The role of transactive memory in the formation of organizational routines

Abstract

This study addresses how individuals combine their diverse skills during the process of forming organizational routines. Our explanation centers on the development of transactive memory, which forms during the initial performances of a routine, as actors search for (and subsequently remember) other actors with the capabilities needed to complete a routine. We present an agent-based model to analyze how the distribution and availability of individual capabilities influence the set of actors involved in performing routines, initially and over time. The model shows that even when the pattern of actions stays the same, the pattern of actors involved in performing an organizational routine can change continuously. Variations in the efficiency of routine formation that are inexplicable in terms of action sequences may be readily explained when we examine actor sequences. Transactive memory contributes to the theory of organizational routines by serving as a bridge between individuals’ skills and collective capabilities.

Keywords

Actor sequence variety agent-based model organizational routines transactive memory

Organizational routines form in settings where groups repeatedly solve similar problems and develop solutions learned from past experience (Cohen et al., 1996). In this way, routines substitute memory for search; rather than searching for a novel solution, organizational participants remember and apply what worked in the past (March and Simon, 1958). The literature on routines has interpreted this as searching for what to do, or how to do it, but it is equally important to search for and learn who can accomplish the required tasks. Learning who has which skills involves forming transactive memory, which contributes to the efficiency gains associated with forming routines (Liang et al., 1995; Ren et al., 2006). This study considers how familiarity with others’ capabilities emerges along with patterns of actions in the process of learning and performing organizational routines.

We address the question: What is the role of transactive memory in the formation of organizational routines? Learning associated with the formation of transactive memory gives rise to ties among organizational members and patterns in workflow relationships. Transactive learning facilitates the efficient combination of individuals’ skills during routine performances. As such, transactive memory serves as a bridge between individuals’ skills and collective performative capabilities.

We propose an agent-based model to examine the development of transactive memory through experience solving repeated problems. Our analyses take into consideration differences across organizations in the number of members, distribution of individuals’ skills, personnel turnover, and occasional unavailability of actors. Across organizations that differ along these dimensions, we examine the role of transactive memory in the efficiency of performed routines and variation in the actors involved over time. We introduce the concept of actor sequence variety (ASV) to capture variability in the actors involved in performing a particular pattern of actions. Our analyses show that despite uniform action sequences, the actor sequences within an organization often exhibit variation across performances of a routine, and variations in routine efficiency that are inexplicable in terms of action sequences may be explained readily when we examine actor sequences. Distributing transactive learning over larger numbers of actors and disrupting established relationships among actors increase variety in actor sequences and impede efficiency gains during routine formation.

Our modeled organization has characteristics consistent with collective-centric organizational cognition as described by Michel (2007). The organization consists of individuals with differing skills who rely upon decentralized coordination to manage their workflow. The organization confronts repeated problems that require members to perform a series of sequential tasks, yet initially, they lack an established response. Actors must search to find out who can perform the required tasks. They store in their transactive memories their learning about others’ skills gained from repeated problem solving. Such learning over time establishes ties among actors that enable efficient execution of the repeated patterns of actions associated with routines. Our analysis complements that of Dionysiou and Tsoukas (2013), who consider how routines form when the actors are known but the necessary actions are unknown. Here, the required actions are known, but the relevant actors must be discovered.

Organizational routines and transactive memory

Organizational context

The context that we consider is an organization facing a recurring problem but lacking an established response. Organizational members must become familiar with the problem that they face and organize themselves to respond (Majchrzak et al., 2007). In the absence of formal existing structures and procedures, coordination occurs through decentralized self-organizing (Lindkvist, 2004). Members search to find out who can carry out the tasks that the situation demands, and the emergence of identifiable roles over time facilitates this search (Bechky, 2006; Brandon and Hollingshead, 2004).

This situation is exemplified in Michel’s (2007) description of “Amp Bank,” an investment bank in which newcomers joining the organization experience ongoing cognitive uncertainty. Michel’s (2007) ethnographic research identified in Amp Bank the following key characteristics of collective-centric organizational cognition: an orientation toward tasks rather than roles and titles, staffing based on employee availability and substitutability, solutions arising through social interactions, and ongoing experiential learning. At Amp Bank, there are no titles on employees’ business cards. In the absence of identifiable roles, bankers may be asked to take on a variety of tasks. When a banker goes on vacation or is occupied with other work, other bankers with similar skills serve as substitutes. Amp Bank puts the onus on individuals to engage in ad hoc search and learning to resolve problems. Bank employees work through their personal social networks to tap expertise and resources in the broader organization. In organizations characterized by distributed knowledge, learning who knows what is crucial to addressing complex problems efficiently (Hutchins, 1995; Lindkvist, 2005). Learning over time establishes patterns of actors who interact to accomplish actions that resolve a problem that the organization confronts repeatedly.

Although Amp Bank may be an extreme case, every organization confronts new problems at its inception and occasionally throughout its history. If the new problem is repetitive, then the organization may form a routinized response (Cohen et al., 1996). In doing so, organizational members may engage one another for help in determining an appropriate course of action (Cross and Sproull, 2004). They may approach others to identify where relevant expertise or skills reside in the organization and to transfer work in progress to those who have appropriate skills and legitimacy to perform particular tasks.

Transactive memory

Individuals develop transactive memories in response to their limited and specialized capabilities (Carley, 1999; Palazzolo et al., 2006; Ren et al., 2006). Organizational members may know what needs to be done to solve a problem but lack the ability to perform the required task themselves, so they need to search for someone capable of performing the task. Transactive memory enables efficient identification of individuals with skills and knowledge uniquely suited to performing particular facets of a multiple-task routine. As such, learning who knows what is an efficient response to the limited capacity of individuals’ procedural (know-how) and declarative (know-what) memories. Transactive memory links tasks to required competencies that are embodied in identifiable individuals (Brandon and Hollingshead, 2004), thereby facilitating coordinated execution of tasks (Reagans et al., 2005).

A transactive memory system consists of the collective-level learning about who knows what that is stored in individuals’ memories for subsequent retrieval (Wegner, 1987). Individuals can access their personal transactive memories directly and others’ transactive memories indirectly by communicating. In the model described below, individuals can access others’ transactive memories by asking for referrals when their personal transactive knowledge is insufficient to guide their search for actors with task-relevant skills. The structure of transactive memory systems can be either differentiated—where individuals possess distinct information and skills—or integrated—where individuals have overlapping expertise (Hollingshead, 2001; Wegner, 1987). The performance implications of differentiated and integrated transactive memory systems depend on the kind of problem that a group faces (Gupta and Hollingshead, 2010). By altering the degree to which organizational members’ skills overlap, our analyses include experimental conditions for both differentiated and integrated transactive memory systems. When their skills are similar, individuals may develop similar transactive memories to compensate for shortcomings in their skill set; when their skills differ, their transactive memories also tend to differ.

Explanations of organizational routines neglect transactive memory, whereas they prominently feature procedural and declarative memory (e.g. Cohen and Bacdayan, 1994; Dionysiou and Tsoukas, 2013; Lazaric, 2008; Lazaric and Denis, 2005; Moorman and Miner, 1998). The emphasis is on what to do, while suitably skilled others are assumed to be present and easily found. The neglect of transactive memory in the organizational routines literature parallels that in research on organizational learning, which “has focused on declarative (know-what) or procedural (know-how) knowledge with little inquiry into organizational learning as a function of relationships (know-who)” (Borgatti and Cross, 2003: 433).

Whereas declarative memory informs what to do, transactive memory is crucial to identifying and involving appropriately skilled individuals to complete the tasks in a routine. Limits to individuals’ procedural and declarative knowledge create interdependence among organizational members, who solve complex problems by developing interpersonal connections to access capabilities complementary to their own. In this sense, transactive memory facilitates connections among the necessary capabilities distributed across organizational members (Argote and Ren, 2012). Accordingly, omitting transactive memory leaves a gaping hole when trying to account for the knowledge reflected in organizational routines. Our model employs transactive memory, as established in prior group and organizational research (e.g. Hollingshead, 1998; Lewis, 2004; Littlepage et al., 1997; Moreland et al., 1996; Wegner, 1987), to explain the set of actors and efficiency gains associated with routine formation.

Actor sequences

Abell et al. (2008) and Felin et al. (2012) argue for an explanatory approach to organizational routines that addresses individuals’ actions and interactions. Their contention is part of a general research agenda that seeks microfoundations that link the individual and collective levels in theories of strategic organization (Felin and Foss, 2005). Likening routines to individual-level phenomena such as skills (Nelson and Winter, 1982), dispositions (Hodgson, 2003), and procedural memory (Hodgson, 2003; Lazaric, 2008) does not acknowledge sufficiently the individuals who carry out routines. Commenting on familiar comparisons of organizational routines to individual habits, scripts, and genes, Feldman and Pentland (2003) note that “ironically, there are no people in these traditional metaphors” (p. 99).

Furthermore, Feldman and Rafaeli (2002) observe that “[p]revious analyses of organizational routines have not focused on the connections that routines make between and for people” (p. 312). They attribute this omission to researchers’ tendency to address “routines in the abstract” instead of studying actual performances. As a corrective, they propose an alternative framing: “Organizational routines entail multiple interpersonal interactions with other organizational members. They specify for organizational members the respective other members with whom behavior needs to be coordinated” (Feldman and Rafaeli, 2002: 312). Every organizational routine requires a set of actors who are connected by ties involving communication and workflow.

Our response to the call for microfoundations is to highlight transactive memory as a bridge between individuals and the collective capabilities expressed in organizational routines. We model the formation of organizational routines by beginning with individuals who are endowed with diverse skills, and the capacity to learn about others’ skills (i.e. who knows what). Problems trigger individuals’ responses and, where individuals’ skills prove insufficient, search provides opportunities for transactive learning. Transactive learning by individuals supports patterned sequences of actors that efficiently solve repeated problems. Patterns of actors, like patterns of actions, come about through individual effort and mindfulness (see Feldman, 2003; Steen et al., 2006).

In keeping with an orientation toward microfoundations, our analyses focus on four organizational demographic characteristics: (1) the number of organizational members, (2) the distribution of individuals’ skills, (3) personnel turnover, and (4) occasional unavailability of actors. These variables are fundamental to describing the pool of actors who will interact to carry out a given routine. The number of organizational members and the distribution of individuals’ skills determine who is capable of performing tasks. Adding organizational members and increasing individuals’ skills expand the pool of agents who potentially can fulfill a task as part of a workflow routine. Personnel turnover and unavailability of organizational members influence who is available to participate in organizational routines at any point in time. Temporary unavailability—due to absenteeism or attending to other duties—diminishes the proportion of an organization’s members who can perform assigned tasks in any period. The disruption is more severe for turnover than for temporary unavailability. With turnover, experienced individuals leave and are replaced by others who are both unknowing (of others in the organization) and unknown (by others in the organization). Turnover and unavailability of individuals with particular skills pose practical challenges because they disrupt established networks for accomplishing routines.

As portrayed in our model, the need (or opportunity) to solve recurring problems requiring the completion of sequential tasks presents the context in which organizational routines emerge (see Cohen and Bacdayan, 1994; Pentland, 2003). The nature of the presenting problem constrains the resulting task sequence, but there is latitude for an organization to involve different actors in completing the routine. In other words, problems underdetermine actor sequences. Each time an organization solves a problem, it involves a series of actors who complete the sequential tasks that make up a routine. In addition to conveying who did what, the actor sequence tells us who handed off the problem to whom.

To summarize the degree of stability or instability in the pattern of actors involved in performing routines over time, we introduce a new measure: actor sequence variety (ASV). The measure is analogous to sequential variety (Pentland, 2003), but it is computed based on the sequence of actors who perform the routine, rather than the sequence of actions. If there is no variation in the pattern of actors from one performance of a routine to the next, ASV is zero. When the actors involved in the routine change from one performance to the next, then ASV is positive. ASV provides a simple way to quantify the extent to which actor sequences vary over routine performances.

Description of the agent-based model

This section specifies the modeled problems, individual behaviors, and organizational features. Our agent-based model builds on prior work by Miller et al. (2012) who examined how different types of individual memory affect the efficiency of organizational routines (as measured by cycle time).¹ The present study analyzes the role of transactive memory in establishing sequences of actors to accomplish routines, and our principal outcome of interest is ASV. The set of organizational demographic factors analyzed here extends beyond those considered by Miller et al. (2012).

Model specification

Problems

An organization faces a repeated problem requiring k different sequential tasks. This problem reflects a recurring demand imposed on the organization by its operating environment. Consider, for example, that Amp Bank needs to develop a new contracting routine made up of tasks requiring the expertise of multiple employees, and this routine will be used repeatedly to meet requests from clients in the future. Our focus on sequential tasks follows routines research on action sequences (e.g. Cohen and Bacdayan, 1994; Pentland, 2003). We assume a standard problem ordering the tasks as 1, 2, …, k. To resolve a problem, tasks must be completed in the exact order in which they occur in the problem. Because of serial interdependence among tasks, the resulting organizational technology is “long-linked” (Thompson, 1967: 15–16). Each task takes an actor one period to complete.

Problems enter the organization via the set of workers qualified to start the problem-solving process. New problems are assigned to an actor selected at random among those able to complete the first required task.² If an actor completes a task and has the skill to complete the subsequent task, it does so in the next period in which it is available. When an actor cannot complete the subsequent task in a problem, it searches for another actor to whom it can transfer the problem for further work (as explained below in “Decentralized search process”). Upon completing all the tasks in a problem, the organization receives an identical new problem in the next period.

Organization and skills

The organization consists of n actors. Actors possess skills that do not change or improve during a simulation run. Assuming stable skills allows us to isolate the effects of transactive learning. Actor i (i = 1,…, n) possesses s_i skills, where $s_{i} \in {1, \dots, k}$ . The k possible skills fit the tasks making up the recurrent problem that the organization must solve (i.e. skill j can accomplish task j for j = 1,…, k). The individuals in the organization collectively must possess all k skills for resolution of the standard sequential-task problem to be feasible.³ When n is larger than k, the organization has agents with overlapping skills. Such is the case among bankers at Amp Bank who are considered fungible.

In any period, actors are unavailable to apply their skills to a problem with probability q (0 ≤ q ≤ 1). The rate of unavailability reflects the extent to which actors are occupied with other activities or temporarily absent and, as such, unavailable to contribute to solving the problem in that period.⁴ If the actor holding the problem becomes unavailable in a given period, it does not perform a task or search during that period. An unavailable actor cannot receive a problem to perform a task or make a referral (as explained below in “Decentralized search process”).

Personnel turnover permanently affects the set of agents capable of performing tasks. In our model, the probability of any actor being replaced between problems by a new actor is v (0 ≤ v ≤ 1). Our approach assumes that turnover occurs after completing a problem, not in the middle of solving a problem. A newly hired actor brings the same number of skills as its predecessor, but only one of its skills is essential to its position and its additional skills are assigned at random. By implication, when actors each possess a single skill, a newly hired actor must bring the same skill as its predecessor. Newcomers have no transactive knowledge; they build transactive knowledge as they gain experience solving problems.

Transactive memory

Following Palazzolo et al. (2006) and Ren et al. (2006), we model changes in the content of agents’ transactive memories as resulting from work experience gained over time. Initially, actors have no knowledge of which actors in the organization possess which skills. Each actor’s transactive memory starts out as a k-dimensional vector of zeroes. Through experience over time, actors come to know who knows what. When an actor transfers a problem to another actor, it has an opportunity to record the problem-receiving actor as demonstrating the particular skill used in the next task. It updates its transactive memory with probability p. A high p reflects, among other things, willingness to assign high task credibility to a new transaction partner (see Liang et al., 1995). An actor’s memory of who has demonstrated a particular skill includes only its most recent remembered experience.

When an actor leaves the organization (due to turnover), what others in the organization know about that person is no longer relevant to accomplishing tasks because the departing actor’s skills are no longer accessible. Hence, our model deletes all references to a departing actor from the transactive memories of the remaining actors. Remaining organizational members must search to find a replacement qualified to complete a task that the departing actor handled previously.

Decentralized search process

If the actor holding the problem is unable to complete the next task, it searches for another actor who can receive the problem and complete the required task. Searching involves (step 1) checking the actor’s own transactive memory to identify another actor with the skill sought, and, if step 1 is unsuccessful, (step 2) approaching an actor chosen at random to find out if it possesses the skill sought. Random selection may reflect search under ignorance or search based on social ties unrelated to skills (e.g. affective ties). During each of these two search steps, if the approached actor is available and can perform the task needed, it accepts the problem and performs the task. In the second step, if the approached actor is unable to perform the task, then it checks its transactive memory to see if it can make a referral to an actor with the task-relevant skill. The latter response lets the searching actor access the transactive memory of the helping actor. At most, only one such referral occurs in a period. If the actor fails to transfer the problem after following the search steps, the period ends and the actor resumes the search process in the next period. The search process repeats until a problem is finally transferred and a task completed, irrespective of the number of periods required. This decentralized search process reflects task orientation, which emphasizes who can perform a task rather than titles or roles, and inductive search as characteristics of collective-centric organizational cognition (Michel, 2007).

Figure 1 presents a flow chart for the decisions and actions of an actor holding a problem. For simplicity, the figure assumes that all actors are available (i.e. q is at its default value of zero). Beginning at the top of the chart (“Do I have the task relevant skill?”), it takes one period to cycle through the sequence of decisions and actions and either return to this starting point or stop (because the final task in a problem was completed). The critical point for connections among actors is when an actor hands off a problem to another actor and updates its transactive memory (with probability p). The actor receiving the problem completes a task and proceeds with the flow of decisions and actions, now acting as the holder of the problem.

Figure 1.

Flow chart for search and task completion.

Outcomes of interest

Actor sequence variety

As noted earlier, the pattern of actions associated with a routine has been a central concern in routines research, but the pattern of actors has been neglected. To address this gap, we measure variation in the actors involved in a routine. ASV reflects the extent to which different sequences of actors get involved in solving each problem within a set of problems occurring over time. If problems stimulate the involvement of the same actors performing the same tasks for each problem, then the links between actors (where one actor hands off a problem to another) are stable, and ASV is minimal. To the extent that different actors get involved in solving problems, or the order in which actors participate differs from one problem to another, ASV increases. Fungibility of actors, as in Amp Bank, is a prerequisite for variation in actor sequences within a routine over time.

Let d(g,h) designate the Hamming distance between actor sequences for problems g and h (g < h). Hamming distance is the number of changes (switches between actors) required to make one actor sequence equivalent to another. The total number of sequences being compared is c, so the mean Hamming distance is

ASV = \frac{1}{c (c - 1) / 2} \sum_{h = g + 1}^{c} \sum_{g = 1}^{c - 1} d (g, h)

We compute the ASV measure using a moving window of c = 10 problems. As such, the first measurement of ASV occurs after problem 10 and is recomputed after each subsequent problem is solved. We use Hamming distance rather than Levenshtein distance because, in this simulation, all actor sequences are of the same length.

Cycle time

March and Simon (1958: 142) portrayed routines as stable responses to particular stimuli that enhance organizational efficiency by reducing search. Subsequent research further associated the formation of routines with efficiency gains (Cohen and Bacdayan, 1994; Cohen et al., 1996; Nelson and Winter, 1982; Winter and Szulanski, 2001). We track cycle time as an indicator of the learning associated with routine formation. Cycle time is the number of periods required to complete a problem consisting of k tasks. Cycle time for a given problem is the number of periods in which tasks are completed plus the number of periods in which the actor holding the problem is unavailable to search or its search did not result in a completed task. To measure changes in efficiency, we track cycle time as a function of the number of problems solved. We present means computed across all runs of a model.

Analysis and result

We implemented our model using MATLAB 7. Table 1 provides a summary of the model parameters, default values, and ranges used in the simulation runs. A model run consists of a series of 100 problems. All of the results presented in this section are averages based upon 100 runs at each parameter combination. We examine first the demographic variables affecting who is capable of performing tasks: organizational size (n) and the distribution of skills (s_i). Then, we consider disturbances affecting actors’ availability to participate in problem-solving routines: the rates of unavailability to assist in problem solving (q) and turnover (v).

Table 1.

Model parameters.

Parameter	Values^a	Meaning
k	10	Number of tasks in a problem
n	10, 20,…, 100	Number of actors in the organization
s_i	1, 2,…, 10	Number of skills for actor i (i = 1,…, n)
p	0.2	Probability that an actor updates its transactive memory
q	0, 0.1, 0.2,…, 0.5	Probability of actor being unavailable to perform a task
v	0, 0.01,…, 0.1	Average rate of actor turnover after each problem.

Values underlined are default settings.

Organization size and skill distribution

We begin by analyzing the effects of changes in the number of actors (n). Figure 2(a) shows how ASV declines over time for organizations of different sizes. Increasing the number of actors (n) increases ASV for early problems and in the long run. When only one unique single-skilled actor is available to solve each of the tasks (i.e. s_i = 1 for all i = 1, …, 10), only one possible sequence of actors can resolve a problem, so ASV starts and remains at zero. By contrast, when there is more than one candidate capable of performing a task, actor sequences can vary across problems.

Figure 2.

Organization size: (a) actor sequence variety over problems, (b) actor sequence variety at specific problems, and (c) cycle time over problems.

Figure 3 provides an example of how the pattern of ties changes over problems. Circles represent actors, which are numbered sequentially from 1 to 20. Single-skilled agents 1 through 10 can fulfill tasks 1 through 10, respectively, as can their redundant counterparts 11 through 20. Arrows indicate the agents involved in transferring a problem and the direction of the exchange. The numbers along each arrow indicate the number of cumulative exchanges between pairs of actors over the range of problems (1 through 5 or 51 through 55). For early problems, the organization explores various actor sequences to solve the repeated problem (Figure 3(a)). However, ASV declines with problem-solving experience and stabilizes as primary actor sequences emerge over time (Figures 2(a) and 3(b)).

Figure 3.

Dynamics of problem-solving networks: (a) problems 1–5 and (b) problems 51–55.

Figure 2(b) depicts the relations between organization size and ASV after problem 10, when the first measurement of ASV occurs, and after problem 100, at the end of each model run. We refer to these as the short-run and long-run relations. Figure 2(b) shows that the relation between organizational size and ASV is positive in both the short and long runs, but the functional form changes from concave to linear. Underlying this relation is the development of actors’ transactive memories. In early periods, as transactive memories are forming, search generates variability in actor sequences. Additional actors have a declining marginal effect on this variability. Early random choices stabilize into fixed network patterns over time. In the long run, ASV is proportional to organizational size.

Figure 2(c) shows the cycle time reductions as actors learn who knows what. Through experience, actors build transactive memory and select dominant actor sequences that produce cycle time reductions by reducing search (as portrayed in Figure 3). The presence of redundant actors lengthens the time required for routine to form (as indicated by achieving and maintaining the minimum possible cycle time). Involving more than one actor to accomplish the same task across problems spreads the need to search and learn among more actors than if only a single individual is skilled for each task. Figures 2(a) and (c) point out the learning resulting from early problem solving and the diminishing marginal learning from further experience.

We took two different approaches to varying the distribution of skills across actors. First, we made one randomly chosen actor multiskilled and varied that actor’s number of skills from one to ten. The distribution of skills in this organization is uniform (one skill per actor) except for a single spike. We maintained the default organization size of 20 actors. The presence of a multiskilled actor increases the number of potential performers for each task. As shown in Figure 4(a), the larger the number of skills possessed by the multiskilled actor, the more slowly ASV declines over time and the higher is its long-run level. Figure 4(b) shows contrasting effects in the short and long runs. The short-run inverted-U relation results from two considerations: (1) as a single actor has a larger range of skills, the number of possible actor sequences for completing the routine increases, while (2) the number of realized sequences diminishes due to centralization of the routine in one actor. In the long run, redundancy in the skills available in the organization increases the instability of actor sequences at an increasing rate. The long-run convex relation reflects exponential growth in the number of possible actor sequences as the number of skills of the multiskilled actor rises.

Figure 4.

Single-spike skill distribution: (a) actor sequence variety over problems, (b) actor sequence variety at specific problems, and (c) cycle time over problems.

Figure 5 illustrates a cumulative pattern of actors for one run (100 problems) in an organization with a single multiskilled actor who has nine skills (s₁₀ = 9). Circles represent the 19 single-skilled actors. A diamond represents the multiskilled actor, number 10, who can complete any task except 7. Figure 5(a) shows the actor involved in performing each of the 10 sequential tasks, with arrows indicating the path that a problem travels toward completion. The numbers along each arrow indicate the number of times (out of 100) that a particular exchange occurred. Figure 5(b) shows the conventional network diagram in which a single node represents each actor, and arrows indicate where dyadic ties (i.e. problem handoffs) occur and their direction. The presence of one multiskilled actor creates additional possible paths for accomplishing each task and thereby increases the variety of problem-solving actor sequences (Figure 5(a)). Because the multiskilled actor performs multiple tasks, its position is central in the resulting network (Figure 5(b)).

Figure 5.

Cumulative network with a single multiskilled actor (s₁₀ = 9): (a) actor–task sequences and (b) exchange relations.

The presence of a multiskilled actor reduces the initial problem-solving cycle time; however, cycle times quickly converge as transactive memory accumulates and search diminishes (Figure 4(c)). There is no long-run cycle-time advantage associated with having a multiskilled actor in the organization.

In the second set of analyses varying the number of skills per actor, we considered different numbers of skills distributed uniformly across all 20 actors (s_i = 1, 2, 3,…, 10 for all i). The longitudinal results show limited declines in ASV (Figure 6(a)) relative to those for the single-spike distribution (Figure 4(a)). When s_i ≥ 2, different actors record and retain different entries for the same task in their transactive memories. The short-run effect on ASV of increasing the number of skills per actor is positive and concave (Figure 6(b)). As the number of skills per actor increases, the probability that an actor searching randomly finds an appropriately skilled actor increases, but the average probability of involving any qualified actor in a particular task declines. The concavity of this relation declines with experience.

Figure 6.

Uniform skill distribution: (a) actor sequence variety over problems, (b) actor sequence variety at specific problems, and (c) cycle time over problems.

In the long run, ASV increases monotonically with the number of skills per agent (s_i) distributed uniformly. The reason for this becomes clear if we consider that ASV measures the average Hamming distance between actor sequences over a set of performed routines. As each actor’s skills cover a wider range of tasks, it performs a wider set of tasks when given the opportunity. For the extreme case of all actors possessing all skills (s_i = k for all i), the agent that starts a problem completes every task in the problem. ASV is high for this case because the actor sequences for problems begun by different agents are completely different from one another. In this extreme case, there is no variation in the actors involved within a particular performance of the routine but maximum variation in the actors involved across performances.

The pattern for cycle time reductions is similar for the uniform skill distribution (Figure 6(c)) and the single-spike distribution (Figure 4(c)). The fewer the skills per agent, the higher is the initial cycle time and the more rapid is the ensuing decline. In the long run, drawing upon transactive memory drives out search, producing a null effect of s_i on cycle time.

Actor unavailability and turnover

We now introduce two realistic complications that impede routine formation and disturb established routines: unavailability and turnover of organizational personnel. We first present the results when actors are not always available to complete tasks. As introduced earlier, q is the probability of any actor being unavailable to search, perform a task, or make a referral in a given period. Our model assigned each actor’s availability as a probabilistic draw at the beginning of each period. If a searching actor finds that the actor to whom it routinely goes for a particular skill is unavailable, it searches by randomly selecting another actor and, if necessary, asking for a referral (see Figure 1). As such, increasing the rate of unavailability (q) increases ASV (Figure 7(a)). In Figure 7(b), the long-run effect of unavailability on ASV is positive and approximately linear. Unavailability continues to disrupt existing actor sequences, and the formation of transactive memory does not offset this effect. By contrast, the short-run effect of unavailability on ASV is positive, but almost negligible. In early periods, when actors still are building transactive memory, they are much less consistent in their patterns of returning to the same actors to accomplish particular tasks. As a result, the unavailability of a subset of actors each period raises ASV very little.

Figure 7.

Actor unavailability: (a) actor sequence variety over problems, (b) actor sequence variety at specific problems, and (c) cycle time over problems.

As Figure 7(c) shows, actor unavailability lengthens both the initial and long-run cycle time for completing the routine. An organization fails to reach minimum cycle time because unavailability forces search for new actors who can perform particular tasks.

Figure 8(a) illustrates that increasing the rate of actor turnover (v) makes actor sequences more turbulent in the short and long runs, as we would expect. Figure 8(b) shows the positive effect of turnover to be nearly linear in the short run, but it becomes concave in the long run. We already have noted that actor sequences tend to be quite unstable in the early periods, so ASV is less sensitive to turnover in the short run than in the long run. In the long run, there is a diminishing marginal effect of turnover on ASV. Once turning over a few actors disrupts the actor sequence, additional departures add marginally less disruption.

Figure 8.

Personnel turnover: (a) actor sequence variety over problems, (b) actor sequence variety at specific problems, and (c) cycle time over problems.

Figure 8(c) shows the failure to reach the minimum cycle time for organizations experiencing turnover. In addition, the constant loss and recovery of transactive memory increases cycle time variance, and this variance is an increasing function of v.

The effect on ASV is much more dramatic for personnel turnover than for actor unavailability. Comparing values of v that are an order of magnitude below those for q, the resulting levels of ASV due to turnover exceed those due to unavailability (compare Figures 7(b) and 8(b)). Likewise, personnel turnover has a more detrimental effect on efficiency gains associated with routine formation than does temporary unavailability (compare Figures 7(c) and 8(c)). Our findings regarding turnover are consistent with empirical evidence on the detrimental effect of turnover on group task performance (Argote et al., 1995; Van der Vegt et al., 2010). Underlying the serious disruptive effect of turnover is the loss of transactive memory acquired through experience with a group of coworkers.

We evaluated the robustness of our findings by varying the default values of each of the demographic variables (n, s_i, q, and v) and checking the implications for the effects of the other three variables. Using widely varying default values did not alter the qualitative outcomes.

We also analyzed the moderating effects of the probability of updating transactive memory (p), and our qualitative findings proved robust to variations in these parameters.⁵

Discussion

Explaining routines

Routines, transactive memory, and networks

As a complement to explanations for routine formation based on procedural (know-how) and declarative (know-what) memory (e.g. Cohen and Bacdayan, 1994; Lazaric, 2008; Lazaric and Denis, 2005; Moorman and Miner, 1998), our study highlights the role of transactive (know-who) memory in the formation and execution of routines. Here we portray organizational routines as emerging when organizational members search for solutions to repeated problems. What distinguishes our analysis from conventional perspectives on organizational routines is the content of the search. Rather than searching for what to do, workers search for who can perform identifiable tasks during a workflow process. Limitations on individuals’ procedural and declarative knowledge lead them to develop transactive memory to access efficiently others’ complementary capabilities (Cross and Sproull, 2004). Organizational members keep track of who has which skills and retrieve this information when they face a problem that resembles one already solved (Nebus, 2006; Wegner, 1987). In so doing, they economize on the time, effort, and other costs involved in searching for help (March and Simon, 1958).

Our model highlights transactive memory as the mechanism that produces the interpersonal networks needed for multiple actors with specialized skills to perform efficiently organizational routines. In this way, our model fits Carley’s (1999) general framing in which network structure emerges endogenously as an implication of learning among actors with diverse knowledge. Transactive memory is the cognitive side and repository of network structure. How organizational members form and access their transactive memories affects the realized interpersonal ties and, by implication, ASV and routine efficiency.

Our model reflects the dynamic relations between individuals, problems, and networks. Networks are contexts for action (e.g. Burt, 1982; Lin, 2001) and also products of action. Networks provide access to people and information relevant to solving problems (Cross and Sproull, 2004). Through networks, actors accomplish work, and exchange and accumulate resources. One particular form of action accomplished through social networks is routinized workflow, which involves a series of individuals fulfilling different tasks within an organizational process. Recurring workflow transactions—receiving inputs and distributing outputs—establish an identifiable network structure. Workflow networks embody what Borgatti and Foster (2003) refer to as a “flow” or “pipes” conception of network ties. Materials and information flow along these ties, as work-in-progress passes from one person to the next. Within this framing, individuals’ efforts to complete the tasks required to solve organizational problems shape interpersonal networks within organizations. In turn, networks influence how work gets done—who participates, the sequencing of that participation, and the problem-solving efficiency.

According to Feldman and Pentland (2003), organizational routines consist of two aspects: the ostensive and the performative. The performative aspect is the particular instantiation of the routine—involving particular people doing particular tasks at particular times and places. The ostensive aspect is an abstract representation or typification of the routine, which may differ across participants. Transactive knowledge contributes to the ostensive aspect of organizational routines by informing who gets involved in problem solving and when their participation occurs. This knowledge is distributed among organizational members, with each actor’s understanding shaped by unique experiences from past routine performances. The ostensive aspect of the routine consists of more than just procedural and declarative knowledge. When knowledge is diverse and distributed (Hutchins, 1995; Weick and Roberts, 1993), organizational members need to collaborate and coordinate tasks. This study presents transactive memory as a key mechanism to realize each performance of a routine by linking individuals with specialized skills.

Examining the role of transactive memory contributes to our understanding of the microfoundations of organizational routines framed in terms of individuals and their interactions (see Abell et al., 2008; Felin et al., 2012). Individuals with diverse and complementary capabilities connect with one another in the course of solving multifaceted problems. Networks arise within organizations out of the networking activity of individuals (Steen et al., 2006). Handling the workflow associated with organizational problem-solving is a primary context for seeking out others for their unique skills. In our model, repeated interactions between actors emerge with experience over time through the development of transactive memory. Transactive memory serves as a bridge between individuals’ skills and the collective capabilities demonstrated in organizational routines.

Levitt and March (1988) contend that “[r]outines are independent of the individual actors who execute them and are capable of surviving considerable turnover in individual actors” (p. 320). Empirical evidence also suggests that routines that are regimented can persist even when there is turnover of key personnel (Aime et al., 2010; Rao and Argote, 2006; Ton and Huckman, 2008). Specifying a formal model allows us to identify two key assumptions underlying claims about the continuity of organizational routines in the face of turnover. The irrelevance of particular individuals to an organization’s ability to perform a routine holds only if (1) there are other individuals available in the organization who possess similar skills and (2) the transactive memory system can readily identify substitutes to replace those who depart. These two conditions assure that the organization avoids costly search for replacements outside or within the organization. Absent either of these conditions, turnover causes a loss of continuity in the network of sequential actors and a loss of transactive knowledge that is costly to replace.

Actor sequences within routines

Our interest in characterizing the degree of variability in the pattern of actors over a set of solved problems led us to introduce a new measure: ASV. Whenever the actor sequence varies, either (a) new ties are being formed or (b) alternative networks, already established through prior experience, can fulfill the same set of sequential tasks. The latter case is one of equifinality among alternative actor networks within the organization due to substitutable actors with redundant skills. ASV correlates with network density when actors have limited skills: high ASV implies that on average actors have high in-degree and out-degree.

Modeling a repeated task sequence that never varies allowed us to isolate the causes of variation in routine efficiency arising solely on the actor side. Despite uniform action sequences, the actor sequence often exhibited variation across performances of a routine. The greater the overlap in agents’ skills, the larger the number of potential ties among actors and the more broadly transactive learning gets distributed across actors. Disruptions in the availability of particular actors force search for new actors who can perform particular tasks. Network patterns change as actors update their transactive memories after finding out about others’ skills. A fixed sequence of actions can involve a continuously changing sequence of actors, resulting in an evolving actor network. Studies examining patterns of action in organizational routines find ongoing change (Feldman, 2000; Pentland et al., 2011; Tsoukas and Chia, 2002). Our study makes a complementary claim about ongoing variation in routines on the actor side.

Variation in the set of actors involved in a routine increases the possibilities for errors, innovation, and improvization. Thus, changing participation should influence what tasks get done, and how and when tasks get done. Newcomers are a source of novelty (Carley, 1991; March, 1991), so variation in the actors involved in a routine should tend to increase the potential for variation in actions. This leads to the counter-intuitive insight that repeatedly solving the same problem can produce ongoing change in how a routine is carried out.

Our model allowed us to pursue an experimental design that systematically varied four key demographic factors affecting actor networks and the efficiency of routine formation. These four organizational demographic factors can be categorized into two types: “who is capable” and “who is available” to perform organization routines. The findings show that “who is capable” factors (the number of organizational members and their skill distribution) are related to spreading transactive learning over larger numbers of actors. “Who is available” factors (unavailability and turnover) disrupt established ties between actors and stimulate search for alternate actors. Both factors can increase variety in actor sequences and reduce the efficiency of performed routines. Table 2 summarizes our findings for ASV from experiments varying each of the organizational demographic variables. The reported short-run effects are the relations for the 10th problem and the long-run effects are for the 100th problem.

Table 2.

Summary of effects on actor sequence variety.

	Short run		Long run
	Relation	Function	Relation	Function
Size (n)	+	Concave	+	Linear
Skills (s_i): single-spike	Inverted-U	Concave	+	Convex
Skills (s_i): uniform	+	Concave	+	Linear
Unavailability (q)	+	Linear^a	+	Linear^a
Turnover (v)	+	Linear^a	+	Concave

Indicates approximate, rather than strict, linearity.

We might expect that ASV would increase with each of the independent variables. In general, this proved to be the case, but there are a couple of noteworthy exceptions. First, for the single-spike skill distribution, as the number of skills increases for the multiskilled individual, ASV rises and then falls in the short run (Figure 4(b)). As a single individual becomes highly skilled, centralization increases, routine performances become less interactive, and the network structure becomes simpler and more stable. Second, the short-run effect of actor unavailability on ASV, although positive, is very weak (Figure 7(b)). Because transactive memory is underdeveloped during early problem solving, the actor sequence is highly unstable from one problem to the next. Search is widespread during these early experiences, and only later does transactive memory substitute for search, increasing the stability of the actor network. Because network structure is unstable in early periods, unavailability of some actors each period adds little to ASV.

Beyond identifying the direction of the effects of demographic variables on variations in actor sequences, our model results identify the functional forms of these relations. Table 2 reports that some relations are concave, others linear (or nearly linear), and still others convex. There are distinct patterns among the short-run results and among the long-run results, and contrasts across the two sets of results. It is unlikely that relying solely on deductive reasoning from the assumptions of our model would lead one to these subtle distinctions in ASV outcomes. A key advantage of modeling over relying solely upon conceptual arguments and one’s own deductive reasoning ability is that deriving the functional form of relations provides more precise guidance for empirical testing than do characterizations of the signs of relations alone.

The effects on cycle time also show distinct patterns. Whereas organizational size and the extent to which actors possess overlapping skills have important effects at the early stage in routine formation, these effects become negligible over time. This change over time reflects the growth of transactive memory, which is the central explanatory mechanism in our model. Distributing learning over many actors produces inefficient routine performances. Nevertheless, as actors learn who knows what, they can perform routines efficiently regardless of the number of actors in the organization and the distribution of their skills. By contrast, actor unavailability and turnover cause ongoing inefficiency in performed routines that never goes away. Because turnover causes permanent loss of transactive memory, its effects are more disruptive of routines than when individuals are temporarily unavailable but remain in the organization.

Model extensions and future research

As with other forms of theorizing, constructing a computational model involves making simplifying assumptions that abstract from reality in order to focus on key aspects of the phenomenon of interest (Burton and Obel, 1995; Midgley et al., 2007). This study depicts organizational routines as patterns of sequential tasks undertaken by multiple actors (see Becker, 2005; Feldman and Pentland, 2003). In specifying our model, we incorporated simplifying assumptions consistent with isolating and understanding the relation between transactive memory and routine formation. Our model reflects a rationale for the formation of organizational routines based on individuals engaging in decentralized search and learning in order to solve repeated complex problems efficiently. By including problems and individuals, our model goes further toward illuminating the microstructure of organizational routines than prior simulation models that purported to address routines but treated them abstractly (e.g. Bruderer and Singh, 1996; Crowston, 1996; Gavetti and Levinthal, 2000; Levinthal and Warglien, 1999; Rivkin, 2001) rather than trying to reflect the internal features that make up routines.

Some of the assumptions that we made could be challenged. We postulated an initial state of ignorance of others’ skills and no preexisting interpersonal ties like Dionysiou and Tsoukas (2013) who assumed, for the sake of theory development, actors who had “no history of prior interaction with each other” (p. 189). This assumption serves theory development by making the resulting network structure entirely a function of search and learning in response to the repeated problems that the organization faces. We recognize that the network structure of any real-world organization is subject to other influences, such as the organization’s formal structure and social ties formed for a variety of reasons other than workflow. As such, network structure is not entirely formed through self-organizing starting from a state of no transactive knowledge. However, our model contributes to the theory of routines by highlighting the critical transactive learning that occurs while forming new routines among organizational members who previously were unfamiliar with each other. Empirical research must take into consideration the preexisting social network structure in order to explain the resulting structure associated with forming a new routine but, for modeling purposes, imposing an initial network structure introduces confounding effects that reduce the explanatory power of transactive learning from repeated problem solving. A preexisting social network would make the resulting actor sequences attributable in part to exogenous causes, rather than isolate the effects of the theorized learning process.

Another point where our model deviates from the reality of many organizations is in the assumed self-organizing search process. Our model portrays features of collective-centric organizational cognition (Michel, 2007) and neglects centralized hierarchical and rule-oriented approaches to establishing routines. Despite limits to its generalizability, our model helps us understand how routines arise endogenously, rather than as exogenous impositions on the participants. A possible extension of our model could be to tease out the tension between the spontaneous and planned order in routine formation. Our model provides a starting point for such work.

How we characterized the problem that the organization faces connects our model to prior research on organizational routines. As such, we focused on repeated problems involving sequential tasks and a work process that includes the possibility of transferring problems from one organizational member to another. By addressing the setting of an organization facing a repeated identical problem, we are able to generate the repeated pattern of actions that is the core characteristic of organizational routines. Nevertheless, instead of a stream of identical problems, an organization could face a stream of different problems (see Gavetti et al., 2005). As an extension of our model, it would be useful to investigate how routines form when organizations confront problems that vary over time. In this situation, transactive memory systems enable an organization to combine knowledge distributed across its members in new ways to develop new solutions that respond to changing demands (Argote and Ren, 2012). Solving a stream of varying problems also involves learning to identify their task requirements through direct assessment of problems and inferential reasoning from past experience. Hence, addressing varying problems calls for learning beyond the development of transactive memory featured in the present model. Specifying this learning process and analyzing its implications in conjunction with transactive learning could generate further insights into the process of routine formation and change.

Although simulation research can never substitute for empirical research, the control afforded by simulation modeling makes it a compelling method for deducing causal relations for complex organizational phenomena (Burton, 2003). We see simulation modeling as a useful complement to the field settings typically studied in routines research. Such settings provide little experimental control or quasi-experimental variability in the organizational demographic characteristics featured in our model. Our model and analyses should motivate empirical studies evaluating our theoretical assumptions and derived testable relations.

Empirical research connecting routine formation and transactive memory can build on this study. Organizational demographic variables, such as the size of the organization or subunit and the rates of worker unavailability and turnover, are observable and measurable constructs. ASV can be measured as Hamming distance, or by applying string matching (Abbott, 1995; Pentland, 2003) if task sequences vary in length across performances of a routine. Even though transactive memory and skill distribution are not directly observable, researchers should be able to design questionnaires or interview protocols to inventory individuals’ skill sets as well as transactive knowledge. Researchers studying groups have developed methods for assessing transactive memory (see Austin, 2003; Lewis, 2004). Settings where organizations face recurring similar problems, such as invoice processing (Pentland et al., 2011), are good contexts to test and extend our theory and findings.

Conclusion

This study offers an original model specifying key variables affecting how organizational routines emerge and how the associated pattern of actors changes over time. Our study highlights the role of transactive memory in forming routines. We responded to calls to pay greater attention to the individual actors who perform routines in organizations (Feldman and Rafaeli, 2002; Felin and Foss, 2005) and their relationships (Cohendet and Llerena, 2008; Dionysiou and Tsoukas, 2013; Turner and Rindova, 2012), and to transactive memory formation as an aspect of organizational learning (Borgatti and Cross, 2003; Palazzolo et al., 2006, Ren et al., 2006). In our model, transactive memory is the key mechanism linking individual skills to collective organizational capabilities.

Even when an organization carries out a repetitive sequence of actions, the sequence of actors can change continuously. Viewing routines from an actor-sequence perspective complements the prevailing emphasis on action sequences in routines research. Our theoretical arguments and model findings illustrate how variations in the efficiency of routines that are inexplicable in terms of action sequences may be readily explained when we examine actor sequences. We hope that our model and findings encourage further theory development integrating research on transactive memory to explain routine formation and change.

Footnotes

Acknowledgements

We thank Daniel Beal, Teppo Felin, Henrich Greve, Daniel Levinthal, Brent Scott, Raymond Sparrowe, and Scott Turner for helpful comments on earlier drafts.

Funding

This material is based upon work supported by the National Science Foundation under grant no. SES-0924786 and Ewha Womans University research grant no. 1-2012-2057-001-1.

Notes

Author biographies

Kent D Miller is a Professor of Management at the Eli Broad Graduate School of Management, Michigan State University (millerk@broad.msu.edu). His current research examines organizational learning and change, as well as methodological and philosophical issues in management research. In recent years, his research articles have appeared in Academy of Management Journal, Journal of Management Studies, Organization Science, Organization Studies, and Strategic Management Journal. He received his PhD from the University of Minnesota.

Seungho Choi is an Assistant Professor in the Department of Business Administration at Ewha Womans University (choise@ewha.ac.kr). He received his PhD in Strategic Management from Michigan State University. His current research examines knowledge combination, ongoing innovation, organizational learning, and technology acquisition. He recently published “Dynamics of Performing and Remembering Organizational Routines” in Journal of Management Studies.

Brian T Pentland is a Professor in the Department of Accounting and Information Systems at Michigan State University (pentland@broad.msu.edu). His current research is focused on the relationship between work and technology, especially in the context of organizational routines. He has also been developing methods for representing and analyzing patterns of action within work processes. His most recent publications have appeared in the Journal of Management Studies, Organization Science, and YouTube. He received his PhD in Management from the Massachusetts Institute of Technology in 1991 and SB in Mechanical Engineering from the Massachusetts Institute of Technology in 1981.

References

Abbott

(1995) “Sequence Analysis: New Methods for Old Ideas,” Annual Review of Sociology 21: 93–113.

Abell

Felin

Foss

(2008) “Building Micro-Foundations for the Routines, Capabilities, and Performance Links,” Managerial and Decision Economics 29: 489–502.

Aime

Johnson

Ridge

J. W.

Hill

A. D.

(2010) “The Routine May Be Stable but the Advantage Is not: Competitive Implications of Key Employee Mobility,” Strategic Management Journal 31: 75–87.

Argote

Ren

(2012) “Transactive Memory Systems: A Microfoundation of Dynamic Capabilities,” Journal of Management Studies 49: 1375–82.

Argote

Insko

C. A.

Yovetich

Romero

A. A.

(1995) “Group Learning Curves: The Effects of Turnover and Task Complexity on Group Performance,” Journal of Applied Social Psychology 25: 512–29.

Austin

J. R.

(2003) “Transactive Memory in Organizational Groups: The Effects of Content, Consensus, Specialization, and Accuracy on Group Performance,” Journal of Applied Psychology 88: 866–78.

Bechky

B. A.

(2006) “Gaffers, Gofers, and Grips: Role-Based Coordination in Temporary Organizations,” Organization Science 17: 3–21.

Becker

M. C.

(2005) “The Concept of Routines: Some Clarifications,” Cambridge Journal of Economics 29: 249–62.

Borgatti

S. P.

Cross

(2003) “A Relational View of Information Seeking and Learning in Social Networks,” Management Science 49: 432–45.

10.

Borgatti

S. P.

Foster

P. C.

(2003) “The Network Paradigm in Organizational Research: A Review and Typology,” Journal of Management 23: 991–1013.

11.

Brandon

D. P.

Hollingshead

A. B.

(2004) “Transactive Memory Systems in Organizations: Matching Tasks, Expertise and People,” Organization Science 15: 633–44.

12.

Bruderer

Singh

J. V.

(1996) “Organizational Evolution, Learning, and Selection: A Genetic-Algorithm-Based Model,” Academy of Management Journal 39: 1322–49.

13.

Burt

R. S.

(1982) Toward A Structural Theory of Action: Network Models of Social Structure, Perception, and Action. New York: Academic Press.

14.

Burton

R. M.

(2003) “Computational Laboratories for Organization Science: Questions, Validity and Docking,” Computational and Mathematical Organization Theory 2: 91–108.

15.

Burton

R. M.

Obel

(1995) “The Validity of Computational Models in Organization Science: From Model Realism to Purpose of the Model,” Computational and Mathematical Organization Theory 1: 57–71.

16.

Carley

(1991) “A Theory of Group Stability,” American Sociological Review 56: 331–54.

17.

Carley

(1992) “Organizational Learning and Personnel Turnover,” Organization Science 3: 20–46.

18.

Carley

(1999) “On the Evolution of Social and Organizational Networks,” in Bacharach

S. B.

Andrews

S. B.

Knoke

(eds) Research in the Sociology of Organizations, vol. 16, pp. 3–30. Stamford, CT: JAI Press.

19.

Cohen

M. D.

Bacdayan

(1994) “Organizational Routines Are Stored as Procedural Memory: Evidence from A Laboratory Study,” Organization Science 5: 554–68.

20.

Cohen

M. D.

Burkhart

Dosi

Egidi

Marengo

Warglien

Winter

(1996) “Routines and Other Recurring Action Patterns of Organizations: Contemporary Research Issues,” Industrial and Corporate Change 5: 653–98.

21.

Cohendet

Llerena

(2008) “The Role of Teams and Communities in the Emergence of Organizational Routines,” in Becker

M. C.

(ed.) Handbook of Organizational Routines, pp. 256–77. Cheltenham: Edward Elgar.

22.

Cross

Sproull

(2004) “More than an Answer: Information Relationships for Actionable Knowledge,” Organization Science 15: 446–62.

23.

Crowston

(1996) “An Approach to Evolving Novel Organizational Forms,” Computational and Mathematical Organization Theory 2: 29–47.

24.

Dionysiou

Tsoukas

(2013) “Understanding the (Re)Creation of Routines from within: A Symbolic Interactionist Perspective,” Academy of Management Review 38: 181–205.

25.

Feldman

M. S.

(2000) “Organizational Routines as a Source of Continuous Change,” Organization Science 11: 611–29.

26.

Feldman

M. S.

(2003) “A Performative Perspective on Stability and Change in Organizational Routines,” Industrial and Corporate Change 12: 727–52.

27.

Feldman

M. S.

Pentland

B. T.

(2003) “Reconceptualizing Organizational Routines as a Source of Flexibility and Change,” Administrative Science Quarterly 48: 94–118.

28.

Feldman

M. S.

Rafaeli

(2002) “Organizational Routines as Sources of Connections and Understandings,” Journal of Management Studies 39: 309–31.

29.

Felin

Foss

N. J.

(2005) “Strategic Organization: A Field in Search of Micro-Foundations,” Strategic Organization 3: 441–55.

30.

Felin

Foss

N. J.

Heimeriks

K. H.

Madsen

T. L.

(2012) “Microfoundations of Routines and Capabilities: Individuals, Processes, and Structure,” Journal of Management Studies 49: 1351–74.

31.

Gavetti

Levinthal

(2000) “Looking Forward and Looking Backward: Cognitive and Experiential Search,” Administrative Science Quarterly 45: 113–37.

32.

Gavetti

Levinthal

Rivkin

J. W.

(2005) “Strategy Making in Novel and Complex Worlds: The Power of Analogy,” Strategic Management Journal 26: 691–712.

33.

Gupta

Hollingshead

A. B.

(2010) “Differentiated versus Integrated Transactive Memory Effectiveness: It Depends on the Task,” Group Dynamics: Theory, Research, and Practice 14: 384–98.

34.

Hodgson

G. M.

(2003) “The Mystery of the Routine: The Darwinian Destiny of an Evolutionary Theory of Economic Change,” Revue Economique 54: 355–84.

35.

Hollingshead

A. B.

(1998) “Communication, Learning, and Retrieval in Transactive Memory Systems,” Journal of Experimental Social Psychology 34: 423–42.

36.

Hollingshead

A. B.

(2001) “Cognitive Interdependence and Convergent Expectations in Transactive Memory,” Journal of Personality and Social Psychology 81: 1080–9.

37.

Hutchins

(1995) Cognition in the Wild. Cambridge, MA: MIT Press.

38.

Lazaric

(2008) “Routines and Routinization: An Exploration of Some Micro-Cognitive Foundations,” in Becker

M. C.

(ed.) Handbook of Organizational Routines, pp. 205–27. Cheltenham: Edward Elgar.

39.

Lazaric

Denis

(2005) “Routinisation and Memorisation of Tasks within a Workshop: An Illustration through a Case Study,” Industrial and Corporate Change 14: 873–96.

40.

Levinthal

D. A.

Warglien

(1999) “Landscape Design: Designing for Local Action in Complex Worlds,” Organization Science 10: 342–57.

41.

Levitt

March

J. G.

(1988) “Organizational Learning,” Annual Review of Sociology 14: 319–40.

42.

Lewis

(2004) “Knowledge and Performance in Knowledge-Worker Team: A Longitudinal Study of Transactive Memory Systems,” Management Science 50: 1519–33.

43.

Liang

D. W.

Moreland

Argote

(1995) “Group versus Individual Training and Group Performance: The Mediating Role of Transactive Memory,” Personality and Social Psychology Bulletin 21: 384–93.

44.

Lin

(2001) Social Capital: A Theory of Social Structure and Action. London: Cambridge University Press.

45.

Lindkvist

(2004) “Governing Project-based Firms: Promoting Market-like Processes within Hierarchies,” Journal of Management and Governance 8: 3–25.

46.

Lindkvist

(2005) “Knowledge Communities and Knowledge Collectivities: A Typology of Knowledge Work in Groups,” Journal of Management Studies 42: 1189–210.

47.

Littlepage

Robison

Reddington

(1997) “Effects of Task Experience and Group Experience on Group Performance, Member Ability, and Recognition of Expertise,” Organizational Behavior and Human Decision Processes 69: 133–47.

48.

Majchrzak

Jarvenpaa

S. L.

Hollingshead

A. B.

(2007) “Coordinating Expertise among Emergent Groups Responding to Disasters,” Organization Science 18: 147–61.

49.

March

J. G.

(1991) “Exploration and Exploitation in Organizational Learning,” Organization Science 2: 71–87.

50.

March

J. G.

Simon

H. A.

(1958) Organizations. New York: John Wiley & Sons.

51.

Michel

A. A.

(2007) “A Distributed Cognition Perspective on Newcomers’ Change Processes: The Management of Cognitive Uncertainty in Two Investment Banks,” Administrative Science Quarterly 52: 507–57.

52.

Midgley

Marks

Kunchamwar

(2007) “Building and Assurance of Agent-Based Models: An Example and Challenge to the Field,” Journal of Business Research 60: 884–93.

53.

Miller

K. D.

Pentland

B. T.

Choi

(2012) “Dynamics of Performing and Remembering Organizational Routines,” Journal of Management Studies 49: 1536–58.

54.

Moorman

Miner

A. S.

(1998) “Organizational Improvisation and Organizational Memory,” Academy of Management Review 23: 698–723.

55.

Moreland

R. L.

Argote

Krishnan

(1996) “Socially Shared Cognition at Work: Transactive Memory and Group Performance,” in Nye

J. L.

Brower

A. M.

(eds) What’s Social about Social Cognition? Research on Socially Shared Cognition in Small Groups, pp. 57–84. Thousand Oaks, CA: Sage.

56.

Nebus

(2006) “Building Collegial Information Networks: A Theory of Advice Network Generation,” Academy of Management Review 31: 615–37.

57.

Nelson

S. G.

Winter

R. R.

(1982) An Evolutionary Theory of Economic Change. Cambridge, MA: Harvard University Press.

58.

Palazzolo

E. T.

Serb

D. A.

She

Contractor

N. S.

(2006) “Coevolution of Communication Knowledge Networks in Transactive Memory Systems: Using Computational Models for Theoretical Development,” Communication Theory 16: 223–50.

59.

Pentland

B. T.

(2003) “Sequential Variety in Work Processes,” Organization Science 14: 528–40.

60.

Pentland

B. T.

Hærem

Hillison

(2011) “The (N)Ever-Changing World: Stability and Change in Organizational Routines,” Organization Science 22: 1369–83.

61.

Rao

R. D.

Argote

(2006) “Organizational Learning and Forgetting: The Effects of Turnover and Structure,” European Management Review 3: 77–85.

62.

Reagans

Argote

Brooks

(2005) “Individual Experience and Experience Working Together: Predicting Learning Rates from Knowing Who Knows What and Knowing How to Work Together,” Management Science 51: 869–81.

63.

Ren

Carley

K. M.

Argote

(2006) “The Contingent Effects of Transactive Memory: When Is It More Beneficial to Know What Others Know?” Management Science 52: 671–82.

64.

Rivkin

J. W.

(2001) “Reproducing Knowledge: Replication without Imitation at Moderate Complexity,” Organization Science 12: 274–93.

65.

Steen

Coopmans

Whyte

(2006) “Structure and Agency? Actor-Network Theory and Strategic Organization,” Strategic Organization 4: 303–12.

66.

Thompson

J. D.

(1967) Organizations in Action. New York: McGraw-Hill.

67.

Ton

Huckman

R. S.

(2008) “Managing the Impact of Employee Turnover on Performance: The Role of Process Conformance,” Organization Science 19: 56–68.

68.

Tsoukas

Chia

(2002) “Organizational Becoming: Rethinking Organizational Change,” Organization Science 12: 567–82.

69.

Turner

S. F.

Rindova

(2012) “A Balancing Act: How Organizations Pursue Consistency in Routine Functioning in the Face of Ongoing Change,” Organization Science 23: 24–46.

70.

Van der Vegt

Bunderson

Kuipers

(2010) “Why Turnover Matters in Self-Managing Work Teams: Learning, Social Integration, and Task Flexibility,” Journal of Management 36: 1168–91.

71.

Wegner

D. M.

(1987) “Transactive Memory: A Contemporary Analysis of the Group Mind,” in Mullen

Goethals

G. R.

(eds) Theories of Group Behavior, pp. 185–208. New York: Springer-Verlag.

72.

Weick

K. E.

Roberts

K. H.

(1993) “Collective Mind in Organizations: Heedful Interrelating on Flight Decks,” Administrative Science Quarterly 38: 357–81.

73.

Winter

S. G.

Szulanski

(2001) “Replication as Strategy,” Organization Science 12: 730–43.

74.

Zoethout

Jager

Molleman

(2006a) “Formalizing Self-Organizing Processes of Task Allocation,” Simulation Modelling Practice and Theory 14: 342–59.

75.

Zoethout

Jager

Molleman

(2006b) “Simulating the Emergence of Task Rotation,” JASSS: The Journal of Artificial Societies and Social Simulation 9(1). Available at: http://jasss.soc.surrey.ac.uk/9/1/5.html

76.

Zoethout

Jager

Molleman

(2007) “Newcomers in Self-Organising Task Groups: A Pilot Study,” in Takahashi

Sallach

Rouchier

(eds) Advancing Social Simulation: The First World Conference, pp. 213–24. Tokyo, Japan: Springer.

77.

Zoethout

Jager

Molleman

(2008) “Task Dynamics in Self-Organising Task Groups: Expertise, Motivational, and Performance Differences of Specialists and Generalists,” Autonomous Agents and Multi-Agent Systems 16: 75–94.

78.

Zoethout

Jager

Molleman

(2010) “When Does A Newcomer Contribute to A Better Performance? A Multi-Agent Study on Self-Organising Processes of Task Allocation,” JASSS: The Journal of Artificial Societies and Social Simulation 13(3). Available at: http://jasss.soc.surrey.ac.uk/13/3/7.html