kane orgsci

8/18/2019 Kane OrgSci

1/17

OrganizationScienceVol. 18, No. 5, September–October 2007, pp. 796–812

issn1047-7039 eissn1526-5455 07 1805 0796

informs ®

doi 10.1287/orsc.1070.0286

© 2007 INFORMS

Information Technology and Organizational Learning: AnInvestigation of Exploration and Exploitation Processes

Gerald C. KaneCarroll School of Management, Boston College, 140 Commonwealth Avenue, Chestnut Hill, Massachusetts 02467,

[email protected]

Maryam AlaviGoizueta Business School, Emory University, 1300 Clifton Road, Atlanta, Georgia 30322, [email protected]

This study investigates the effects of information technology (IT) on exploration and exploitation in organizationallearning (OL). We use qualitative evidence from previously published case studies of a single organization to extendan earlier computational model of organizational learning (March 1991) by introducing IT-enabled learning mechanisms:communication technology (e-mail), knowledge repositories of best practices, and groupware. We find that each of theseIT-enabled learning mechanisms enable capabilities that have a distinct effect on the exploration and exploitation learningdynamics in the organization. We also find that this effect is dependent on organizational and environmental conditions, aswell as on the interaction effects between the various mechanisms when used in combination with one another. We explorethe implications of our results for the use of IT to support organizational learning.

Key words: knowledge management; organizational learning; exploration; exploitation; simulation; groupware; knowledgerepositories; knowledge portals; electronic communities of practice

The ability of organizations to learn and acquireknowledge has emerged as a key factor influencing orga-nizational performance and survival (Argote et al. 2003,Grant 1996). Many organizations allocate dedicatedresources to organizational learning (OL) and knowledgeacquisition. Examples include research staff positions,research and development (R&D) departments, formaltraining programs, hiring employees with specializedknowledge, and the development and use of informationtechnology (IT)-enabled learning support mechanisms.An emerging research area in the information systems(IS) field focuses on the application of advanced IT toolsto support the underlying processes of knowledge shar-ing and organizational memory (Alavi and Leidner 2001,Sambamurthy and Subramani 2005, Robey et al. 2000).

A number of factors can influence how these IT-en-abled learning mechanisms support OL. The character-istics of the mechanisms themselves, of the individualswho use those mechanisms, and of the organizationalenvironment in which those mechanisms are used caneach impact their influence. Little is understood abouthow the introduction of IT-enabled learning mechanismsinto organizations influences complex OL processes. Forinstance, various IT-enabled learning mechanisms mayeach support distinct OL processes; but, they also mayinteract with one another to create different effects thanwhen used alone. Individuals who use these IT-enabledlearning mechanisms may have different learning char-acteristics than others who use the same mechanisms,leading certain individuals to benefit more from one typeof mechanism than another type. The environmental con-text in which these tools are applied also may influ-

ence the way they can support OL in terms both of theorganizational environment in which the mechanisms areapplied and of the environment in which the organiza-tion operates.

In this manuscript, we explore the impact of IT onOL. More specifically, we use computer simulations,both individually and in combination, to investigate theimpact of three types of IT-enabled learning mecha-nisms on learning processes: communication technology(e-mail), knowledge repositories (KRPs) of best prac-tices, and groupware. We examine how different char-acteristics of the IT tools, the individuals using them,and the environment in which they are used each influ-ences their impact. We rely on qualitative evidence andpublished accounts of a single organization as the basisfor simulating these technologies and their associatedcapabilities. We then use findings of this case study toextend a previously published computational model of OL (March 1991).

Information Technology andOrganizational LearningSeveral organizational researchers have defined learningin terms of acquiring, retaining, and transferring knowl-edge at the individual and group levels (Huber 1991,Robey et al. 2000). We define OL as the dynamic processof creating new knowledge and transferring it to whereit is needed and used, resulting in the creation of newknowledge for later transfer and use. Knowledge cre-ation, transfer, and retention can be largely regarded associal processes involving communication, interaction,

796

A d d i t i o n a l i n f o r m a t i o n , i n c l u d i n g r i g h t s

a n d p e r m i s s i o n p o l i c i e s , i s a v a i l a b l e a t h t t p : / / j o u r n a l s . i n f o r m s . o r g / .


2/17

Kane and Alavi: An Investigation of Exploration and Exploitation ProcessesOrganization Science 18(5), pp. 796–812, ©2007 INFORMS 797

collaboration, and discourse among organizational mem-bers. OL is related to the concept of knowledge manage-ment (KM), which is also primarily concerned with theorganization’s ability to create and transfer knowledge.KM, however, tends to emphasize the static stocks of knowledge held by an organization and the characteris-

tics of that knowledge, rather than the dynamic processesthrough which knowledge is developed by organizations(Vera and Crossan 2002). Given their shared focus, natu-ral points of comparison and commonality exist betweenOL and KM, particularly in the types of IT tools used.Nevertheless, since we are most interested in the use of IT to support dynamic knowledge creation processes, wehave chosen to adopt the terminology of OL consistentlythroughout the paper.

We consider two forms of OL: exploration andexploitation. Exploration involves the development of new knowledge or replacing existing content within theorganization’s memory (Abernathy 1978, March 1991,Pentland 1995). Exploitation refers to incremental learn-ing focused on diffusion, refinement, and reuse of exist-ing knowledge (Larsson et al. 1998, March 1991, Smithand Zeithaml 1996). Previous literature has noted thatIS can influence both exploration and exploitation inOL (Attewell 1992, Gray 2001, Pentland 1995). Miss-ing from these investigations, however, are studies thatsystematically examine how organizations may use theirIT-enabled learning mechanisms to affect the desiredbalance between exploration and exploitation in OL.Throughout the paper, we use the term IT-enabled learn-ing mechanisms to refer to both the technologies them-selves (e.g., e-mail, listservs) and the organizationalcapabilities and structures these technologies enable(e.g., an online community).

A number of important factors must be consideredwhen assessing the influence of IT on OL. IT-enabledlearning mechanisms are not applied generically toOL—rather, they are used in distinct ways to supportlearning processes. Certain mechanisms may enable dif-ferent types of learning. For instance, KRPs may be bet-ter used for sharing structured or explicit knowledge,but communication technologies (e.g., e-mail, listservs)may be better suited for sharing unstructured or tacitknowledge (Goodman and Darr 1998). The effects of IT-enabled learning mechanisms on OL may also changeover time. People may learn to use these tools to supportOL more effectively (Carlson and Zmud 1999) or timemay render some tools less effective as the amount of knowledge available through them becomes overwhelm-ing for users (Hansen and Haas 2001). Understand-ing both the short- and long-term effects of differentIT-enabled learning mechanisms on OL is important asorganizations use these tools toward the attainment of various OL outcomes.

Most organizations, however, do not use only a singleIT-enabled learning mechanism to support OL; instead,most use a combination of mechanisms (Goodman

and Darr 1998, Vertegaal 2003). Because individualIT-enabled learning mechanisms have distinct influenceson OL processes, using multiple mechanisms together tosupport OL may result in unpredictable outcomes. Thiscombination of IT-enabled learning mechanisms may notinvolve a simple aggregation of their effects, since the

relationship between different types of OL processesare non-linear and may involve significant interactions(Katila and Ahuja 2002). Furthermore, individuals oftenuse IT-enabled learning mechanisms differently thanintended by their designers. For instance, researchershave found that individuals used groupware systems tosupport OL in significantly different ways than theirmanagers or the system designers intended (Orlikowski1996), and these variations may increase as different sys-tems are used in conjunction with one another.

Previous literature has provided mixed messages onthe combined use of different IT-enabled learning mech-anisms to support OL. The notion of balance betweenexploration and exploitation has been a consistent theme(March 1991, Teece et al. 1997, Weick 1992). It is gen-erally believed that short- and long-term performancesof organizations depend on achieving an effective bal-ance between these two processes. Other researchers,however, have suggested that blending different types of IT-enabled learning mechanisms to support OL is detri-mental, suggesting that organizations choose a singletype of IT-enabled learning mechanism (Hansen et al.1999). It is unclear how these seemingly contradictoryprescriptions might be reconciled with one another orwhich prescription is preferable. Nevertheless, under-standing whether and how to use IT-enabled learningmechanisms in conjunction with one another is a criticalissue in OL.

The effects of IT-enabled learning mechanisms onOL can also be significantly influenced by the individu-als who use them. Mechanisms might support individu-als who have different learning capabilities in differentways. This effect might be complementary, IT enhanc-ing learning capabilities that are already well developedin the individual. For instance, people who are morecreative are better able than people who are less cre-ative to leverage groupware to develop innovative ideas(Garfield et al. 2001). In contrast, this effect might becompensatory-IT augmenting underdeveloped individ-ual learning capabilities. Faced with large volumes of information, individuals with limited memory may ben-efit significantly from knowledge repository technolo-gies that enable them to compensate for this limitation.Understanding how different IT-enabled learning mech-anisms are influenced by the learning characteristics of the individuals who use them may be an important factorin understanding their effect on OL.

The influence of particular IT-enabled learning mech-anisms may also be dependent on organizational andenvironmental conditions. Organizational turnover, forinstance, has been shown to be an important condition




3/17

Kane and Alavi: An Investigation of Exploration and Exploitation Processes798 Organization Science 18(5), pp. 796–812, © 2007 INFORMS

for understanding OL, but its effect is somewhat para-doxical. Organizational turnover can result in the loss of key organizational knowledge because employees taketheir knowledge with them when they leave the organi-zation, but it can also bring critical new knowledge to theorganization through the contributions of new employ-

ees (Carley 1992, March 1991). IT-enabled mechanismsmay have similar seemingly contradictory effects in rela-tion to environmental conditions (Robey and Boudreau1999). For instance, KRPs may preserve knowledgedespite organizational turnover by retaining knowledgeheld by the individuals after they leave (Griffith 1999).By storing vast amounts of knowledge, however, repos-itories may also make the organization less sensitive toand aware of new knowledge, restricting the assimilationof new knowledge from the environment or the knowl-edge brought by new employees (Gill 1995). Under-standing how IT-enabled learning mechanisms operatedifferently in both high- and low-turnover environments

is a critical factor in determining how to use these mech-anisms to support OL.

Another condition that may influence the effective-ness of particular IT-enabled learning mechanisms isenvironmental turbulence that leads to changing knowl-edge requirements for an organization. In some situa-tions, the knowledge requirements of the organizationare relatively stable. Certain industries such as con-struction or manufacturing experience relatively minorchange in knowledge requirements over time. In otherindustries, the knowledge requirements are highly tur-bulent. The biotechnology industry, for instance, expe-riences relatively significant changes in both the core

knowledge and the means of obtaining that knowl-edge (Enriquez et al. 2002). Organizations in turbu-lent environments need to accommodate radical changesin knowledge requirements. Certain IT-enabled learningmechanisms may function better under different levelsof environmental turbulence. Understanding how dif-ferent IT-enabled learning mechanisms function underboth internal and external environmental conditions suchas organizational turnover and environmental turbulencemay be critical for understanding their influence on OL.

Research Method and SettingTo explore the effects of IT on exploration and exploita-tion, we replicate and extend a computational model of OL developed by March (1991). Building on existingcomputational models, rather than developing new onesfrom scratch, is an effective method for validating exist-ing work, developing a cumulative research tradition,and enabling deeper exploration of foundational ideasthan would be possible through the continual creation of new models (Prietula and Watson 2000). A significantamount of research has extended the theoretical compo-nents of March’s exploration and exploitation framework (Benner 2002, Lee and Lee 2003), but surprisingly lit-

tle work has extended the original model on which thetheory of exploration and exploitation was based.

Computational modeling is an established but oftenoverlooked organizational research method in whichprobabilistic parameters of a theoretical model are builtinto a computer software program and executed, followed

by analysis of the program’s output (Carley 1995). Thesemodels represent theoretical abstractions of real-worldorganizations and focus on simulating general organi-zational principles and processes. Comparing and con-trasting ideal type models yield results that are some-what limited in their realism. What is lost in the level of detail, however, is gained in the degree of control overthe research environment. Researchers can model ele-ments of the research environment that are often difficultto observe, such as dynamic and nonlinear processes overtime. Within a model, the researcher can vary the param-eters of the simulation to assess the effects of variousmanipulations on specific processes and their outcomes.

Because of the control the researcher enjoys over theresearch environment, computer simulations have oftenbeen called “virtual experiments” (Prietula et al. 1998).

Because computational modeling is a theoretical ab-straction and simplification of reality, the researcher canuse empirical observation to inform the developmentand implementation of model parameters and their val-ues. Recent research has combined qualitative case studywith computational modeling to develop an effectiveresearch method that is likely greater than the sum of itsparts (Rudolph and Repenning 2002, Sastry 1997). Qual-itative evidence provides rich insight into organizationalprocesses, and this insight can be effectively used to

develop the computational model. Computational mod-eling formalizes the observations made as a result of thequalitative analysis, extending and manipulating theseresults in ways that would be difficult to obtain throughqualitative observation alone.

To extend March’s (1991) model to account for theintroduction of IT into OL, we rely on a series of pub-lished case studies of a single organization (Alavi et al.2005, Gongla and Rizzuto 2001, Huang 1998, Mack et al. 2001). The company studied will be known in thispaper by the pseudonym Company Z. Company Z reliesheavily on, and is regarded as a worldwide leader in,applying IT tools inside its organization. In addition to

using published accounts, we were granted access to theprimary qualitative interview data collected by some of these researchers (Alavi et al. 2005), providing us withrich insight regarding the role of IT tools in learningprocesses for this organization.1 The interview data werecoded by both of the researchers independently, and theresults were compared and reconciled to provide greaterreliability. The results were then placed in a content-analytic summary to facilitate analysis and provide adata structure on which to base our extensions to themodel (Miles and Huberman 1994).




4/17


Company Z uses three types of IT to support OL,which are roughly equivalent to general classes of IT-enabled learning mechanisms identified elsewhere inthe IS literature. First, Company Z has developed arobust system for knowledge storage and retrieval usingdisparate knowledge repositories and portals (KRP)which provide efficient access to the contents of theserepositories. Knowledge repositories have been regardedby IS researchers as the cornerstone of OL initiatives(Argyres 1999, El Sawy and Bowles 1997). Second,Company Z uses groupware (in form of a combina-tion of communication technology and KRPs) to cre-ate a secure learning environment, known as virtualteam rooms (TRs), in which project teams can shareand discuss engagement-specific knowledge with oneanother. IS researchers have investigated the role of sim-ilar types of groupware to support team-level learning inthis way (Dennis et al. 1988, Orlikowski 1996). Third,Company Z relies heavily on communication technolo-gies (e-mail and instant messaging) to connect employ-ees who share common interests and expertise fromacross the globe into electronic communities of practice(ECOPs). These ECOPs have drawn considerable inter-est from researchers in recent years (Ahuja and Carley1999, Wasko and Faraj 2005). Further details regardinghow Company Z uses these IT-enabled learning mecha-nisms to support OL can be found in Table 1.

Replicating March’s ModelMarch (1991) constructed a computational model of an organization to explore how different factors influ-enced the nature and effectiveness of learning. March’s

model of OL is parsimonious, comprising three pri-mary components: (1) an external reality (reflectingwhat the knowledge or beliefs of the organizationshould be), (2) an organizational code representing theorganization’s perceived beliefs about that reality (i.e.,the organization’s approximation of it), and (3) indi-vidual knowledge representing individual beliefs aboutthat reality (i.e., an individual’s perception of it). SeeAppendix A for a detailed description of March’s origi-nal model.

March (1991) observed the tendency for the knowl-edge levels of the code and of individuals to converge assuccessive iterations are performed and a stable knowl-

edge equilibrium is achieved. (However, the equilibriumvalue may actually differ from reality.) March found thathigher individual learning rates and higher learning ratesby the code resulted in quicker convergence of knowl-edge levels to equilibrium (exploitation). March alsofound that slower individual learning rates accounted formany of the higher equilibrium levels, especially whencoupled with fast code learning. This result occurredbecause slow individual learning extended heterogene-ity of the individuals’ knowledge, therefore disallow-ing premature convergence of the organizational codeto lower knowledge equilibria (exploration). Thus, the

rapid diffusion of knowledge may not necessarily bea desirable organizational characteristic if the level of knowledge matters more than quicker resolutions of individual heterogeneity.

March (1991) also extended this basic model toaccount for the effects of organizational turnover and

environmental turbulence. To simulate turnover, Marchallowed the possibility for any given individual’s knowl-edge to revert to an initial random state according to aset probability, representing the replacement of one indi-vidual with another. To simulate environmental turbu-lence, March introduced the possibility of the knowledgedimensions of reality to change according to a differentparameter. March found that organizational turnover andenvironmental turbulence each had a detrimental effecton average knowledge levels in the population, althoughin certain situations these environmental conditions hada positive effect on the code knowledge by introducingan additional exploratory effect into the learning pro-

cesses. We first replicated these models and validatedour efforts by comparing these results with March’s.

Extending the ModelUsing Company Z as a guide, we modified March’s(1991) model in three primary ways to account forthe introduction of IT in OL. First, we made baselinemodifications to his original model—modeling learn-ing between individuals and project teams—to moreclosely replicate the organizational conditions repre-sented by Company Z and most contemporary organiza-tions. Second, we modeled the three types of IT-enabledlearning mechanisms used by Company Z to support

OL—KRP, TRs, and ECOP. KRP uses KRPs to store andmake valuable knowledge available across the organiza-tion, TR uses a combination of IT tools to support OLat a local level, and ECOP use communication technolo-gies to transfer knowledge between individuals. Third,we modeled a number of different possible configura-tions of these tools. These different configurations alsoenabled us to isolate the particular features of each tool,helping us to understand how these tools are used inconjunction with one another and permitting a greaterdegree of generalization of our findings beyond Com-pany Z. Table 2 illustrates and describes our extensionsin relation to March’s (1991) original model.

Baseline Extensions. We first modified March’s (1991)model in two key ways—modeling learning between indi-viduals and modeling project teams in the population—tofacilitate our later introduction of IT-enabled learningmechanisms to support OL.

First, we extended March’s (1991) model to allowindividuals to learn from one another. Although Marchdid not permit learning between individuals in his orig-inal model, he explicitly addressed this characteristic of his model and introduced the possibility for individualsto interact with one another (p. 75). A long and robust




5/17


Table 1 Summary of IT-Enabled Learning Mechanisms Used by Company Z and Implemented in Simulation

IT-enabledlearning General observationsmechanism from case study Examples from case study evidence Model specifications

KRP 1. Employees contribut e knowledgeto repositories maintained by

project teams.2. Contributed knowledge is vettedby subject matter experts withinthe team.

3. Access knowledge through portalthat aggregates and organizesthe knowledge.

4. Employees noted that KRPs oftenbecame out of date, since noformal procedures to removeknowledge once contributed.

1. “The repositories are independentlyorganized, each one of them. There are

some standard things that they have to havein there, but then each team can apply itsown categorization to it.”

2. “An important part of this is contentevaluation. We do not put anything on therepository that is not vetted by the subjectmatter experts. All content in the repositoryhas been reviewed for quality.”

3. “It is basically a huge database that tiestogether all these other databases. You canget to it through the Intranet and you cango on and do a search like a megasearch,where you can search all the differentdatabases.”

4. “Now with the knowledge in the repositorieswe have a whole lot of noise and nobodyhas the time to search it all. It takes a lot of

effort to get what you want at times.”

1. Each project team (t)maintains independent

repository.2. Individual knowledge iscompared to those withabove-average knowledgelevels in team. Only tuplesmatching the majority arecontributed to KRP.

3. Knowledge portal synthesizesknowledge in each repositoryinto single vector. Theindividual searches portalsto assemble KRP-group ofvectors from which individuallearns.

4. Once knowledge contributedto a repository, remains inrepository for duration of

simulation.

TRs 1. TR used primarily to supportlearning within a project team.Password protected to excludenonteam members.

1. “The process to set up a TR is fairly easy.It’s a really helpful way to coordinateproject-specific knowledge because youdon’t have to worry about sanitizingknowledge.”“Because we know the room is secure, wecan challenge each other in there and geta sense of what is really contributed.”

1. TR-group assembled onlyfrom members of projectteam (t.

ECOP 1. ECOP assembled on the basisof shared interests amongemployees.

2. Employees learn from asubnetwork of individualswithin the ECOP.

3. ECOP reported as relatively leanchannel compared to other tools.

1. “The membership of the community isvoluntary. Members have a ‘day job’. itis the passion for the knowledge domainthat motivates people to participate [inthese communities]. Members are willingto spend personal time on that.”

2. “People typically rely on people they havea previous relationship with for example,if they’ve worked together before. Theysimply trust the responses more if theyknow the people. The higher [the] levelof trust, also the more quickly peoplewill respond.”

3. “In the tools used for ECOP, knowledgesharing can be limited because there’snot very much you can actually say on aparticular subject. It’s small in that respect.If you want more on a particular subject,then you would rely on other tools like the[knowledge repository].”

1. Individuals randomly assignedto one of four interest groupsat beginning of simulation.

2. Individuals assemblesubnetwork of otheremployees based on whetheror not individuals sharedinterest parameter.

3. Exchange only limited portion(R ct = 033020) of knowledgevector via ECOP.

Notes . We model three approaches at Company Z. (1) Knowledge repositories (KRPs), which are our more formal, explicit tools for

sharing knowledge (2) TRs, which are a very unstructured environment in which team members share with one another, and (3) electronic

communities of practice (ECOP), which involve people from many different teams all living under one roof.

research stream examines how individuals learn directlyfrom one another in interpersonal settings (Borgatti andFoster 2003), and OL literature has used similar sorts of models that embed learning between individuals (Changand Harrington 2005). Since learning is operationalizedwithin March’s simulation as a function of the individ-ual and not of the environment or of the organization(pp. 76–77), introducing learning between individuals isconsistent with the theoretical basis of learning found inthe model.

When seeking to learn from other individuals, peo-ple tend to look for knowledge from those they per-ceive as having superior knowledge (Perry-Smith andShalley 2003). For consistency, we sought to implementindividual learning in ways as similar as possible tothe original model. The algorithms for individual learn-ing are replicated from March’s (1991) original model,in which the code assembles knowledgeable individu-als from which to learn. In our case, though, we sub-stitute the individual learning parameter for the OL




6/17


Table 2 Illustration and Description of Extended Model

Team

number

Interest

number

Knowledgedimension # 1 2 3 4 5 6 7 … 25 26 27 28 29 30 KL

Reality 1 1 1 1 1 …

1 1…

1 …

1 …

1 …

…

…

…

…

1 1 1 1

Code 1 1 0.47

Individual 1 1 1 1 1 1 1 1 1.0

Individual 2 1 2 0 1 1 1 1 0.97

Individual 3 1 3 0 1 1 1 1 0.93

1 0 –1 … 1 1

1 1 0 … 0 –1

(KRP for Group 1)

… …

Individual n 5 2 –1 –1 –1 –1 –10

–1 –1 0

–1 –1 0

–1 0

–1 –1 0

–1 –1 0

–1 –1

–1

–1

–1

–1

–1

–10

0

0

0

0 0 11 1 1 1 0.50 –1 0 –1 … 0 1

(KRP for Group 5 )

… …

Individual 98 10 2 0 1 1 1 0.07

Individual 99 10 3 0 1 0 1 0.03

Individual 100 10 4 0 1 0 0

1 0 –1 … 1 1

1 1 0 … 0 –1

(KRP for Group 10 )

TR Learn only from

individuals with sameteam number

KRP Team members contribute to KRP

at team level. Knowledge synthesized

across repositories through portal fordissemination across population and

stored for use in later rounds.

Φ -group All eligible sources with

higher KL (here, Φ-group forcode = individuals with KL > 0.47)

KL Percent of

individual/code knowledge

dimensions match reality

–1

1 –1

1 –1

1 –1

0 –1

–1

1 1 –1 –1 –1 –1 –1

1 1 –1 –1 –1

–1 –1

–1 –1

1 1 –1

1 1 –1

–1 –1

–1 –1

ECOP Learn from network

of others based on sameinterest number.

Note . KL = knowledge level.

parameter. Thus, when a given individual seeks to learnfrom others in the population, that individual assemblesa group of knowledgeable individuals (-group) fromwhich to learn (Fang et al. 2006). When assemblingthe -group, an individual cannot know which knowl-edge that another individual or source possesses maybe valuable (i.e., the same as reality), but the individ-

ual simply knows and chooses to learn from particularindividuals or nonhuman repositories that are generallymore knowledgeable. Parameter specifications were cho-sen for the three IT-enabled learning mechanisms so thatthe number of available sources from which an indi-vidual could assemble the -group was approximatelyequal for each. Controlling for the number of knowl-edge sources allowed us to focus on how the particularfeatures of each mechanism influenced outcomes.

Second, employees in most large organizations arenot uniformly distributed, but rather are organized intosome form of organizational structure. Company Z isno exception. These organizational structures can influ-ence the nature and effectiveness of OL and also howIT is used to enhance it (Carley 1992, Siggelkow andRivkin 2005). Given the team-based structure preferredby Company Z, we organized the population into a sim-ple team-based structure consisting of an equal numberof identically sized and configured teams. This relativelysimple structural form was chosen to maintain simplicityin the model (Taber and Timpone 1996), while permit-ting us to capture ways in which the use of IT-enabledlearning mechanisms may be influenced by these orga-nizational structures. At the beginning of the simulation,

we partitioned the population into 10 distinct projectteams (t of 10 individuals each. Each individual wasplaced in one project team and remained a member of that team throughout the duration of the simulation. Thenumber of teams into which the population was parti-tioned was based on trial simulations conducted duringthe initial construction of the model. We experimented

with various team sizes (n= 25102550) and foundthat all but the largest and smallest teams performed rel-atively consistently with one another. Thus, we chose amiddle value (n= 10) on which to base the remainderof our extensions.

Knowledge Repositories and Portals. A combinationof technology and processes governed how KRPs wereused by Company Z to support OL. Individuals con-tributed to KRPs at the team level. Through the processof repackaging knowledge to be made more relevant toothers in the organization, subject matter experts vet-ted the repackaged knowledge to ensure its quality and

applicability. The knowledge portal then provided pow-erful tools to search and access the knowledge held inthe disparate team-level repositories across the organi-zation. Once knowledge was added to the repository,no formal procedures were established to remove out-dated knowledge. Although recognized by users as hav-ing notable limitations, this practice characterized theway Company Z used KRP.

These features were implemented into our simulation.The process of using the KRP for knowledge sharinginvolved three stages: (1) individuals contributing KRPs,




7/17


(2) synthesizing knowledge, and (3) portals disseminat-ing knowledge. First, individuals contribute their exist-ing knowledge to the repository maintained by theirteam t. This knowledge is vetted by team members bycomparing an individual’s knowledge to that of the mostknowledgeable members of the team (those with knowl-

edge levels above the group average) that representedthe subject matter experts. Only those knowledge dimen-sions that matched the majority values of this group of knowledgeable individuals were included in the KRP.The KRP retains all knowledge contributed to it for theremainder of the simulation. Second, each repositorymaintained by a project team t synthesizes the knowl-edge it contains to produce a single knowledge vector.At the beginning of each round, the repository aggre-gates all of the knowledge contributed to it and selectsa majority value for each knowledge dimension. Thesevalues become the knowledge vector for the repository.This knowledge vector is compared to reality, and the

knowledge level for the repository is determined. Third,the knowledge portal disseminates the knowledge byallowing individuals to search knowledge held in each of the independent repositories maintained by teams acrossthe organization. When individuals access knowledgethrough the knowledge portal, they search the reposito-ries by assembling a KRP-group of those repositoriesthat have knowledge levels that are higher than the indi-vidual’s level. The individual then adopts the knowledgecontained within these repositories according to a simi-lar selection algorithm implemented for each of the otherforms of learning used in the simulation.

Therefore, when learning via KRP, individuals con-

tribute knowledge to the KRP by (1) identifying whichmembers of the team have a higher knowledge level thanthe individual (subject matter experts) and (2) contribut-ing to the team-level repository only those dimensionsof the knowledge vector that match the majority value of these subject matter experts (vetting knowledge). Eachteam-level repository synthesizes all of the knowledgecontained within it, contributed in this and previousrounds, into a single vector. Individuals adopt knowledgefrom the KRP by (3) assessing which of the knowledgevectors in the team-level repositories has a higher knowl-edge level than the individual (KRP-group), and (4), foreach dimension of the knowledge vector, adopting the

majority value of the KRP-group of repositories accord-ing to a given probability set by the individual learningparameter (p1).

Team Rooms. Company Z used both knowledgerepository and communication technologies to supportlearning within a given project team, restricting learningin TR to occur only within the project team. In our sim-ulation, when individuals choose to learn via TR, theyassemble their superior group (TR-group) and learnonly from the individuals who have been assigned to thesame project team (t). Learning in TR is thus confined to

a local scope, compared to the other tools, which permita wider exchange of knowledge.

When learning via TR, individuals (1) assess whichof all individuals in the project team t have a higherknowledge level than the individual (TR-group), and(2) for each dimension of the knowledge vector, adopt

the majority value of the TR-group according a givenprobability set by the individual learning parameter p1.

Electronic Communities of Practice. The case stud-ies of Company Z revealed that ECOP were organizedalong particular interest groups and involved individu-als learning from a subnetwork of employees within theECOP. In terms of our simulation parameters, individualswere randomly assigned to have one of four general inter-ests that represent those topics around which the ECOPare organized. Individuals assemble a subnetwork of oth-ers from which to learn within the ECOP according toa set probability (pi = 025. Although individuals can-

not belong to more than one ECOP in our simulation,we did introduce the possibility for individuals to learnfrom other individuals in the population according toanother probability (pni = 002 to represent the perme-able boundaries between the ECOP that were accountedfor in the case studies. Parameter values were variedsomewhat and results were robust with respect to thesevariations (pi = 035025015pni = 001002003).We also imposed alternative network structures on theseECOP. These had no significant influence on the resultsof our simulation, so we chose a simple random network structure.

ECOP members also claimed that communication

technologies used as the basis for ECOP are regarded asa relatively lean learning mechanism when compared toother mechanisms. We sought to model this character-istic of communication technologies by specifying thatindividuals could only exchange a portion of their knowl-edge dimensions with others when using communica-tion technologies for learning. If the interest parameterdefines who an individual can learn from using ECOP,the mechanism richness parameter governs what can belearned via these channels. In a given period, the individ-ual randomly selects a portion of the code to learn fromthe ECOP-group. Since the reputation and credentials of individuals in a network of practice are not limited by

the communication medium, this channel leanness didnot affect the formation of the ECOP-group. Individualknowledge levels are still based on the entire individualknowledge, even though only a portion is exchanged.

It is interesting to note that respondents perceived thetools used to support ECOP as a lean mechanism but didnot so perceive the others, such as KRP and TR. It maybe that ECOP enables types of knowledge sharing inpractice other than the other IT-enabled learning mecha-nisms. This difference could affect the amount and typeof knowledge exchanged through a given mechanism.




8/17


TR were used to augment face-to-face relationships,and this inherent multiplexity may have provided sig-nificant contextual information when used in practice.The KRP also contained a high volume and variety of knowledge (including multimedia), perhaps also lend-ing itself to perceptions of greater richness. TR and

KRP are able to provide a considerable amount of back-ground information in ways that ECOP does not. Fur-thermore, knowledge in the KRP was repackaged forconsumption by a more general audience (e.g., project-specific information was removed, keywords added, keypoints synthesized, etc.) and thus the KRP was explic-itly and intentionally made more lean. Employees mayperceive this synthesized knowledge differently. ECOP,in contrast, enabled people with common interests andknowledge to exchange knowledge regarding relativelyspecific and deep issues, without the benefit of signif-icant contextual or shared background knowledge. Themain differences may be that the former two spaces

are used to share codified materials while the ECOPis used to ask or discuss highly tacit questions andissues. Each technology provides affordance to differ-ent types of knowledge sharing. This difference wouldaccount for the variance in perceptions of richness foreach mechanism.

In sum, when learning via ECOP individuals (1) iden-tify individuals that belong to their ECOP subnetwork,(2) assess which of these individuals have a higherknowledge level than the individual (ECOP-group),(3) select a portion (Rct of the knowledge vector tolearn from the -group, and (4) for each dimension of the knowledge vector, adopt the majority value of the

ECOP-group according a given probability establishedby the individual learning parameter p1.

Combining IT-Enabled Learning Mechanisms for OL.

Company Z uses all three IT-enabled learning mecha-nisms to support OL. Once we modeled the three mech-anisms in the simulated organization, we varied thedegree to which the organization used each for learning.This variation enabled us to isolate the distinct featuresof each, as well as examine how they function in com-bination with one another.

We implemented these variations through a series of probabilities to represent the likelihood that an individ-ual will choose one of the three learning forms in a givenround. At the beginning of a given round of knowledgeexchange, each individual chooses to use one of thesemechanisms, according to probabilities established at thebeginning of the simulation. The choice of an IT-enabledlearning mechanism in one round does not influence thechoice of other mechanisms in further rounds. At thebeginning of the simulation, one mechanism was identi-fied as the dominant tool and was selected by membersof the population according to a particular probability(pd. The other mechanisms were identified as sec-ondary tools, and were selected according to a separate

probability (ps. Several different configurations of pri-mary and secondary forms were modeled. In the pureconfiguration (pd = 1ps = 0), the organization reliedexclusively on one form of learning. In the augmented configurations (pd = 08ps = 01pd = 06ps = 02),the organization preferred a particular technology but

augmented it with each of the other forms of learning.In the blended configuration (pd = 033ps = 033, theorganization used each of the forms equally. These con-figurations were modeled to test both the main and theinteraction effects for each form of learning.

Evaluating Computational Models. Several criteriahave been forwarded to evaluate the validity and con-tribution of computational models (Taber and Timpone1996, Burton and Obel 1995). First, choices regardingmodel parameters should be grounded in existing the-ory or empirical observation, or both, to ensure that themodel captures and simulates the critical features of real-ity in appropriate ways. Table 1 shows our model param-eters and the empirical observations on which they arebased. Second, parameters must be subjected to sensi-tivity analyses in which they are varied within a givenrange to ensure that the results of the model do notstem from arbitrary choices by the researcher duringmodel construction. Table 3 details our parameter valuesand the sensitivity analyses conducted. Third, compu-tational models should be made as simple as possi-ble to isolate the most important theoretical featuresbeing analyzed and to minimize extraneous influence.(Appendix B contains a flowchart of our programminglogic.) Fourth, results of the model should have face

validity, but should also provide new and interestinginsights not immediately obvious during model construc-tion. These implications are addressed in our results sec-tion and our discussion section.

ResultsThe simulation was executed 30 times for each set of parameter specifications. Each of the conditions was sys-tematically analyzed and evaluated. Once values andparameters included for sensitivity analysis were deter-mined to have no significant effect on model outcomes,they were discarded. On completing specification anal-ysis, we focused on a few primary areas of interest

for detailed analysis. We focused most intently on themost interesting, surprising, and important outcomes of our model: exploration and exploitation characteristicsof specific IT-enabled learning mechanisms individuallyand in combination, the influence of individual learn-ing rates on particular mechanisms, and the influenceof the mechanisms under different environmental condi-tions. Following March (1991), we examined the effectsof parameter variation on average population knowledgelevel (K pop, code knowledge level (K code, and knowl-edge variance in the population (K var. Also consistent




9/17


Table 3 Parameter Values and Sensitivity Analysis

Additional values

Values used in included for Values reported

Parameter final model sensitivity analysis by March (1991)

Dominant tool TR, ECOP, KRP

Tool configuration pd /ps1/ps2 100/0/0, 80/10/10, 60/20/20, 80/0/20, 80/20/0

33/33/33Individual learning rate p1 0.1, 0.5, 0.9 0.3, 0.7 0.1–0.9OL rate p2 1 0.1, 0.3, 0.5, 0.7, 0.9 0.1–0.9

Turnover p3 0, 0.1, 0.3 0.2, 0.5, 0.9 0.1, 0.9

Turbulence p4 0, 0.01, 0.03 0.01, 0.02, 0.04, 0.05, 0.1 002

Mechanism richness R ct 0.2, 0.33 0.1, 0.5, 1Number of teams t 10 2, 5, 20, 50

ECOP structure Random network Core-periphery structure

Probability of including those 025 0.15, 0.35

with shared interest inECOP subnetwork p i

Probability of including those 002 0.01, 0.03

not with shared interest in

ECOP subnetwork pni

with March, we look at these outcomes in terms of bothshort-term dynamics (Rounds 1–30) and long-term equi-librium (Round 80).

The value of computer simulation as a researchmethodology rests less on the actual results obtainedand more on the process of understanding and explain-ing why the model operates in the way it does. There-fore, in addition to presenting the results, we also seek to understand why our model behaves as it does. If anexamination of our simulation results can uncover thecausal mechanisms behind them, the simulation providesresults that can both inform real-world situations andgeneralize these findings into less-studied domains of IT

and OL. Exploration and Exploitation. The first observation

from our simulation is that, when examined alone, differ-ent IT-enabled learning mechanisms appear to have dif-ferent effects on exploration and exploitation processesin organizations. Figure 1 demonstrates the effects of IT-enabled learning mechanisms on the knowledge levelof the average population. Both KRP and TR yield sim-ilar learning patterns: very rapid short-term performancebenefits that tend to plateau within the first few roundsof the simulation. March (1991) labeled these patternsexploitation, and our simulation suggests that these toolstend to have an exploitatory effect on OL. Conversely,

knowledge levels under ECOP tend to increase moreslowly but do not plateau in the same way as the otherforms of learning, surpassing the overall knowledge levelof the other forms by about Round 15, then continuingto improve. March labeled this pattern exploration, andour simulation suggests that ECOP has an exploratoryeffect on OL.

March (1991) argued that these exploration andexploitation effects are driven by the heterogeneity of knowledge in the population. Figure 2, which demon-strates the effect of each mechanism on knowledge

variance within the population, supports this interpre-tation. KRP results in rapid reduction in knowledgevariance within the first two rounds, virtually elimi-nating the knowledge variance in the population byabout Round 5. ECOP initially introduces an increasein knowledge variance, followed by a gradual reduc-tion in variance in successive rounds. Although thevariance appears to be continually and steadily decreas-ing with the ECOP, an examination of knowledge vari-ance at Round 80 (not shown) demonstrates that ECOPwill never entirely eliminate knowledge variance withinthe population (K var = 002). TR seem to maintain afairly consistent degree of variance within the popula-

tion across all rounds. Further analysis demonstrates thatthis variance exists across rather than within the projectteams. Requisite knowledge variance exists within thepopulation, but the organization lacks the means to getthe knowledge to the individuals who can benefit fromit. Thus, consistent with March, exploitation tools arecharacterized by a rapid homogenization of knowledge,whereas exploration tools tend to preserve the hetero-geneity of knowledge longer in the process.

Combining Mechanisms. Figure 1 also demonstratesthe results of combining IT-enabled learning mech-anisms on average population knowledge levels. TRdemonstrate the most substantial effect of blending

mechanisms. When other tools are combined with TR,this configuration avoids the knowledge plateau charac-teristic of exploitation, continuing to improve averagepopulation knowledge levels into later rounds and tooutperform predominantly exploration tools (ECOP-dominant) in the long run. We do not see the samemarked improvements, however, when other tools areblended with KRP. KRP appears to perform largely thesame regardless of the particular blend of IT-enabledlearning mechanisms. All the simulations in whichKRP was dominant (pd = 10806) performed nearly




10/17


Figure 1 The Effect of IT-Enabled Learning Mechanisms on Knowledge Levels K pop , Round 1–30, p1 = 05, R ct = 033)

Pure config

(100/0/0)

Round: Round: Round: Round:5

1015

2025

30

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

K p o p

Augmented config

(80/10/10)

510

1520

2530

Augmented config

(60/20/20)

510

1520

2530

Blended config

(33/33/33)

510

1520

2530

TR

ECOP

KRP

Note. Config= configuration.

equally with one another, except for the blendedconfiguration in which all three tools were used equally.ECOP behaves still differently when other mechanismsare added. In these cases, the blending of mechanismstends to have a detrimental effect on overall knowledgelevels. This effect is only marginal in relation to aver-age population knowledge level (Figure 1), but becomes

Figure 2 The Effect of IT-Enabled Learning Mechanisms on Knowledge Variance (K var

, Round1–30, p1

= 05R ct

= 033)

Config 1

Round: Round: Round: Round:

5

10

15

20

25

30

0.00

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.10

K v a r

Config 2

5

10

15

20

25

30

Config 3

5

10

15

20

25

30

Config 4

5

10

15

20

25

30

TR

ECOP

KRP

–0.01

Note. Config= configuration.

more dramatic in relation to levels of knowledge heldby the organizational code (Figure 3).

One key to understanding how blending IT-enabledlearning mechanisms promotes or reduces knowledgeheterogeneity can be explained by the knowledge trans-formation cycle identified by Carlile and Rentibish(2003). This cycle involves three stages: individuals




11/17


Figure 3 The Effect of Combining IT-Enabled Learning

Mechanisms on Knowledge Levels (K pop , Round 80,

p1 = 09, R ct = 02)

Pure (100/0/0)

Augmented 1 (80/10/10)

Augmented 2 (60/20/20)

Blended (33/33/33)

0.76

0.78

0.80

0.82

0.84

0.86

0.88

0.90

0.92

0.94

0.96

0.98

K c o d e

TR

ECOP

KRP

(1) accessing knowledge, (2) transforming that knowl-edge according to their own experience (and their expe-rience is in turn transformed by that knowledge), andthen (3) contributing the transformed knowledge forlater use or for use by others in the organization. Ourresults support their framework, but also extend it bysuggesting differences between closed and open knowl-edge transformation cycles. In a closed transformationcycle, all individuals have access to the same knowl-edge, which then transforms and is transformed by thesame population, and finally, is contributed back tocommon IT-enabled learning mechanisms for access bythe entire population in the future. As this knowledge

transformation cycle repeats itself in successive rounds,knowledge converges quickly and exploitation occurs. Inan open transformation cycle, individuals have differ-ent sources of and recipients for knowledge. Individualsdraw knowledge from heterogeneous sources, combineit with their own knowledge in personalized ways, andthen transfer that knowledge to a heterogeneous set of recipients, who will repeat the cycle in later rounds. Asindividuals continue to draw on and contribute to diverseknowledge sources, knowledge converges more slowly,resulting in exploration.

This difference between the closed and open knowl-edge transformation cycle is clearest in our simulationwhen we examine the effects of combining IT-enabledlearning mechanisms with one another. We observe astrong positive effect of adding other mechanisms toTR, but we observe little if any benefit to addingother tools to KRP. This difference can be traced tothe type of knowledge these additional tools bring tothe exploitation process. In the case of TR, additionalIT-enabled learning mechanisms bring new knowledgeto the exploitation process. As the TR continues torapidly synthesize and disseminate knowledge, addi-tional tools infuse this process with knowledge from

other parts of the organization. The closed knowledgetransformation cycle, in which the source and recipi-ent of knowledge are the same, is broken by addinga new knowledge source independent from any learn-ing that occurs within the team. The resulting learn-ing curve begins to demonstrate some characteristics

of exploration—slightly underperforming other exploita-tion processes (KRP) in the short term and then surpass-ing its knowledge as the KRP plateaus (Figure 1).

Adding other IT-enabled learning mechanisms to theKRP has little, if any, effect on the exploitation pro-cess. In this case, the additional tools do not add newknowledge to the closed knowledge transformation cycleand offer little to no additional benefit to the OL pro-cesses, since the entire population (or a subset of thatpopulation) remains both the source and recipient in theknowledge transformation cycle. Additional IT mecha-nisms circulate existing knowledge through the popu-

lation differently, but this process adds little value asthe KRP continues to homogenize the knowledge acrossthe organization. Figure 2 supports this interpretation,since the population knowledge variance follows nearlythe same pattern and rate of reduction under pure KRPconditions as when KRP is blended with other tools.Therefore, adding more IT-enabled learning mechanismsin this circumstance does not benefit the organization.

Concurrent use of ECOP with other IT-enabled learn-ing mechanisms demonstrates still different effects fromeither condition observed in the exploitation tools. Theoverall results of ECOP degrade when other tools areadded to the process (Figure 3). The knowledge hetero-

geneity cultivated by the ECOP is easily affected by theintroduction of exploitation tools that are used to reducethat variance. As knowledge exploitation tools are addedto the learning process, they provide a common source of homogenous knowledge that erodes the knowledge vari-ance on which the exploration tools rely for later gains.March (1991) and other researchers (Benner 2002) havepreviously noted this tendency for exploitation to domi-nate and crowd out exploration. It is interesting to notethat the open knowledge transformation cycle in whichindividuals learn differently from one another across thepopulation results in similar outcomes, as when March

introduced heterogeneous learning rates into the popula-tion of his simulation (pp. 76–77). We are able to recreatehis results by modeling IT-enabled structures to introducethese heterogeneous learning rates, rather than relying onan inherent feature of the population of employees. Sincewe controlled for and conducted sensitivity analyses onthe size of the network in relation to specific IT tools, ourresults support this interpretation. ECOP provides accessto different sources of knowledge, not simply more of them, a finding consistent with previous studies of infor-mation networks (Constant et al. 1996).




12/17


Figure 4 The Effect of IT-Enabled Learning Mechanisms

Under Different Individual Learning Rates

(K pop , Round 80, R ct = 02)

0.1 0.5 0.9

p 1 (Individual learning rates)

0.60

0.65

0.70

0.75

0.80

0.85

0.90

0.95

K p o p

TR

ECOP

KRP

Individual Learning Rates. Figure 4 compares the

effect of individual learning rates on the performance of IT. The learning rate influences the performance of theIT-enabled learning mechanisms somewhat differently.ECOP is significantly influenced by variation in the indi-vidual learning rate, yielding significantly higher resultsunder high learning rates p1 = 09 than under low ratesp1 = 01. In contrast, KRP and TR appear less influ-enced by individual learning rates, performing relativelysimilarly across all learning rates. Mechanisms that pro-mote exploitation tend to be robust in light of changes toindividual learning values in the population. Using thesemechanisms, high learning rates demonstrate roughly thesame outcomes as low learning rates (Figure 4).

March (1991) found that high learning rates werethe key sources of exploitation in his original model.Using ECOP, high learning rates (p1 = 09 performconsistently and substantially better than low learningrates (p1 = 01. In our simulation, high learningrates (exploitation) combined with IT-enabled learn-ing mechanisms that promote knowledge heterogeneity(exploration) lead to a blend of forces that results inperformance superior to other mechanisms under simi-lar conditions. In contrast, when ECOP operates underconditions of slow learning, the exploratory forces of slow learning redouble the exploratory tendencies of ECOP, resulting in inadequate exploitation and inferioroutcomes when compared to other mechanisms undersimilar conditions.

Since the refined and homogeneous common knowl-edge synthesized by the closed knowledge transformationcycle is the primary source of learning in exploitation, itmatters little whether individuals quickly or slowly adoptknowledge through the exploitation tools. All individuals(the team for TR and the population for KRP) simul-taneously transform and are transformed by the sameknowledge. Knowledge will, therefore, ultimately con-verge to the same levels regardless of learning rates. In a

closed knowledge transformation cycle, slower learningdoes not provide more opportunity for knowledge hetero-geneity to exist in the population.

Organizational Turnover. In his original model,March (1991) found that the introduction of bothorganizational turnover and environmental turbulence

enhanced the knowledge level of the organizational codeby introducing an exploratory influence, but consistentlydeteriorated average knowledge levels in the popula-tion. We find in our simulation that IT-enabled learningmechanisms behave differently in relation to the nega-tive effects of these environmental influences found inMarch’s original model.

First, March (1991) found that organizational turnoverdecreased overall equilibrium knowledge levels in thepopulation by shortening the tenure of the individualswithin the organization and providing these individualswith less opportunity to learn from others and to refinetheir knowledge (p. 79). Our simulation of IT-enabledlearning mechanisms in the population showed that cer-tain mechanisms were less susceptible than others tonegative first-order effects of turnover in the populationwhen compared to other mechanisms in the short term.Figure 5 displays the effect of organizational turnoveron average population knowledge level at Round 80.ECOP is radically affected by the introduction of organi-zational turnover. This effect is strong. Under conditionsof no turnover (p3 = 0, ECOP outperforms other mecha-nisms. This performance drops under increasing turnover,underperforming all tools with only moderate turnover(p3 = 02. ECOP performance continues to deterio-

rate more rapidly than other mechanisms under stillhigher turnover. Although average population knowledgelevels of exploitatory tools (KRP and TR) are still neg-atively affected by organizational turnover, they are lessaffected than ECOP, even under relatively high levels of

Figure 5 The Effect of IT-Enabled Learning Mechanisms

Under Organizational Turnover (K pop , Round = 80,

p1 = 09, r ct = 02)

0 0.1 0.2 0.3

Turn

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

K p o p

TR

ECOPKRP




13/17


Figure 6 The Effect of IT-Enabled Learning Mechanisms Under Environmental Turbulence (K pop , Round 1–30, p1 = 09, r ct = 02)

TR

ECOP

KRP

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

K p o p

Turb: 0

Round:5

1015

2025

30

Turb: 0.01

Round:5

1015

2025

30

Turb: 0.03

Round:5

1015

2025

30

turnover (p3 = 03. TRs are the most resistant to highturnover levels.

Exploratory tools are more susceptible to the effectsof organizational turnover as a result of the open knowl-edge transformation cycle. As new individuals enter theorganization, the exploratory tools do not afford them theopportunity to quickly assimilate organizational knowl-edge from a single source. Each new individual is forcedto learn “from scratch” by assembling knowledge fromtheir ECOP network. Each must determine which knowl-edge is valuable in a given reality without the benefitof knowledge improvements made by the organizationprior to his arrival. It takes the individual many roundsto accumulate, assess, and accept organizational knowl-edge, even if the individual is a fast learner. In the mean-time, more employees are turning over, continuing tolower the overall knowledge levels of the organization,and leaving less valuable knowledge in the populationavailable for learning in later rounds. As the rate of turnover increases, the ECOP becomes a less effectivetool for OL because it simply cannot process knowledgefaster than it is being lost via turnover.

The rapid homogenization of organizational knowl-edge permits mechanisms that promote exploitationto better resist the impact of organizational turnover.Because the closed knowledge transformation cycleenables only a single available source from which tolearn, exploitation tools homogenize the knowledge heldby new individuals to that of the population morerapidly. These exploitation mechanisms also limit thedegree to which new knowledge brought via turnoverdisseminates to the rest of the population. Whether dueto the subject matter experts (KRP) or to the locally

defined scope of the TR-group, new individual knowl-edge is weighed against that of an established set of localexperts. Only knowledge that approximates local per-ceptions of existing organizational knowledge is consid-ered for learning in successive rounds. Thus, by limitingthe sources of knowledge on which individuals can rely,exploitation tools resist the effects of turnover by ensur-ing that individuals with new knowledge are quicklybrought into line with organizational expectations.

Environmental Turbulence. March (1991) simulatedenvironmental turbulence by allowing the dimensionsof reality to change according to a particular proba-bility (p4 = 002 and found that there was also aninevitable deterioration of average population knowl-edge within the organization (p. 80). March found thathis simulation resisted the degrading effects of environ-mental turbulence only by introducing new knowledgethrough organizational turnover. We found that certainIT-enabled learning mechanisms were more suscepti-ble than other mechanisms to the environmental effectsof turbulence (Figure 6). Knowledge exploitation toolsexhibited a susceptibility of environmental turbulencesimilar to March’s original model, rapidly convergingto equilibrium knowledge and then slowly deteriorat-ing as the knowledge held in these technologies becameirrelevant with changing reality. ECOP, however, demon-strated considerable resistance to these effects. Evenunder conditions of environmental turbulence higherthan March’s original model (p4 = 003, knowledgecontinues to improve throughout the early rounds of the simulation and does not demonstrate the degradationobserved in the knowledge exploitation tools.

Because ECOP cultivates and preserves knowledgeheterogeneity in later rounds of the simulation, it




14/17


performs better than exploitatory mechanisms in high-turbulence situations. The population still possesses rel-evant knowledge to adapt to a changing reality underenvironmental turbulence. As reality changes, individu-als who possess different knowledge that is now more(or less) relevant in relation to the new reality are readily

identified by the population, and -groups are updatedaccordingly. Through this process, the organization canreevaluate the value of the knowledge possessed by indi-viduals. Average population knowledge levels can con-tinue to improve in the face of a changing reality andgenerally resist environmental turbulence relatively bet-ter than other mechanisms.

Exploitation mechanisms do not perform well in theface of environmental turbulence. The rapid and uni-form homogenization of knowledge makes it difficultfor alternative knowledge to exist within the population.Thus, the organization has little knowledge resourceswith which to adapt to a changing reality. As reality

changes sufficiently from the state to which the organi-zational knowledge has been optimized, reality becomesless and less relevant to the knowledge that the organi-zation has accumulated. As March (1991) noted, knowl-edge improvements devolve toward random chance thatreality will change to match what had previously beenestablished as organizational knowledge—which is a sit-uation analogous to the proverbial stopped clock beingright twice per day.

DiscussionA direct quantitative comparison between our extendedmodel and March’s (1991) original is difficult, becausemany of the baseline changes required to implement ourextensions affect the interpretation of model dynamics(e.g., introducing learning between individuals). Never-theless, we can make some qualitative comparisons of the original model with our extended model that enhanceour understanding of the results. One observation is thatour model provides further evidence to support March’sassertion that knowledge heterogeneity is the source of exploration and exploitation dynamics. We find that thesame fundamental mechanism that drove March’s modelalso drives ours. March, however, attributed most of thedifference between organizations to differences in theemployees themselves (e.g., individual learning rates).We found that IT-enabled learning mechanisms by whichindividuals learn also significantly influence the explo-ration and exploitation balance in firms and should notbe overlooked in future examinations of OL.

Another comparison between our model and March’s(1991) can be made in examining the general influenceof individual learning rates on overall knowledge lev-els. In March’s model, higher learning rates generallydecrease the overall knowledge levels of the organization(p. 76), but in our model of IT-enabled learning mech-anisms, higher learning rates either have no effect orincrease the overall knowledge levels of the organization

(Figure 4). Comparing our results with March’s on thispoint suggests that IT-enabled learning mechanisms maybe effective in some types of organizations (i.e., withfast learners) but may be ineffective or even detrimentalin other types of organizations (i.e., with slow learners).Combined with our other results demonstrating the rel-

ative performance differences between IT-enabled learn-ing mechanisms under various environmental conditions,our results suggest that if the right IT-enabled learningmechanisms are used for learning under the right envi-ronmental and organizational conditions, they can havea considerable benefit for OL. If an organization usesthe wrong type of IT-enabled learning mechanisms givenenvironmental and organizational conditions, however,those mechanisms can be severely detrimental to OL.Thus, IT can be a double-edged sword in relation toOL—either helping or hindering learning.

This work provides some initial insights on whichIT-enabled learning mechanisms should be used under

given conditions (if at all). First, IT-enabled learningmechanisms have different effects on exploration andexploitation processes. Managers would be well advisedto consider their short- and long-term learning goalswhen selecting the IT tools to make available to theiremployees. Certain tools (KRP and TR) tend to pro-mote exploitation by reducing knowledge heterogeneity,leading to improved results for the short term and lowerresults for the long term. Other tools (such as ECOP)cultivate exploration by preserving knowledge hetero-geneity, resulting in better long-term results, but maybe less effective leveraging knowledge in the short-term.IT tools should be selected to support OL with these

knowledge goals in mind.Second, this research has implications for how orga-

nizations can select and combine IT-enabled learningmechanisms. Previous research offers somewhat contra-dictory prescriptions, some suggesting that a combina-tion of mechanisms should prove superior (March 1991),while others advocate a limited combination of differ-ent IT-enabled learning mechanisms in conjunction withone another (Hansen et al. 1999). Our results suggestthat each of these seemingly contradictory prescriptionscan be valid under particular conditions. Organizationsthat primarily pursue a knowledge exploitation strategywould be advised to augment their exploitation tools with

exploration tools, particularly if those tools can bringnew knowledge to the knowledge transformation cycle.Augmenting knowledge exploitation tools with othermechanisms appears to have considerable potential forimprovement with very little downside risk. We also findsome support, however, for the position that organiza-tions choosing an exploration strategy through IT-enabledlearning mechanisms such as ECOP might be advisednot to augment these ECOP with other IT-based mech-anisms. These strategies thrive on an ability to cultivateknowledge heterogeneity within a population, providing




15/17


greater opportunity for future learning and holding upwell under conditions of environmental turbulence. Intro-ducing additional IT-enabled learning mechanisms to theorganization, particularly those that promote knowledgehomogenization, can erode the benefits of ECOP. In thiscase, a pure strategy may prove advisable.

Third, this research offers considerable insight regard-ing how to use IT-enabled learning mechanisms underparticular environmental conditions. Our results sug-gest that organizations can use IT to compensate forthe effects of turnover and turbulence, without neces-sarily resorting to more radical organizational solutionssuch as forced turnover. Thus, an effective portfolio of IT-enabled learning mechanisms might give the orga-nization the opportunity to respond to various envi-ronmental conditions, and the organization would bewell advised to develop its learning mechanisms withthese environmental conditions in mind. For instance, weobserved that ECOP performed well under conditions of low organizational turnover, but that these relative ben-efits rapidly deteriorated as turnover increased. In com-parison, results of TR and KRP are relatively stable inthe face of various rates of turnover. Thus, organizationsthat typically experience a high degree of turnover orvariable turnover rates might cultivate IT-enabled learn-ing mechanisms that function better under these condi-tions. These mechanisms function not only by preservingthe valuable knowledge for later use by the organiza-tion, but also by assimilating new employees quicklyinto the organization and imparting on them the relevantknowledge to address the environmental conditions. Incontrast, organizations that experience substantial envi-ronmental turbulence would benefit from exploration

mechanisms that preserve knowledge heterogeneity inthe population. These tools better retain essential diver-gent knowledge in the organization that provides aknowledge base from which to effectively respond toinevitable changes in future knowledge requirements.

Limitations. Like all research, this study has limi-tations. Most significantly, simulation is a simplifiedrepresentation of a real-world environment. All sim-plification of real-world phenomena for research pur-poses involves drawbacks. One risks oversimplifying agiven process (e.g., knowledge represented in 30 discretedimensions) and idealizing imperfect processes (e.g.,

individuals identifying those with superior knowledge).We have sought to mitigate these limitations by extend-ing an established and highly regarded computationalmodel (March 1991) and by grounding our extensionsto this model using previously published case studies ina single organizational setting. Nevertheless, other orga-nizations may use different IT-enabled learning mecha-nisms or use the same technologies to support differentlearning structures not simulated here. Furthermore, ourmodel suggests that the ECOP strategy consistently out-performs a number of other approaches and may seemto be the logical default mechanism. Although some

researchers have suggested that ECOP is, in fact, thesuperior OL mechanism (Cross and Baird 2000), it isimportant to recognize that we model some importantand common conditions where ECOP is not superior(e.g., high turnover, low individual learning rates) thatmust be considered when introducing IT to support OL.

Summary. This study investigates the effects of IT onexploration and exploitation in OL. We relied on qual-itative case evidence in a single organization to extendan earlier computational model of OL (March 1991) byintroducing learning between individuals in an organi-zation through three distinct IT-enabled learning mecha-nisms: KRPs, TRs, and ECOPs. We found that each of these mechanisms has a distinct effect on the explorationand exploitation dynamics in OL. We further found thathow these tools are blended together, the environmentalconditions under which they operate, and the character-istics of the individuals who use the tools all influence

the impact of IT-enabled learning mechanisms on OL interms of the exploration or exploitation dynamics.Understanding how various IT-enabled learning mech-

anisms support OL when used in different combinationsand under particular environmental conditions can helpus better grasp their effect on exploration and exploita-tion dynamics. The results from computer simulation canalso help extend our understanding of under-researchedareas such as unintended effects of IT in OL.

AcknowledgmentsThe authors thank Mike Prietula for his guidance at the ear-liest stages of this project. The authors also thank Nicolaj

Sigglekow, Mike Gallivan, and Rob Fichman for their insight-ful comments on earlier drafts of this manuscript.

Appendix A. Description of March’s Original ModelReality is modeled as a single external vector of 30 integers,where each integer i presents an orthogonal and independentelement of reality, consisting of 1 or −1. The organizationalcode is similarly modeled as a single vector of beliefs, andeach element of the vector can additionally take on the value 0(representing no belief). Thus, i ∈ −101. Individual knowl-edge is modeled similar to the organizational code (a vectorof 30 beliefs). There is a distinct vector for each individual inthe simulation (100 individuals), each vector representing anorganizational member. Consistent with March’s (1991) origi-

nal model (p. 75), the number of individuals simulated in theorganization had little qualitative influence on model outcomes.

Both organizational and individual knowledge change vialearning, and learning is represented in this model as specificinteractions between the individuals and the code occurringover 80 periods (iterations). Each period, every individual willalter any given belief to conform to that of the organizationalcode with some probability (p1), reflecting the learning rateof individuals in the organization. Likewise, the organizationalcode will also alter any given belief based on the dominantbelief of a knowledgeable group of individuals, hereafter calledthe -group, based on a factor that reflects the learning rate of




16/17


Appendix B. Simulation Flowchart

Begin simulation.Set initial conditions.Determine parameter values for iteration (see Table 2 for parameter values).

Set ECOP networks.

Individuals choose IT tool (repeat for

each of 100 individuals).

Choose IT tool (ECOP, KRP, TR).

AssembleΦ-group of knowledge

sources with higher KV using chosen tool.

Update knowledge values.Determine KV for code, individuals, KRP.

Individual learn (repeat for 30 knowledge

dimensions).

Compare knowledge dimension to Φ-group.

If prob. < p 1, change to majority value.

Send changes to “t +1 array.”a

Code assembles Φ-group.

Environmental conditions.

Determine if organizational turnover.

Determine if environmental turbulence.

Conclude round.Output results for round to file.

“t +1 array”a values become t values.

Clear “t +1 array.”

R e p e a t f o r

1 0 0 i n d i v i d u a l s

d i m e n s i o n s

Code learn (repeat for 30 knowledge dimensions).

Compare knowledge dimension to Φ-group.

If prob. < p 2, change to majority value.

Send changes to “t +1 array.”

a

R e p e a t 8 0 t i m e s f o r

e a c h i t e r a t i o n ( r o u n d ) .

R e p e a t 3 0 t i m e s f o r e a c h

s e t o f p a r a m e t e r v a l u e s ( i t e r a t i o n ) .

Organizational turnover: If prob.< p 3, replace

individual with random knowledge value, reset

ECOP network for new individual.

Environmental turbulence: If prob. < p 4, change

single dimension of reality to opposite value.

R e p e a t

f o r 3 0 k n o w l e d g e

d

i m e n s i o n s .

at+1 array: All changes in individual and code knowledge are first recorded in the “t+1 array,” then simultaneously updated at the end

of the round.

the code p2. This -group is defined as all those individualswho have a higher knowledge level than the code.

The knowledge level is the proportion of reality correctlyrepresented by either the organizational code or an individual.If 10 of an individual’s 30 knowledge dimensions match thatof reality, that individual has a knowledge level of 0.33; if 15 of 30 knowledge dimensions match reality, the individ-ual’s knowledge level is 0.5; and if there is a perfect matchbetween an individual’s knowledge and reality, the individualhas a knowledge level of 1.0. Neither the code nor the individ-uals directly observe reality, but reality influences learning bydefining the -group of individuals in the organization withvaluable knowledge. The code can identify which individu-

als have knowledge that more closely approximates reality inaggregate, but it cannot assess which elements of an individ-ual’s knowledge vector match reality and which do not.

Endnote1We obtained data from semistructured telephone interviewswith 20 employees of Company Z at various locations andhierarchical

kane orgsci

Documents