2. Let us focus on real routines

Without question, the term ‘organizational routine’ refers to a complicated phenomenon. As Cohen (Reference Cohen2007) notes, the concept is widely misunderstood. To avoid confusion, it is useful to consider a concrete example, such as invoice processing. Pentland et al. (Reference Pentland, Feldman, Becker and Liu2010) studied invoice processing in four small Norwegian organizations, all of which used the same workflow technology, as described here:

Invoice processing satisfies the definition of an organizational routine: it has ‘repetitive, recognizable patterns of interdependent actions, carried out by multiple actors’ (Feldman and Pentland, Reference Feldman and Pentland2003: 93). It satisfies each part of the definition:

This definition provides an easily applicable test of whether a given phenomenon qualifies as an organizational routine. If any piece of the definition is absent, the phenomenon is not an organizational routine.

Readers who prefer a different example might like to consider Target the Two, the experimental card game used by Cohen and Bacdayan (Reference Cohen and Bacdayan1994) and subsequently analyzed by Egidi and Narduzzo (Reference Egidi and Narduzzo1997). Or you may prefer to think of your own examples, from personal experience, popular press, or published research. Any example will do. The key to avoiding Wittgenstein's trap is to focus on the phenomenon.

3. What is the problem?

With the phenomenon clearly defined and some concrete examples fresh in our minds, let us consider the central issue raised by Felin and Foss:

To address this concern, Felin and Foss offer the poverty of stimulus argument, which they have borrowed from Chomskian linguistics. They introduce this ‘in hopes of re-directing extant work on organizational routines and capabilities’. Since this argument forms the core of their paper, it is worth examining it in some detail. There are several problems that readers should take into consideration in evaluating Felin and Foss's use of this argument.

First, Felin and Foss characterize the logic of this argument incorrectly, mainly by exaggerating it. They state:

A more accurate statement of the poverty of stimulus argument would be as follows:

It is clear from this statement that linguists do not assert that the environment ‘does not in any meaningful way contain lessons learned’. Quite the contrary – it contains lots of lessons – but linguistic nativists (such as Chomsky) believe that those lessons are insufficient to explain certain specific aspects of language acquisition.

Second, it is important to emphasize that the poverty of stimulus is an argument about language acquisition, not about organizations, routines or capabilities. Felin and Foss are attempting to make meta-theoretical, level-independent claims, but the basis for their argument comes from a very specific aspect of human child development. To generalize from language acquisition in two-year-olds to new product development in auto companies or mortgage origination in banks, we need to accept the implicit analogy that banks and auto companies are like two-year-olds. This seems like a stretch. While metaphor and analogy can be a useful source of theory, Tsoukas (Reference Tsoukas1991) cautions that one must be careful about constructing analogies between domains.

The analogy between routines and Chomskian universal grammar was analyzed by Pentland (Reference Pentland1995). While some aspects of the analogy are fruitful, Pentland (Reference Pentland1995) concludes that nativist ideas such as universal grammar and ‘deep structure’ have no meaningful analogy in organizations. Organizations are filled with structures, of course, but they are always historically and culturally contingent, so the analogy to Chomskian universal grammar (which is claimed to be biological in origin) is misleading. This distinction is critical here because in linguistics, the poverty of stimulus argument is specifically intended to support the claim that universal grammar is innate and biological in origin.

Finally, even in linguistics, the poverty of stimulus argument is at best controversial and probably incorrect. Parents will always be impressed by the remarkable things that their children can say, but the nativist position described by Felin and Foss is not supported by neurological, computational or linguistic evidence. Writing in The Linguistic Review, Pullum and Scholz (Reference Pullum and Scholz2002) offer a rigorous examination of the evidence for and against this argument. They examine various versions of this argument and conclude that none of them holds up to scrutiny. Further, they argue that additional research on the topic is likely to ‘further undermine the linguistic nativist position’. Far from being some kind of intellectual bedrock upon which a new theory of routines can safely be built, the poverty of stimulus argument is a controversial conjecture that is quite possibly wrong, even in its source domain.

Nevertheless, Felin and Foss suggest that this conjecture should be used to re-direct research on organizational routines. They argue that it would provide an antidote to what they see as an excessively behaviorist emphasis on experience and empirical data in our theories and in our research. While experience has always been seen as important in forming routines, Winter (forthcoming) does a very convincing job of demonstrating that strict behaviorism has never been the foundation for any theory of routines. In short, Felin and Foss appear to be advocating the application of a misleading, erroneous conjecture to solve a problem that never existed.

Rather than engaging the details of an argument whose premise is fatally flawed, it may be more productive to direct interested readers to an alternative perspective. While the ideas outlined here are just one of many possible perspectives, they do address some of the issues raised by Felin and Foss.

4. Current foundations

Some recent theories of organizational routines have been grounded in theories of structuration (Giddens, Reference Giddens1984) and practice (Bourdieu, Reference Bourdieu1990). Routinization is at the core of structuration theory (Giddens, Reference Giddens1984). Likewise, Bourdieu's (Reference Bourdieu1990) concept of practice centers on the repetitive, patterned activities that constitute habitus. These theories use different vocabulary and emphasize different things, but they have made a fruitful foundation for a theory of organizational routines (e.g., Pentland and Rueter, Reference Pentland and Rueter1994; Feldman, Reference Feldman2000, Reference Feldman2003; Feldman and Rafaeli, Reference Feldman and Rafaeli2002; Feldman and Pentland, Reference Feldman and Pentland2003; Pentland and Feldman, Reference Pentland and Feldman2005). This theoretical view is successful because it is supported by detailed ethnographic observation of a wide range of actual work practices, such as filing papers in an office (Suchman, Reference Suchman1983), photocopier repair (Orr, Reference Orr1996), medical imaging (Barley, Reference Barley1986), technology roadmapping (Howard-Grenville, Reference Howard-Grenville2005), software support (Pentland and Rueter, Reference Pentland and Rueter1994), navigating a ship by sight (Hutchins, Reference Hutchins1991), and many others.

In light of this corpus of research, it is difficult to understand how Felin and Foss can argue that the extant literature on routines is based on simple behaviorism. For example, Feldman and Pentland (Reference Feldman and Pentland2003) show that subjectivity, agency and power are essential to organizational routines. What counts as a routine – the boundaries, the patterns of action, and so on – is inherently subjective and agentic (Emirbayers and Mische, Reference Emirbayers and Mische1998). Feldman and Pentland (Reference Feldman and Pentland2003) use the term ‘ostensive aspects’ to refer to the abstract boundaries and patterns underlying every routine. The term is plural because organizational routines involve multiple actants whose perspective may differ. For example, one of the abstract patterns for invoice processing can be described as: ‘enter the invoices, then approve them, then pay them’. Particular people who are only involved in payment or approval may have a different understanding of ‘invoice processing’. In any case, without an abstract pattern or patterns, routines cannot be recognized or reproduced.

5. Necessary and sufficient conditions

Felin and Foss (Reference Felin and Foss2011) are concerned that experience alone cannot account for an organizational routine. This is surely true, but this problem has been specifically addressed by existing theory. In the practice-based view articulated by Feldman and Pentland, routines have two aspects that exist as a duality:

If we examine actual routines, it is easy to see that abstract patterns are necessary, but not sufficient. Experience is necessary, but not sufficient. Together, they are necessary and sufficient.

For example, in invoice processing, the abstract pattern (enter, approve, pay) is an essential aspect of the work. At the same time, the work cannot be performed at all without some experience or training. Similarly, one cannot play Cohen and Bacdayan's (1994) card game, Target the Two, without knowing the rules. But the play does not become routinized unless the players have some experience. Skeptical readers are invited to try and think of counterexamples, but I do not think that there are any. This is because, as current theory tells us, these two aspects are inherently present in every routine.

6. So where do routines come from?

Felin and Foss have argued that there is no satisfactory explanation in the literature of how routines are formed. I respectfully disagree. It is difficult to generalize about all routines, but it is easy to explain any particular routine. For example, in the Norwegian organizations mentioned above, the design of their invoice-processing routine came from Compello Software (Fornebu, Norway), the vendor that sells the invoice-processing workflow system. The particular pattern mentioned earlier (enter, approve, then pay) comes from decades of institutionalized accounting practice for the prevention of fraud.

Real routines are usually formed for simple, obvious reasons. It is hard to make a single, universal statement that covers every case, but here are some very common scenarios:

While the details differ considerably, these explanations share one basic idea: routines form in context (Cohen et al., Reference Cohen, Burkhart, Dosi, Egidi, Marengo, Warglien and Winter1996). In any given context, when people need to get their work done, they adopt whatever strategies (patterns of action) that seem best, and adapt them as necessary. For the people involved, there are no deep philosophical issues, and no shaky foundations. Of course, for any routine, historical research could be conducted to learn more about its origin, and in principle, we can take that back as far in time as we want (Inca knot tying? Heterogeneity after the Big Bang?). The search for historical origins is a worthy pursuit, but, in practice, the origins of routines are not problematic.

7. Routines form regardless of where experience comes from

Some of the issues raised by Felin and Foss have been addressed in recent computer simulations. Simulations are used to create testable hypotheses and to guide theorizing. We can run many thousands of different scenarios – far more than would be possible in field research. These particular simulations model the formation and dynamics of organizational routines (Miller et al., Reference Miller, Pentland and Choi2010) and path dependence in repetitive patterns of action (Pentland et al., Reference Pentland, Haerem and Hillison2010).

In one simulation (Miller et al., Reference Miller, Pentland and Choi2010), an organization is subjected to a stream of problems – totally exogenous – that can be varied from perfectly identical (same problem every time) to perfectly random. This is a traditional agent-based simulation, where agents with heterogeneous skills collaborate to solve the stream of problems. In another simulation (Pentland et al., Reference Pentland, Haerem and Hillison2010), the stream of actions is completely endogenous. This simulation uses a very different kind of model, where individual agents are not explicitly represented. Rather, the simulated organization generates a stream of actions from an abstract pattern. Initially, this pattern can be completely random (a ‘clean slate’) or it may be some a priori structure. That stream of actions is then subject to random variations that range from zero to 100%. The (partially randomized) actions then form the basis for the subsequent patterns.

In both simulations, the model forms an abstract pattern for the routine that is a simple function of the stream of experiences. Among other things, the results show that any source of experience – whether it is endogenous or exogenous, whether it is totally random or highly structured – can result in the formation of routines. Repetitive, recognizable patterns of action will form as long as the organization has some memory, so that past actions can be repeated.

These simulations provide another important point of reference with respect to Felin and Foss's argument. In each of these simulations, there are no a priori internal structures that represent the patterns of action. The software has algorithms that model certain kinds of behavior, but at the start of the simulation, they are truly tabula rasa concerning the overall pattern of the routine (more so than any real organization ever could be). Yet, when given a repetitive stream of problems, routines form. These results directly contradict the poverty of stimulus argument as articulated by Felin and Foss.

Still, Felin and Foss raise a good point: it is important to understand how organizational routines form, and more research is needed into this topic. However, for scholars that are interested in addressing this question, it is important to avoid confusion and distraction. The current paper by Felin and Foss (Reference Felin and Foss2011) asks a question that seems important (‘What are the origins of heterogeneous experience. . .?’). Our simulation results suggest that this question is probably irrelevant to the real, substantive question of how routines form.

Of course, simulations do not constitute proof, so it is best to consider actual empirical examples. For example, in Target the Two (Cohen and Bacdayan, Reference Cohen and Bacdayan1994), the experience comes from shuffling the deck of cards and the actions of the other player. For the invoice routines, it comes from the vendors. Readers are invited to think of examples of routines where the source of experience cannot be identified, or where that experience does not contribute to the ongoing conduct of the routine in some way.

8. Routines are sociomaterial

The most serious flaw in the current Felin and Foss paper (Reference Felin and Foss2011), and in their so-called ‘microfoundations project’ in general, is that it fails to incorporate current thinking about the ontology of organizational routines.

First of all, by definition, organizational routines are not individual-level phenomena. They involve patterns of interdependent actions carried out by multiple actors. The involvement of multiple actors is not a recent addition to the idea of routines. Consider, for example, the discussion of routines as truce in Nelson and Winter (Reference Nelson and Winter1982). ‘Truce’ is not an individual-level concept; it is inherently relational. Yet, as Zbaracki and Bergen (Reference Zbaracki and Bergen2010) show in their field study of a sales process, when truces break down, so do routines.

Second, routines are not exclusively a function of human psychology. While research has shown the importance of psychological phenomena, such as declarative and procedural memory (Cohen and Bacdayan, Reference Cohen and Bacdayan1994; Lazaric and Denis, Reference Lazaric and Denis2005; Lazaric, Reference Lazaric and Becker2008), Schulz (Reference Schulz and Becker2008) offers a long list of factors that explain why routines ‘stay on track’, only some of which are psychological. Other factors include physical artifacts, institutionalization and coercion. Or if we return to Giddens (Reference Giddens1984) for guidance, we would need to consider signification, legitimation and domination (see, for example, Feldman, Reference Feldman2003).

Third, while it may be tempting to look for ‘first principles’ at the level of individual psychology, it is important to realize that relational phenomena (e.g., network ties) are also first principles. In the same way that Chomsky demonstrated the limits of behaviorism in the early 1960s, network theorists have effectively demonstrated the limits of methodological individualism since at least the early 1980s (Mayhew, Reference Mayhew1980). And since organizational routines are defined as patterns of interdependent action, this includes the relations between actions, not just relations between actors.

But the larger concern here is the implicit notion that humans are the only relevant actants. Winter (forthcoming) jokes about the need to specify H. sapiens in our theories, but in typical routines, many actants are not human. Consider the invoice-processing routine described above – the computerized workflow system itself performed many of the actions. Even Cohen and Bacdayan's (Reference Cohen and Bacdayan1994) Target the Two requires two players, plus a deck of cards. It is easy to overlook, but without the cards (a physical artifact), the game cannot be played. When Felin and Foss make appeals to ‘human nature’ (e.g., linguistic nativism) as a key to explaining routines, it seems like a quaint anachronism.

Contemporary scholars have adopted the term ‘sociomateriality’ to refer to this phenomenon (Orlikowski, Reference Orlikowski2007; Leonardi and Barley, Reference Leonardi and Barley2008). Building on the work of Latour (Reference Latour2005), Law (Reference Law2004) and others, Orlikowski has argued that:

As Winter (Reference Winter2011) notes, strict behaviorism is not a good foundation for this kind of phenomenon, but the alternative suggested by Felin and Foss seems worse. To see how this is so, consider some current empirical research on routines.

9. Implications for empirical research

Felin and Foss want to shape future research on routines so that it is less behavioristic. Towards that end, they argue that repetition and experience should be treated as ‘epiphenomenal’. In contrast, in practice-based theory described here, repetition and experience are almost definitional of the phenomenon. And contrary to the impression left by Felin and Foss, empirical research that embraces repetition and experience has been making progress.

Consider, for example, several recent studies that seek to explain how routines form and change. Zbaracki and Bergen (Reference Zbaracki and Bergen2010) use longitudinal fieldwork to show how pricing routines depend on maintaining ‘truces’ between different functional areas in an organization. When the political relationships within the organization break down, the routines can no longer be maintained. Goh et al. (forthcoming) use observational data to examine the evolution of routines that form when new technology is introduced in a hospital. Their findings highlight the co-evolution of the patterns of action and the functional affordances of the technology. Leonardi (forthcoming) uses data collected through ethnographic observation of automotive design teams to unpack the relationship between technical artifacts and routines. Pentland et al. (forthcoming) use archival data to study endogenous change in the invoice-processing routines described above. They find that in some circumstances, increased automation increases the variability in the patterns of action, while, in other circumstances, automation decreases variability. Rerup and Feldman (forthcoming) use a combination of fieldwork and archival sources to show how repetitive trial and error influence the formation of organizational routines and schemata. Their analysis provides a particularly direct illustration of why the poverty of stimulus argument makes no sense when applied to organizations, because (among other things) the focal organization (Learning Lab Denmark) constructs its own ‘stimulus’ through trial and error.

All of these studies use longitudinal data about repetition and experience to develop theoretical explanations of how routines form and change over time. As a group, these studies clearly demonstrate the sociomateriality of routines. And in each of these studies, the main source of data consists of the specific actions taken by participants as they do their work. This is what Feldman and Pentland (Reference Feldman and Pentland2003) call the ‘performative’ aspect of the routine. And while they focus on actions, none of these studies could be construed as ‘behaviorist’. However, because these studies are based on ‘epiphenomenal’ aspects of the routines (repetition and experience), they exemplify the kind of research that Felin and Foss find problematic and apparently want to discourage.

The idea that empirical data on performances of routines is an inappropriate basis for research on routines illustrates Tsoukas's (Reference Tsoukas1991) point about the need to be careful when using analogies to build theory. Felin and Foss's error can be traced back to the analogy that underlies their use of the poverty of stimulus argument. To see how this is so, readers need to understand how classical nativist grammarians think about data.

In linguistics, grammarians such as Chomsky have often argued that empirical data (e.g., actual spoken language) are epiphenomenal. This is because people utter a lot of things for a lot of reasons that have nothing to do with universal grammar. So, for that research community (especially in the 1960s and 1970s), intuitive judgment about the correctness of a sentence was considered better evidence about universal grammar than the utterances that people make in normal conversation (Newmeyer, Reference Newmeyer1983).

When they invoke the poverty of stimulus argument, Felin and Foss are implying that organizations are like two-year-olds. On the basis of this implicit analogy, they seem to be suggesting that organizational research should adopt the same epistemology as nativist linguists from the 1970s. This might make sense if the underlying analogy made sense, but organizations are not like two-year-olds. There is no universal grammar for routines (Pentland, Reference Pentland1995), and there is no poverty of stimulus for organizations.

10. Let us focus on real problems

To summarize, organizational routines are a well-defined phenomenon: repetitive, recognizable patterns of interdependent actions, carried out by multiple actors, not all of which are human. Current theory emphasizes that, in practice, routines exist as a duality of ostensive and performative aspects (or, if you prefer, abstract patterns and specific actions). And while each routine has its own unique history, they are all formed in context.

While the phenomenon is complex, we have started to make progress on modeling routines using network methods (Pentland, Reference Pentland, Baum and McKelvey1999; Pentland and Feldman, Reference Pentland and Feldman2007; Pentland et al., Reference Pentland, Haerem and Hillison2010; Goh et al., forthcoming). Networks are a kind of grammar (Chomsky, Reference Chomsky1956), but they allow us to express the ‘pattern-in-variety’ that is inherent to all routines (Cohen, Reference Cohen2007) more succinctly than other kinds of grammar. They allow us to compare routines and measure changes over time; and they provide a simple, rigorous model of the latent dispositions. The new computerized sources of data, such as workflow event logs, make it possible to observe real patterns of action in ways that were never possible before. Thus, the field is poised to make some significant progress. If we focus attention on actual routines and current theory, important research issues are easy to find.