. The informants' ratings ofthe incidents on a 7-point scale indicating the degree towhich an action was extemporaneous was 4.52 (s.d. = 1.98).Our separate ratings as site observers of the same incidentswas 4.01 (s.d. = 1.54) on a scale on which 7 indicated "highimprovisation." This small, random sample was limited towithin-project actions and therefore did not match the set ofexamples of improvisation we detected in our review of thetranscripts and described in this paper. Nonetheless, the con-vergence of these ratings seems promising for both self-report measures and observational measures of organization-al improvisation. Future work can clearly enhance theirdevelopment.

Our observation that organizational improvisation can occur atdifferent levels of analysis raises another crucial measure-ment issue. We noticed that we needed to pay close atten-tion to the organizational level and time frame used as aframe of reference for judging whether improvisation hadoccurred. It was crucial, for example, to contrast scientistspecials with FastTrack's original strategic plan, which includ-ed approved product families and specific lines within thosefamilies, all based on prior reflection about how the firmshould deploy its attention and assets. Scientist specialsmeant that the overall strategic product development pro-gram was being modified as these products were created.Within the implementation of these scientist specials, therewere varying degrees of concurrent product design and exe-cution. Future research could investigate whether firms dis-play a consistent tendency to improvise at many levels atonce or whether they improvise at one level but not atanother.

Further, our interaction with the data underscored that onemust pay careful attention to the level and temporal pace ofregular organizational planning and innovation as part of a reli-able method to assess the occurrence of improvisation. Thisdoes not mean that assessment of improvisation is subjec-tive and depends on any special viewpoint of the observer. Itimplies that a standard step in assessing improvisationshould be to assess explicitly the level of organizational

330/ASQ, June 2001

Improvisation and Lewning

design or p!anning involved. One heuristic is to consider whatlevel of formal planning would be relevant to the activity athand and then contrast the action to that level of forma!design. We clearly saw that improvisation at different levelsof action can be independent.

Extensions and Implications for Organizational Learning

Our findings extend the menu of distinct organizational learn-ing processes by explicating improvisation as a special typeof learning. This raises the issue of how improvisational learn-ing contrasts with experimentation and trial-and-error learn-ing, because al! three can involve the creation of novelactions or knowledge. These contrasts suggest that improvi-sation represents a form of real-time, short-term learning thatmay or may not influence other learning processes. Second,improvisation is a double-edged sword that has the ability toenhance other teaming processes and to detract from theirvalue or operation. Finally, organizations appear to be able todeve!op competencies in generating and deploying improvisa-tion in spite of its dangers, although these competenciesmay be localized.

As described, both companies we studied had extremelywell-developed routines for knowledge discovery throughexperimentation. Because learning was the specific goa! ofexperimentation, transcripts revealed that outcomes wereoften carefully observed and recorded. The focus on inten-tional learning also encouraged reflection on potential mean-ings of different outcomes (FT 4/17/95, FT 5/15/95); in somecases, those involved noticed no differences between condi-tions. In contrast, as we noted earlier, the goa! of improvisa-tion was not learning per se. Rather, improvisation arose asthe firms struggled to deal with surprises. The intense atten-tion to one specific context and the often implicit rule of end-ing the improvisation when a problem was solved or an unex-pected opportunity was exploited meant that improvisationallearning was far more likely to produce context-dependentknowledge. Moreover, improvisational learning could not beused to test interaction effects in the way that experimenta-tion could. Finally, because the goal was not learning, knowl-edge outcomes were often not recorded or preserved forretention or transmission in the organization.

On the one hand, improvisation not only contrasted sharplywith but, in some cases, seemed to drive out experimentallearning, making the balance between them an importantissue—given their different value to organizations—^that hasbeen understudied. Improvisational learning episodes did notnecessarily drive out trial-and-error learning, however, but, infact, sometimes served as trials for such learning. Thesequence in which an improvisational activity was repeatedafter the organization registered its outcome amounts to triat-and-error learning in the end. We suspect this not uncommonsequence accounts, in part, for some of the conceptual andempirica! confusion between improvisation and trial-and-errorlearning. Both trial-and-error !earning and improvisational-!earning processes were grounded in a specific context,invo!ved actions with real consequences, and initia!!y lackedthe off-!ine qua!ity of experimenta! !eaming.

331/ASQ, June 2001

On the other hand, repeated improvisationat learning withouttrial-and-error learning can create "opportunity traps" thatproduce unintended drift or fragmentation over time. Withinan individual project, improvised improvements of specificfeatures can have an aggregate impact by the end of theprocess such that the product no longer matches the originalmarket niche. At the level of whole new products, uncheckedimprovisational activity can generate excessive numbers ofinconsistent products over time, a standard dilemma in thestrategic management of product development (Burgelman,Maidique, and Wheelwright, 1996). More broadly, then, ourobservations imply that improvisational learning that exploitsunanticipated opportunities may generate a chain of oppor-tunistic actions, each of which appears to embody adaptivelearning or benign opportunism (Miner, 1987) at the time.When taken together, however, the actions eventually canproduce de facto strategies that are not adaptive for theentire organization.

Finally, as noted above, because of their nonsystematicnature and one-at-a-time contrasts, both improvisational andtrial-and-error learning are especially likely to produce disap-pointing learning when there are strong interaction effects(i.e., complementarities or epistasis) in the learning context.

Considerable prior research has argued that preexisting mem-ory influences the execution and quality of improvisation, aswe noted in our introduction (Berliner, 1994; Pressing, 1984;Hatch, 1997a). Our study completes the loop by exploringways in which improvisation can in turn influence memory,thereby contributing to theories of organizational memory(Walsh, 1995). The observations point to no single inevitableway in which improvisation affects organizational memory,however, but highlight instead a variety of possibilities. Thisinvites further work on when and how such improvisation-induced changes in long-term memory occur.

Study Limitations

This study remains exploratory, in part because we limitedour observations to two firms in order to get sufficient detailsto observe improvisation directly over nontrivial lengths oftime. By finding common themes between a consumer firmand a scientific-product firm, we believe some generality wasgained, but clearly it would be helpful to examine improvisa-tion in a range of firms.

We also focused on improvisation specifically within productdevelopment, although prior work has also described improvi-sation (although not necessarily using this term) in production(Stoner, Tlce, and Ashton, 1989), job creation (Miner, 1987,1989), crisis response (Hutchins, 1991; Preston, 1991), orga-nizational technological change (Barley, 1986), and even mili-tary activities (Metcalf, 1986). It might be that improvisationis more likely within product development than within otherorganizational activities, although it is not clear that such afrequency bias would alter the potential for varied types andimpacts of improvisational learning.

Our focus on varieties rather than degrees of innovation inimprovisation, along with the fact that both firms had highly

332/ASQ. June 2001

and Lsaming

forma!ized product development procedures, reduces con-cern that the study's setting overdramatizes improvisation'spresence. Nonetheless, dose observation of improvisation inother organizationa! activities cou!d extend and, perhaps,even challenge the types of improvisation we observed.

Neither firm we studied was an artistic or high-technologyfirm, reducing the chances that the links to learning repre-sented idiosyncratic patterns for unusual firms. Both firmsrepresented thriving, actively adaptive firms, however, whichmay have enhanced their chances of deve!oping competen-cies at improvisation. Given our inductive approach, futureresearch should engage in careful testing of our conceptsand relationships in a genera!izab!e samp!e of firms.

Implications for Other Literatures

New product deve!opment represents an important mecha-nism through which organizations adapt to their contexts andsometimes shape the settings in which they compete. Eisen-hardt and Tabrizi (1995) argued that organizations can speedup product development by compressing distinct phases orby merging them in ways that inc!ude improvisationa! activity,although they did not direct!y observe improvisation itself.We did observe improvisation in the product developmentprocess and support their supposition that it occurs. Otherobservers have also noted improvisational processes withinthe technical development of products (Dougherty, 1990,1992; Leonard-Barton, 1995), but our study provides addition-al insight into the question of how improvisation occurs andmay even be institutiona!ized.'* Although this juxtaposition ofinnovation and order is not new (Burns and Stalker, 1961;Nord and Tucker, 1987; Je!inek and Schoonhoven, 1990),much literature still tends to dichotomize the two (Tushmanand Anderson, 1986; iVlarch, 1991). Therefore, by providinginsight into how firms sometimes operate to harvest andretain improvisation and by describing some of the mecha-nisms by which firms generate improvisations, we haveadvanced the literature in this area. Further, our resu!ts add toand reinforce trade-offs between the advantages and disad-vantages of improvisation in product development, creatingan area ripe for empirica! research.

Entrepreneurship. Improvisation a!so offers a ven/ promising!ens for investigating entrepreneuria! processes. Many entre-preneurs do not formulate careful plans when starting theirbusinesses but, instead, respond to sudden opportunities inways that fit our definition of improvisation (Baker andA!drich, 1994; Baker, 1995). Careful app!ication of the improvi-sation framework may he!p i!luminate key determinants oforganizations! founding and ear!y grov r̂th processes. Ski!lfulimprovisation in making do with materials at hand, for exam-ple, may cause troub!e for a new firm making the transitionto appropriate strategic planning and implementation (Baker,Miner, and Eesley, 2001).

At the same time, the entrepreneuria! context may benefitempirica! research aimed at further deve!oping theories ofimprovisation. !f access can be gained, start-ups offer a morecompact setting than do large, complex organizations forexamining specific links between design and implementation

333/ASa June 2001

of both strategies and tactics (Aldrich, 1999). We also sus-pect that new firms will vary in their improvisational capabili-ties, making this a promising area for examining factors thatenhance or detract from improvisation's potential value. Final-ly, close examination of founding processes may shed lighton the origins of improvisation in existing large firms. In ourfield encounters, we tended to assume that the firms haddeveloped their improvisational competencies over time.Empirical research on new organizations might reveal thatthe improvisational competencies we observed representoriginal competencies of firms that persisted as the firmsmatured, rather than new patterns the firms acquired overtime.

Models of emergent change. Our study also helped us seethat some apparent contradictions between planning andimprovisation are illusory. Although it seemed counterintu-itive, for example, our field encounters convinced us thatorganizations can plan to improvise, routinize processes tostimulate improvisation, and routinize the observation of theirown improvisational activities, ail without the actual contentof the improvisation being planned in advance. Our findingshave implications for literatures that emphasize the fusion ofunplanned innovation and order. These include models thatdescribe how emergent aggregate patterns arise from ratio-nal actions at a lower level of activity (e.g., Schelling, 1978;Arthur, 1987; Stein, 1989; Baum and Singh, 1994) and workthat describes unintended outcomes and actions that occurat the same level of analysis (e.g., Follett, 1930; Alinsky,1969; March and Olson, 1976). In contrast, the scientist spe-cial or improvised team behavior are not emergent outcomesof lower-level actions of which the improvisers are unaware.They are designed and enacted at the same level of analysis.They are also not an unintended outcome of other actions;they are the deliberate creation of a novel production.Emphasis on unplanned, novel, but orderly actions marksimprovisation as a narrow yet important type of emergentorder (Berger and Luckmann, 1967; Giddens, 1984; Feldman,2000). Its most distinct feature is that the design unfolds dur-ing and is fundamentally shaped by the interaction of thedesigner and the immediate moment and materials.

Improvisation involves a genuine fusion of design and execu-tion, of unplanned change and order, but it was not all-perva-sive in these organizations. We noticed that a particularepisode of improvisational activity was often nested withinother, nonimprovisational activities. Teams tended to impro-vise in one domain, such as product features or teamprocess, at a given point while following plans or prior rou-tines in other areas. At SeeFoods, the team improvised anew focus group approach. The area of application was notnew, but the process was. Likewise, we saw teams usingwell-established processes but improvising a new feature orproduct-related outcome, such as the ideation sessions atSeeFoods and the scientist specials at FastTrack.

Weick (1998: 551) has expressed concern with assertionsthat organizations can develop competencies in improvisa-tion, suggesting that efforts to fuse routinization and improvi-sation may involve "lost precision" and reflect an effort to

334/ASQ, June 2001

Improvisation and Learning

REFERENCES

Adler, P. A., and P. Adter1998 "Observational techniques."

In N. K. Danzin and Y. S. Lin-coln (eds.}, Collecting andInterpreting Qualitative Mate-rials: 79-109. Thousand Oaks,CA: Sage.

Aldrich, H. E.1999 Organizations Evolving, Thou-

sand Oaks, CA; Sage.

Alinsky, S. D.T969 Rules for Radicals. New York:

Random House.

Amabile, T.1983 The Social Psychology of Cre-

ativity. New York: Springer-Verlag.

Argote, L1999 Organizational Learning: Cre-

ating, Retaining and Transfer-ring Knowledge. Boston:Kiuwer Academic.

Arthur, W. B.1987 "Competing technologies: An

overview-" In G. Dosi (ed.),Technical Change and Eco-nomic Theory: 590-607- NewYork: Columbia UniversityPress.

Baker, E. A., Jr.1995 "Stodgy too fast: Human

resource practices and thedevelopment of inertia instart-up firms." UnpublishedMaster's thesis, University ofNorth Carolina.

Baker, T , and H. E. Aldrich1994 "Friends and strangers: Early

hiring practices and idiosyn-cratic jobs." In W. Bygrave etal. (eds.}, Frontiers of Entre-preneurship Research: 75-67.Wellesley, MA: Babson Col-lege.

embrace improvisation "without giving up the prior commit-ment to stability and order." We agree that careless researchon improvisation could feed a pervasive thirst for predictabili-ty and functionalism by disguising the essentially creativeaspects of improvisation. At the same time, our field observa-tions opened the door to a vision of organizations thatembraces both their crucial protean aspects and their widelyobserved inertial qualities. Improvisational episodes werecommon but often ephemeral. From time to time, their pro-ductions or collateral insights found their way into variousparts of the organization's enduring memory; at other times,they did not. The careful delineation of improvisationalepisodes and tracing of their fates seems to us to offer apromising approach to probing longstanding puzzles in organi-zational learning and change.

Baker, T., A. S. Miner, and D. T.Eesley2001 "improvising firms: Bricolage,

account giving and improvisa-tional competencies in thefounding process." ResearchPolicy, vol- 31 (in press)-

Barley, S. R.1986 "Technology as an occasion

for structuring: Evidence fromobservations of CT scannersand the social order of radiol-ogy departments." Adminis-trative Science Ouarterty, 31:78-108.

Bastien, D. T, and T. J. Hostager1992 "Cooperation as communica-

tive accomplishment: A sym-bolic interaction analysis of animprovised jazz concert."Communication Studies, 43:92-104.

Baum, J. A. C, and J. V. Singh1994 Evolutionary Dynamics of

Organizations. New York :Oxford University Press.

Berger, P. L, and T. Luckmann1967 The Social Construction of

Reality. New York: Doubleday,

Berliner, P. F.1994 Thinking in Jazz: The Infinite

Art of Improvisation, Chicago:University of Chicago Press,

Borkin, J.1999 "Immersion/crystallization,"

(n B, F. Crabtree and W. L,Miller (eds,}. Doing QualitativeResearch, 2d ed.: 179-194,Thousand Qaks, CA; Sage.

Borko, H., and C. Livingston1989 "Cognition and improvisation;

Differences in mathematicsinstruction by expert andnovice teachers." AmericanEducational Research Journal,26: 473-498.

Brown, J., and P. Duguid2001 "Knowledge and organization:

A social-practice perspec-tive," Organization Science,12: 198-213.

Brown, S. L., and K. M. Eisen-hardt1995 "Product development; Past

research, present findings,and future directions," Acade-my of Management Review,20: 343-378.

Burgelman, R. A.1994 "Fading memories: A process

theory of strategic businessexit in dynamic environ-ments." Administrative Sci-ence Quarterly, 39: 24-56.

Burgelman, R. A., M. Maidique,and S. C. Wheelwright1996 Strategic Management of

Technology and innovation,2d ed. Chicago; Irwin.

Burns, T., and G. Stalker1961 Management of Innovation.

London: Tavistock,

Campbelf, D. T.1969 "Variation and selective reten-

tion in socio-cuitural evolu-tion." General Systems, 16;69-85.

Cheng, Y.-T., and A. H. Van de Ven1996 "Learning the innovation jour-

ney; Order out of chaos?"Organization Science, 7;593-^19,

Cook, T. D., and D. T. Campbell1979 Quasi-experimentation:

Design and Analysis Issuesfor Field Settings, Boston:Houghton Mifflin,

Crabtree, B. F., and W. L. Miller(eds.)1999 Doing Qualitative Research,

2d ed. Thousand Qaks, CA;Saga.

335/ASQ, June 2001

Cro^an, M., and M. Sorrenti1997 "Making sense of improvisa-

tion." in J. P. Walsh and A. S.Huff (eds,), Advances inStrategic Management, 14:155-180. Greenwich, CT: JAIPress.

Cunha, M. P. e, J. V. da Cunha,and K. Kamoche19M "Organizational improvisation:

What, when, how and why."Intemationa! Journal of Man-agement Reviews, 1 (3)299-341.

Cyert, R. M., and J. G. March1992 A Behavioral Theory of the

Firm, 2d ed. O>cford: Black-well.

Denzin, N. K., and Y. S. Uncoinieds.)1998 Collecting and Interpreting

Qualitative Materials. Thou-sand Oaks, CA: Sage.

Dougherty,. D.1990 "Understanding new markets

for products." Strategic Man-agementJournal, 11 (Sum-mer): 59-78.

1992 "Interpretive barriers to suc-cessful product innovation inlarge firms." Organization Sci-ence. 3: 179-202.

Eisenhardt, K. M.1989 " Making fast strategic deci-

sions in high-velocity environ-ments." Academy of Man-agement Journal, 32:543-576,

Eisenhardt, K. M., and B. N.Tabrizi1995 "Accelerating adaptive

processes: Product innovationin the global computer indus-try." Administrative ScienceOuarterly, 40: 84-110.

Embrey, D. G., M. R. Guthrie,0. R. White, and J. Dietz1996 "Clinical decision making by

experienced and inexperi-enced pediatric physical ther-apists for children withdiplegic cerebral patsy." Phys-ical Therapy, 76: 20-33.

Feldman, M.1995 Strategies for Interpreting

Oualitative Data. ThousandOaks, CA: Sage,

2000 "Ogani2ationa! routines as asource of continuouschange." Organization Sci-ence, 11:611-629.

Fetterman, D. M.1989 Ethnography Step by Step.

Newbufy Park, CA: Sage.

FoHett, M. P.1930 Creative Experience. New

York: Longmans, Green,

Fontana, A., and J. H. Fray1998 "Intervievwng: The art of sci-

ence." In N. K. Denzin andY. S, Uncoin {eds.). Collectingand Intetpreting QualitativeMaterials: 47-78. ThousandOaks, CA: Sage.

Frankel, R. M.1999 "Standards in qualitative

research." In B. F. Crabtreeand W. L. Miller (eds.). DoingQualitative Research, 2d ed.:333-346. Thousand Oaks,CA: Sage.

Garud, R,, S. Jain, and C. Phelps2000 "Unpacking Internet time

innovation." Working paper.Stern School of Business.

Gavetti, G., and D. Levinthal2000 " Looking forward and looking

backward: Cognitive andexperiential search." Adminis-trative Science Quarterly, 45:113-137.

Giddens, A.1984 The Constitution of Society.

Berkeley, CA: University ofCalifornia Press.

Glynn, M. A., T. K. Lant, and F. J.Milliken1994 "Mapping learning processes

in organizations: A multi-levelframework linking learningand organizing." In J. Meindl,J. Porac, and C. Stubbart(eds.). Advances in Manageri-al Cognition and Organization-al Information Processing, 5:43-83. Greenwich, CT: JAtPress.

Hatch, M. J.1997a "Jazzing up the theory of

organizational improvisation."In J. P Walsh and A. S. Huff{eds.), Advances in StrategicManagement, 14: 181-191.Greenwich, CT: JAt Press.

1997b"Exploring the empty spacesof organizing: How improvisa-tional jazz can help reconcep-tuatize organizational struc-ture. " Paper presented at theAnnual SCQS meeting, War-saw, Poland.

1998 "Jazz as a metaphor for orga-nizing In the 21st century."Organization Science, 9:556-557.

Hawkins, S. A., and R. Hastie1990 "Hindsight-biased judgments

of past events after the out-comes are known." Psycho-logical Bulletin, 107: 311-327.

Huber, G. P.1991 "Organizational learning: The

contributing processes andthe literatures." OrganizationScience, 2:88-115.

Hutchins, E.1991 "Organizing work by adapta-

tion." Organization Science,2: 14-39,

Irby, D. M.1992 "How attending physicians

make instructional decisionswhen conducting teachingrounds." Academic Medicine,67: 630-638,

Jelinek, M., and C. Sc^oonhoven1990 The Innovation Marathon:

Lessons from High Technolo-gy Firms. Oxford: Basil Black-well.

Kauffnnan, S. A.1996 The Origins of Order: Self-

organization and Selection inEvolution. New York: OxfordUniversity Press.

Kernfeld, B.1995 What to Listen for in Jazz.

New Haven, CT: Yale Univer-sity Press.

Leonard-Barton, D.1995 Weilsprings of Knowledge.

Boston: Harvard BusinessSchool Press.

Levi-Strauss, C.1967 The Savage Mind. Chicago:

University of Chicago Press,

Levinthal, D. A.1997 "Adaptation on rugged land-

scapes," Management Sci-ence, 43: 934-950.

Levinthal, D. A., and J. G. March1993 "The myopia of leaming."

Strategic Management Jour-nal, 14:95-112.

Levitt, B., and J. G. March1988 "Organizational learning," In

Annual Review of Sociology,14:319-340. Palo Alto, CA:Annual Reviews.

March, J. G.1991 "Exploration and exploitation

in organizationai learning."Organization Science, 2:71-78,

1998 The Pursuit of Organizationalintelligence. Maiden, MA:Blackwell.

March, J. G., and J. P. Olsen1976 Ambiguity and Choice in

Organizations. Norway: Uni-versitetsforlaget.

336/ASQ, June 2001

ImiH'ovisation and Learning

Metcalf, J.1986 "Decision making and the

Grenada rescue operation." InJ. G. March and R.Weissinger-Bayon {eds.).Ambiguity and Command:Organizational Perspectiveson Military Decision Making:ni-2°n. Marshfieid, MA: Pit-man.

Milgrom, P., and J. Roberts1990 "The economics of modern

manufacturing: Technology,strategy, and organization."American Economic Review,80:511-528.

Miner, A. S.1987 "idiosyncratic jobs in formal-

ized organizations." Adminis-trative Science Quarterly, 32:327-351.

1989 "Structural evolution throughidiosyncratic jobs: The poten-tial for unplanned learning."Organization Science. 1:195-210.

Miner, A. S., and P. Anderson1999 "Industry population-level

learning: Organizational,interorganizational and collec-tive learning processes." inA, S. Miner and P Anderson(eds.), Advances in StrategicManagement: Population-level Learning and IndustryChange, 16: 1-32. Stamford.CT JAI Press.

Miner, A. S., C. Moorman, and P.Bassoff1997 "Organizational improvisation

in new product develop-ment." Working paper, Man-agement Science Institute,Report No. 97-110.

Mintzberg, H., and A. McHugh1985 "Strategic formation in an

adhocracy." AdministrativeScience Quarterly, 30:160-197.

Moorman, C, and A. S. Miner1997 "The impact of organizational

memory on new product per-formance and creativity."Journal of MarketingResearch, 34:91-107.

1998a "The convergence of planningand execution: Improvisationin new product develop-ment." Journal of Marketing,61: 1-20.

1998b "Organizationa! improvisationand organizational memory."Academy of ManagementReview, 23: 698-723.

Nonaka, I.1990 "Redundant, overlapping

organization: A Japaneseapproach to managing theinnovation process." Califor-nia Management Review, 32(3): 27-38.

Nord, W., and S. Tucker1987 Implementing Routine and

Radical Innovations. Lexing-ton, MA: Lexington Books.

Pressing, J.1984 "Cognitive processes in

improvisation." In W. R.Crozier and A. J. Chapman(eds.). Cognitive Processes inthe Perception of Art:345-363. Amsterdam: North-Holland.

1988 "Improvisation: Methods andmodels." In J. A. Stoboda(ed.). Generative Processes inMusic: The Psychology ofPerformance, Improvisation,and Composition: 129-178,Oxford: Oxford UniversityPress.

Preston, A.1991 "improvising order." In I. L.

Mangham (ed.). OrganizationAnalysis and Development:8 V I02 . New York: Wiley.

Ragin, C. C.1987 The Comparative Method:

Moving Beyond Qualitativeand Quantitative Strategies.Berkeley. CA: University ofCalifornia Press.

Scbeliing, T.1978 Micromotives and Macrobe-

havior. New York: W. W. Nor-ton.

Schon, D. A.1983 The Reflective Practitioner:

How Professionals Think inAction. New York : BasicBooks.

Scboonhoven, C. B., K. M. Eisen-hardt, and K. Lyman1990 "Speeding products to mar-

ket: Waiting time to first prod-uct introduction in newfirms." Administrative Sci-ence Quarterly, 35: 177-207.

Spolin, V.1963 Improvisation for the Theater:

A Handbook of Teaching andDirecting Techniques.Evanston. IL: NorthwesternUniversity Press.

Stein, D. L.1989 "Preface." In D. L. Stein (ed.).

Lectures in the Sciences ofComplexity: xiii-xxiii. Red-wood, CA: Addison-Wesley.

Stoner, D. L, K. Tice, and J. E.Ashton1989 "Simple and effective cellular

approach to a job shopmachine shop." Manufactur-ing Review, 2: 119-128.

Strauss, A., and J. Corbin1998 Basics of Qualitative

Research, Thousand Qaks,CA: Sage.

Tushman, M., and P. Anderson1986 "Technological discontinuities

and organizational environ-ments." Administrative Sci-ence Quarterly, 31: 439-465.

Tushman, M., and P. Anderson(eds.)1997 Managing Strategic Innova-

tion and Change: A Collectionof Readings, New York:Qxford University Press.

Van de Ven, A., and D. Polley1992 "Learning while innovating."

Qrganization Science, 3:92-116.

Walsb, J. P.1995 "Managerial and organization-

al cognition: Notes from a tripdown memory lane." Organi-zation Science, 6: 280-321.

Weick, K. E.1993a "The collapse of sensemak-

ing in organizations: TheMann Gulch disaster." Admin-istrative Science Quarterly,38: 628-652,

1993b"Organizational redesign asimprovisation," In G. P. HuberandW. H. Glick (eds.). Orga-nizational Change andRedesign: 346-379. NewYork: Oxford University Press.

1993c"Managing as improvisation:Lessons from the world ofjazz," Working paper. Univer-sity of Michigan GraduateSchool of Business Adminis-tration,

1996 "Drop your tools: An allegoryfor organizational studies,"Administrative Science Quar-terly, 41: 301-313.

1998 "Improvisation as a mindsetfor organizational analysis."Qrganization Science, 9:543-555.

Zaitman, G., R. Duncan, and J.Holbek1973 Innovations and Organiza-

tions, New York: Wiley.

337/ASQ, June 2001