1

THE DYNAMICS OF COORDINATION REGIMES:

IMPLICATIONS FOR ORGANIZATIONAL DESIGN

Phanish Puranam

And

Michael G. Jacobides

London Business School University of London

Sainsbury 317, Sussex Place Regent’s Park, London NW1 4SA U.K.

Tel: + 44 (0) 20 7262 5050 Fax: + 44 (0) 20 7724 7875

ppuranam@london.edu

This Draft – April 2006

We thank Martin Kunc for assistance with simulation models.

2

To the extent that contingencies arise, not anticipated in the schedule, coordination requires communication to give notice of deviations from planned or predicted conditions, or to give instructions for changes in activity to adjust to these deviations. We may label coordination based on pre-established schedules coordination by plan, and coordination that involves transmission of new information coordination by feedback. [March and Simon, 1958; p182]

The idea that coordination – the alignment of actions among interdependent

actors (Gulati, Lawrence& Puranam, 2005; Heath & Staudenmayer, 2000) can occur

through two distinct “regimes” - either through ongoing communication and mutual

adjustment or through predefined plans and standards - is hardly novel or exciting in

itself. However it invites intriguing speculations when coupled with the insight that

organizational boundaries often divide activities that are coordinated through plans

and schedules, but enclose activities that require feedback and mutual adjustment to

be coordinated with each other (Thompson, 1967). Arguments that link optimal

organizational forms to the nature of coordination requirements have been the

mainstay of structural contingency theory (Barnard, 1938; March and Simon, 1958;

Lawrence and Lorsch, 1967; Thompson, 1967) and under a broader definition of

coordination requirements, indeed also of transaction cost theorizing (Coase, 1937;

Williamson, 1976; 1985).

These ideas have of late been rejuvenated by the contributions of several

scholars who have used the principles of modularity to understand the internal

organization and boundaries of firms (Baldwin & Clark, 2000; Ethiraj & Levinthal,

2004; Galunic & Eisenhardt, 2001a; Helfat & Eisenhardt, 2000; Langlois, 2001;

Sanchez & Mahoney, 1996; Schilling, 2000). Within this literature, it is often

suggested that modular interfaces that partition complex activity systems into parts

with dense interdependencies within themselves but relatively fewer

interdependencies between them, can also function as organizational boundaries. In

3

particular, there are two ways in which this idea is presented in the literature. The first

proposes a link between product and organizational architecture (Baldwin & Clark,

2000; Puranam, 2001; Sanchez & Mahoney, 1996; Schilling & Steensma, 2001),

whereas the second extends this link to ownership (Hoetker, 2002; Schilling &

Steensma, 2001), suggesting that modular production enables market based linkages

between activities.

The proposition that product level modularity can influence modularity in

organization and ownership is a valuable contribution to the study of organizations,

and refocuses our attention towards the role of organizational and ownership

boundaries in demarcating distinct regimes of coordination. While undoubtedly

useful, this line of thinking inherits some of the methodological weaknesses identified

with older structural contingency theorizing, as well as generates some new points of

contention. In our view, there are two critical questions that the literature on

organizational modularity must engage with in order to contribute effectively to our

understanding of organization design.

First, what is the link between organizational and product interfaces? The

carry-over of ideas about technological interfaces to organizational interfaces usually

occurs with a corresponding assumption that interfaces either exist and partition

activity completely, or do not. In contrast, in the tradition of organization design, the

division of labor has not (necessarily) been equated with standardized interfaces and

simplified interactions between the components of labor so divided. For instance,

Lawrence and Lorsch (1967) make clear that task decomposition and the creation of

modules (differentiation) goes hand in hand with mechanisms for coordination that

penetrate the boundaries of the resulting modules (integration). In the context of inter-

firm relationships, a number of authors have commented on the existence of fairly

4

thick forms of coordination between as well as within firms – far from being

partitioned off from each other through thin interfaces, there may often be substantial

permeability of firm boundaries (Bradach & Eccles, 1989; Grandori, 1997; Gulati,

Lawrence& Puranam, 2005b).

Second, where do organizational or product interfaces come from? In the

modularity literature, the interfaces that partition activity into modular subsystems are

treated more or less exogenously, with little indication of how they are developed

much as the comparative discussion of coordination by plan and feedback largely fails

to explain where plans and standards come from. This omission makes the modularity

literature vulnerable to a charge of technological determinism (as was structural

contingency theory).

Our goal in this paper is to tackle these issues with a view to assist in a fuller

realization of the potential for ideas about modularity to advance theory on

organization design. We approach this task by first bringing some definitional clarity

to what interfaces mean in organizational contexts, and how product interfaces differ

from organizational interfaces. Second, we propose a dynamic model of how

interfaces are formed, that rests on the concept of systemic knowledge. Systemic

knowledge refers to what is known about the underlying patterns of interdependence

between tasks. As systemic knowledge accumulates, it is possible to define

organizational interfaces (i.e. rely on plans and standards based coordination) that

allow the separation of activity without having to rely on mutual adjustment

(Jacobides and Billinger, 2006). However, systemic knowledge itself accumulates

through mutual adjustment experiences. Therefore, coordination through interfaces

must be preceded by a regime of unstructured coordination involving communication

and mutual adjustment. Further, once adopted an interface-based coordination regime

5

actively impedes the return to a unstructured coordination regime as organizations

become less experienced at unstructured coordination. We study the descriptive and

normative implications of this negative feedback process, and draw implications for

how organizational boundaries are formed and dissolve (Jacobides, 2005). Our

analysis thus suggests that rather than simply view plan and feedback as two alternate

regimes of coordination which organizational designers select from, fresh insights are

generated by focusing on the linkages between the two.

Organizational Interfaces: What they are and how they work

In most general terms, an interface describes how elements of a system

interact with each other. When the system is a piece of hardware, such as a PC, the

elements are various components such as motherboard, graphics chip, LCD screen

etc. and interfaces refer to the physical and electronic connections between these

components. When the system in question is a software program, the elements of the

system are various sub-routines and program modules, while the interfaces refer to

sequencing of execution, and information transfer between software modules through

the use of common variables or data. Most relevant to our discussion, when the

system refers to an organization, the elements are various organizational sub-units,

and interfaces describe the nature of communication and shared decision-making

across these sub-units, as embodied in plans, standard operating procedures,

integrating managers and cross-unit teams (Galbraith, 1974; Tushman & Nadler,

1978).

Interfaces are useful concepts to understand the architecture of complexity,

because they demarcate the sub-systems that comprise the system as well as the

manner in which sub-systems link to each other. It is important to note that interfaces

6

refer not only to where the partitions in a complex system lie, but also the nature of

those partitions. Two attributes of interfaces are of particular interest to students of

modularity: a) the extent to which interfaces are “thick” or “thin” - which captures

the amount of interaction between elements, b) the extent to which interfaces are

“well-specified” vs. “poorly specified”- the degree of detail with which interactions

between the linked elements of a system can be described before they occur.

To understand the distinction between thickness and specification, consider

standard operating procedures that may allow two units to exchange large volumes of

information through technological systems or engage in complex coordination tasks

but in a very structured, pre-determined format. In contrast, committees or cross-

functional teams allow for a relatively smaller volume of interaction, but of a highly

non-structured nature, which allows for unstructured coordination (Langlois, 2001).

This distinction between the thickness of interfaces and their specification mirrors the

distinction drawn by Daft and Lengel (1986) between information volume and

richness. Daft and Lengel refined earlier views on information processing (Galbraith,

1977; Tushman & Nadler, 1978) to suggest that increasing volume of information

exchange is only useful when problems can be framed in such a way that more

information leads to resolution. However, when problems are unstructured, equivocal

and ambiguous, then it is the richness of information rather than its volume that

matters. Communication media that permit rich information transfer are those that

permit real-time human interactions, such as collocation, video-conferencing and

telephones.

Concepts similar to unstructured coordination have been discussed by other

scholars under names such as integrated problem solving (von Hippel, 1998),

technical dialog (Monteverde, 1995) or qualitative coordination (Langlois, 2001).

7

Thus the “thickness” of an interface between any two units may characterize the

amount of information exchange and coordination between two units, but the fact that

the interface is “poorly specified” refers to the large volume of information flow and

coordination that is unstructured. Thus, thickness and specification are distinct but not

orthogonal concepts; interface thickness refers to the total amount of coordination,

whereas (the lack of) interface specification refers to the volume of unstructured

coordination; the latter is a subset of the former.

The distinction between thickness and specification of interfaces suggests an

important distinction between modularity in technology and organization. In designed

artifacts such as hardware and software, regardless of their thickness, interfaces must

always be well-specified in a functioning product. In contrast, both thickness and

specification of organizational interfaces matter, and the benefits of separating

activity within organizations through interfaces depends on both aspects. If anything,

it is possible that the critical parameter may be the degree of specification of

organizational interfaces – the extent to which the interface avoids the need for rich,

unstructured coordination. This is because given the constantly falling costs of

Information Technology (IT), the costs of transmitting large volumes of structured

information are unlikely to prove to be binding constraints in organization design;

however, rich unstructured coordination continues to rely significantly on face-to-face

communication, for which the key constraints remain human information processing

capacities.

We therefore argue that in organizations, as opposed to technological systems,

the key issue is not complete separability and independence of action across sub-

systems but rather the extent to which the nature of interdependence is fully

predictable. Situations of complete independence and the resulting ability to operate

8

in a perfectly parallel way are rare in organizations (Thompson, 1967) and even rarer

in terms of the ability of organizational architects to discover such perfect de-

couplings (Ethiraj and Levinthal, 2004). Interdependence between organizational

units is ubiquitous, and may even require ongoing coordination and mutual

adjustment between them (Thompson, 1967) but if such patterns of adjustment are

fully predictable ex ante, then coordination is a “routine” affair (Gulati et al, 2005). 1

Thus, the degree of specification of organizational interfaces is the key theoretical

variable of interest.

Well-specified interfaces work by specifying in advance exactly what each

sub-system that links to another must do. For instance, two interdependent individuals

may coordinate their activities by communicating and through joint decision-making;

alternately a well-specified interface such as a standard operating procedure may

simply specify exactly what each must do individually so that their joint actions are

coordinated. It is in this sense that well-specified interfaces “embed” coordination –

they enable linked systems to act in a coordinated manner without having to explicitly

engage in coordination, by specifying exactly what each sub-system must do in order

for the aggregate system to work effectively (Baldwin & Clark, 2000; Ethiraj &

Puranam, 2004; Sanchez & Mahoney, 1996).

Table 1 provides some examples of well-specified and poorly specified

organizational interfaces. Intermediate degrees of interface specification are also

possible, arising from different weighted combinations of the high and low

specification mechanisms. For instance, one can envisage organizational interfaces

1 We think that Thompson’ s classic treatment of interdependence does not make this point clearly enough, to the detriment of further development of these ideas. Reciprocal interdependence refers to the circular direction of information flows without reference to the intrinsic difficulties involved in coordination. For extended discussions of these points, see Gulati, Lawrence and Puranam, 2005.

9

that rely primarily on unstructured, face–to-face coordination, on anonymous,

standardized operating procedures, and some combination of the two. In fact, except

in the most routinized situations, we believe that organizational interfaces are more

often somewhere in the “swollen middle” (Hennart, 1993). This underlines the

important distinction between modularity in products and organizations- interfaces of

intermediate specification may be useful in organizational settings, while

inconceivable in the product domain.

Interface specification vs. Thickness: An example

To crystallize the distinction between interface specification and thickness, we

present an example of a hypothetical organizational modularization problem, using

the commonly used device known as a task structure matrix (Baldwin & Clark, 2000).

The problem is a stylized example of product commercialization, with the various

component tasks and their interdependencies indicated as usual in Table 2a. As is

standard in such representations, an x in row j column k indicates that task j depends

on task k. The representation allows for one-way dependence (a single x below/above

the diagonal) or interdependence (with symmetric x’s below and above the diagonal).

Table 2b shows the same problem after it has been “modularised” into three groups of

tasks- module 1 (product design), module 2(media) and module 3(packaging). An off

diagonal x pertains to a sequencing issue; activity i cannot be performed till activity j

is performed. A set of x’s symmetrically above and below the axes indicates need for

ongoing coordination- “cycling”-activity i depends on j and j depends on i.

There are three interfaces in this representation (between module 1 and 2, 2

and 3 and 1 and 3). The one between module 2 and 3 is the thinnest- there is no

dependence between them at all. The activities in these two modules can in fact

proceed in parallel, require no ongoing coordination, can be measured and rewarded

10

separately, can benefit from specialization and can benefit from rapid reconfiguration.

The interface between module 1 and 2 is still relatively thin, though thicker than that

between modules 2 and 3. The interface between modules 1 and 3 are the thickest,

with the most number of dependencies and interdependencies.

However, note that from this representation alone, we cannot say whether the

benefits of modularity are greater at the 1-3 interface or at the 1-2 interface (though

both are certainly lower than at the 2-3 interface). This is because this representation

tells us little about how specified these interfaces are – the extent to which the cross-

module interactions represented by the x’s in Figure 2 can be described in detail

before they occur. For instance, it is clear from Figure 2b that the design of the media

campaign cannot occur before engineering design, prototyping and manufacture; but

what is not clear is the extent of unstructured coordination required between the

media campaign designers ad the engineering team for the former to design an

appropriate campaign that highlights technical features of the product appropriately.

Indeed it is possible that this single x may represent more equivocality and need for

unstructured coordination than the three x’s that appear in the 1-3 interface.

We have consciously chosen an illustration that pertains to organizational

rather than product modularity. As a contrast, one can imagine a similar diagram

showing the sub-modules within a software programme or a piece of hardware; in

such designed artefacts, if they are functioning well, there can be very little doubt

about the nature of interactions. Physical tolerances, information flows, sequencing

must all be fully specified for the artefact to function.

In sum, because of the separation of activity they enable, well-specified (as

opposed to thick or thin) organizational interfaces are the basis for defining

organizational and ownership boundaries. Understanding how such interfaces are

11

formed can therefore generate insights on changes in the division of activity within

and between firms.

A dynamic model of organizational interface creation

Our goal in this section is to outline a model that describes the dynamics of

interface formation. Our model represents an organization with two organizational

sub-units. The two units are interdependent, so that their activities must be

coordinated in order to optimize organizational performance. The two sub-units may

be related vertically or horizontally. For instance, these may be design and

manufacturing units, or two product development groups working on sub-systems

within a complex system. We define “systemic knowledge” as the knowledge about

the nature of interdependence between sub-systems that is essential for defining

interfaces between the two units. Such knowledge cannot be assumed; often the

nature of interactions between sub-systems, and indeed of the location of boundaries

between sub-systems may be unknown to an organization designer. Yet over time, as

the personnel in the two units interact, they are likely to build a stock of systemic

knowledge as they begin to understand the true nature of interdependence between

activities in the two units.

We define “unstructured coordination” as a process of mutual adjustment

supported by frequent and bi-directional communication. The hallmark of

unstructured coordination is that it is effective even when systemic knowledge is

imperfect or non-existent. Thus, the individuals in the two sub-units of our

organization may be able to coordinate their activities through a process of trial-and –

error learning even in the absence of detailed knowledge about how their actions are

interdependent. This process of collective trial-and-error learning is what we refer to

as unstructured coordination. While unstructured coordination can take place even in

12

the absence of systemic knowledge, it is also a basis for building systemic knowledge.

Indeed such knowledge is more likely to be gathered through unstructured

coordination efforts than through highly structured information exchange;

understanding the full ramifications of changes in one sub-system to another sub-

system is more likely if communication between them is not constrained by

predefined topics and channels of information. We therefore assume that systemic

knowledge tS is a concave increasing function of cumulative experience at

unstructured coordination tV . The concavity builds in diminishing marginal returns,

which is a ubiquitous assumption about knowledge building processes. Thus, the

relationship between systemic knowledge and coordination experience must satisfy

0,0 2

2

<∂∂

>∂∂

t

t

t

t

VS

VS

(1)

While systemic knowledge about the nature of interdependence is being

accumulated through unstructured coordination, over time the organization also

becomes more adept at unstructured coordination itself. Repeated interactions give

rise to trust, interpersonal routines, and common language, which together enhance

the efficacy of mutual adjustment between them. Therefore, over time, an

organization’s competence at unstructured coordination improves. Consistent with

standard learning curve formulations, we model this effect through a reduction in the

costs of unstructured coordination per unit tc as a convex decreasing function of

cumulative experience. The relationship between the per unit costs of coordination

and coordination experience must satisfy:

0,0 2

2

>∂∂

<∂∂

t

t

t

t

Vc

Vc

(2)

13

Finally, we assume that the extent of unstructured coordination tU necessary

to manage inter-unit interdependence at any point of time decreases with systemic

knowledge. As the individuals in the two sub-units build systemic knowledge and

gain a better understanding of how their actions are interdependent, they are able to

embed coordination into standards, rules and procedures, instead of relying on

unstructured coordination. We therefore assume that the extent of unstructured

coordination required between the units to manage interdependence reduces as a

convex decreasing function of systemic knowledge; the convexity reflects the

assumption that it is never possible to eliminate entirely the need for unstructured

coordination, because systemic knowledge is ultimately the product of learning by

bounded rational individuals – it is unlikely to be perfect. Further, we allow for a

distinction between the existence of systemic knowledge and its utilization to define

interface standards. We denote by tt Si φ= the systemic knowledge that is used to

define standardized coordination patterns between the sub-units through standards,

operating procedures, rules etc. The parameter � functions as a “switch”- when �=0,

then the organization may possess systemic knowledge but does not utilize it to define

interfaces; when �=1 then this knowledge is utilized. The relationship between the

extent of unstructured coordination and systemic knowledge used to define interfaces

must satisfy:

0,0 2

2

>∂

∂<

∂∂

t

t

t

t

i

Ui

U (3)

The preceding conditions embody the assumptions on which we build our

theoretical arguments. To complete the specification of our model, we also state two

identities. The first equation simply specifies the accumulation function for

coordination experience

14

�+=t

tottot dtUVV (4)

The second states that coordination costs tC are equal to the volume of unstructured

coordination multiplied by the cost per unit of unstructured coordination.

ttt cUC .= (5)

The model specification implied by 1-5 captures a fundamental asymmetry between

the two coordination regimes. In an unstructured coordination regime (i.e. no

interfaces) the total costs of coordination must decline over time simply because of

increasing coordination competence resulting from accumulated coordination

experience (2), but the volume of unstructured coordination remains constant. In

contrast, defining interfaces affects both the volume of unstructured coordination (4)

as well as the cost per unit of coordination. Specifically, because interfaces curtail the

volume of coordination at any point in time, they also result in slower accumulation

of coordination experience (4). Investing in interfaces therefore not only reduces the

extent of coordination required at any point in time but also a) slows down the rate at

which per unit coordination costs decline c) as well as the rate at which systemic

knowledge is built.

There are three important implications of these effects. First, the total costs of

coordination may be lower or higher in an interface-based coordination regime than in

an unstructured coordination regime. Relative to an unstructured coordination regime,

introducing interfaces decreases the extent of coordination necessary, but also raises

the per unit cost of coordination. This suggests a boundary condition on when an

interface based coordination regime is preferable to an unstructured coordination

regime (purely on a coordination cost basis- this does not account for the benefits of

interface specification such as gains from specialization or superior monitoring and

15

measurement). Put simply, interfaces improve coordination only when the “volume

reduction” effect dominates the “cost increase” effect.

Second, delaying interface definition (i.e. “turning on ��later”� may lead to

higher levels of systemic knowledge. This is because the level of systemic knowledge

at time t will be higher in an unstructured coordination regime than in an interface-

based coordination regime, because of the slower accumulation of coordination

experience in the latter. An interface introduced at time t will therefore be better

specified – have more systemic knowledge to draw on- than one that was introduced

in t0, and from time t onwards, will always have lower levels of unstructured

coordination compared to a system in which the interface was introduced at time t0.

Delayed transition to interface-based coordination may therefore help to control

coordination costs.

Third, coordination competence (as measured by per unit costs of

coordination) is typically likely to be lower in an organization that adopts an interface

based coordination regime. This turns on its head the conventional wisdom that

organizations with high competence at unstructured coordination are less likely to

invest in interface-based coordination. This is a particularly important consideration if

we believe that rarely but surely, exogenous shocks to the system change underlying

patterns of interdependencies and destroy existing systemic knowledge (and therefore

interfaces). Building interfaces impedes the formation of competence at unstructured

coordination, which organizations must fall back on when interfaces are destroyed.

Thus, shocks to systemic knowledge may not only lower coordination performance in

an interface based regime, but may lower performance relative to an unstructured

coordination regime.

16

We explore these intuitions about the viability of an interface-based

coordination regime rather than rely on an unstructured coordination regime more

rigorously by simulating the dynamic model specified in equations 1-5. The specific

functional forms used for equations 1-3 are:

tVt eS θ−−= 1 (1’)

tVt ec α−= (2’)

tit eU λ−= (3’)

We rely on exponential forms for all three equations to maintain

comparability. The parameters in these functions affect the second derivatives in

meaningful ways. For instance, the parameters ��and � have meaningful

interpretations as “investments in building systemic knowledge and coordination

competence” respectively. Increasing these parameters increases the rate at which

coordination experience is converted into systemic knowledge and coordination

competence respectively. Since our focus in this study is not on the effectiveness with

which systemic knowledge embedded into interfaces helps to reduce unstructured

coordination, we set �=1 in the simulation experiments described below. Unless

otherwise mentioned, the settings for the simulation runs used to produce graphs are

��������and ��������though in our analysis we verify results for both parameters in the

range [0,1].

The value of adopting interfaces for coordination

If 0.. >∂

∂+

∂∂

tt

tt c

iU

Ui

c then coordination costs ( tt cU . ) may increase through

the adoption of an interface. When is such a condition likely to hold? Both tU and tc

17

are always positive and i

Ui

c tt

∂∂

∂∂

, are positive and negative respectively. A simple

intuition is that for interfaces to begin pushing up per unit costs of coordination by

curtailing cumulative coordination experience (i.e. 0>∂

∂i

ct ) some time must elapse,

so that significant differences in coordination experience will result from adopting an

interface. Further, the magnitude of i

U t

∂∂

will decrease over time given the convex

decreasing nature of this function. It therefore seems likely that the “volume

reduction” effect precedes the “per unit cost increase” effect, so that interface

adoption is followed initially by lower coordination costs relative to an unstructured

coordination regime. Subsequently coordination costs may rise.

Figure 1 shows coordination costs tC in an unstructured (�����line 1) and

interface based coordination regime respectively (�����lines 2 and 3). Line 3

represents a higher ���������than line 2 (0.10). As expected, the adoption of an interface –

based coordination regime improves performance initially, but eventually worsens it

with respect to an unstructured coordination regime. Both the initial improvement and

eventual worsening are stronger for higher ���reflecting increases in both the “volume

reduction” and “per unit cost increase” effects with investment in building systemic

knowledge��

There are both descriptive and normative implications of these results. From a

descriptive standpoint, the adoption of an interface-based coordination regime is

likely to result in initial improvements in coordination performance, but eventual

declines. The decline is inevitable because of the increase in per unit coordination

costs (relative to an unstructured coordination regime) that occurs over time. Investing

significantly in extracting systemic knowledge from the lessons provided by

18

coordination experience is a double-edged sword, as it results in grater initial

improvements but worsens subsequent coordination performance. From a normative

viewpoint, one would expect that firms with a shorter time horizon and/or high

discount factors are more likely to adopt an interface-based coordination regime, all

else being equal.

Figure 2 shows the impact of higher levels of investment in building

coordination competence (�� on coordination costs in the two regimes. Lines 1 and 3

show unstructured coordination regimes (����, 2 and 4 show interface- based regimes

(�����with a constant ��������. The difference between lines 1 and 2 represent the

gains/losses from interface adoption when ��is low (0.05); the difference between

lines 3 and 4 for when ��is high (0.10). Higher rates of coordination competence

building bring the crossover point (at which the interface-based regime begins to

under-perform the unstructured coordination regime) forward, while also dampening

both the gains and losses from interface adoption. Thus, all other things (such as time

horizons and discount rates) being equal, organizations that can build coordination

competence rapidly are less likely to adopt an interface based coordination regime.

If we thought about both ��and ��as endogenous variables, the preceding

discussion also raises the question whether investments in building systemic

knowledge complement or substitute investments in building coordination

competence. The tentative answer we suggest is that it depends on the time horizon.

In the short run, investments in building systemic knowledge enhance the value of

investments in building coordination competence; in the longer run, such investments

may work at cross purposes to each other, as interfaces (and the systemic knowledge

they rest upon) slow down the rate at which coordination competence is accumulated.

19

The timing of interface adoption

Next, we explore the impact of interface adoption on systemic knowledge, and

implications for timing interface adoption. Figure 3 shows the level of systemic

knowledge tS over time, for a low and high value of �������������and for ������Lines 1 and 3

represent systemic knowledge for low and high values of � in an unstructured

coordination regime (i.e. with ����; lines 2 and 4 for low and high values of � in an

interface-based coordination regime (����. This reduction in systemic knowledge in

an interface-based regime suggests that there may be gains from delaying the adoption

of such a regime as the delay translates into enhanced levels of systemic knowledge.

To model delayed adoption of an interface, we introduce a threshold in systemic

knowledge. This modifies the expression for the systemic knowledge used to define

interfaces to tt Si φ= if χ>tS , else 0=ti . Higher threshold values (�� imply later

adoption.

Figure 4 shows coordination costs tC with different threshold levels (0.2,0.4,

0.6). Increasing the threshold in general increases coordination costs relative to a no-

threshold regime (i.e. one in which tt Si φ= at 0t ) up to the time it at which the

threshold is crossed; beyond that point, the higher the threshold, the lower the

coordination costs at every point in time after it . Delayed adopters of interfaces may

therefore enjoy coordination cost advantages over early adopters, after the transition

in coordination regime.

Figure 5 explores the impact of different rates of building coordination

competence on the timing of interface adoption. Lines 1 and 3 are coordination costs

with interface adoption at 0t , 2 and 4 with delayed adoption (������� The effect of

increasing the rate at which coordination competence is built (������������ is to reduce the

20

area between the lines for initial and delayed adoption prior to it and increase it after

it . The intuition is simple- before it , increasing coordination competence implies

lower costs of coordination for the same volume of coordination; after it coordination

competence with delayed interface adoption will be higher than with adoption at

0t (because the interface slows down the rate of building coordination competence).

Thus, delayed adopters of interfaces enjoy even larger coordination cost advantages

over early adopters after the transition in coordination regime if both accumulate

coordination competence rapidly.

The impact of shocks to systemic knowledge

To explore the impact of periodic shocks to systemic knowledge- such as

technological or organizational innovations which transform the nature of

interdependence between units- we modify our model specification to allow for

periodic destruction of the value of systemic knowledge gained through coordination

experience. Accordingly, we distinguish between tsV , the stock of coordination

experience relevant to building systemic knowledge, and tcV , the stock of

coordination experience relevant to building coordination competence. At intervals

drawn from a distribution, tsV is depleted by an amount t

sε , whereas tcV is

unaffected. This corresponds to shocks that destroy the systemic knowledge gained

from coordination experience, without damaging the coordination competence that

also arises from coordination experience. Equation 4 is accordingly modified to

dtUVVt

to

ts

ttos

ts ][� −+= ε (4a)

�+=t

totto

ct

c dtUVV (4b)

21

Note that when tsε =0, then t

sV is always equal to tcV . We draw t

sε from a

Montecarlo distribution.

In Figure 6, we first plot coordination costs during the time window when

specifying interfaces improves performance relative to an unstructured coordination

regime (���������������. This is shown by the difference between the lines “with

interfaces” and “without interfaces”. The third line in yellow represents the average

coordination costs for 50 runs of the model, with a 15% probability of a shock in each

period that completely destroys the value of systemic knowledge gained through

coordination experience. The impact of the shocks, as we conjectured, is to lower

performance below that of an unstructured coordination regime towards the end of

this time window (see periods 20-40). Put differently, shocks serve to bring forward

the crossover point (at which the interface-based regime begins to under-perform the

unstructured coordination regime).

This result can be understood in terms of the impact of shocks on the twin

consequences of defining interfaces- “volume reduction” and “cost increases”. With

shocks, the specification of interfaces periodically goes to zero, leading to spikes in

the extent of unstructured coordination. This weakens both the coordination volume

reduction effect (as shocks increase the extent of unstructured coordination required)

as well as the unit cost increase effect (since cumulative experience at unstructured

coordination is likely to be higher with shocks). Since the impact of differences in

cumulative coordination experience on the unit costs of coordination are slight

initially, but only build up over time, shocks effectively weaken the volume reduction

effect more strongly than the cost increase effect; the cross over point therefore

“moves” to the left.

22

Figure 7 shows the effect strengthening as the frequency of shocks increases

(from a 10% probability of a shock every period -“I-rare” to 30% - “I-frequent”). As

shocks become more frequent, the performance of the interface based coordination

regime worsens and the crossover point moves to earlier periods. Thus, shocks to

systemic knowledge not only lower coordination performance in an interface-based

regime, but lower performance relative to an unstructured coordination regime. Put

differently, when systemic knowledge is unstable and the underlying patterns of

interdependence are frequently changing, building systemic knowledge and defining

interfaces that are valid for brief intervals may be worse than sticking with an

unstructured coordination regime.

Discussion: Interfaces, Boundaries of Organization & Boundaries of Ownership

Our analysis points to some limits to the usage of interface-based coordination

regimes- the advantages of interfaces relative to unstructured coordination decline

over time, and when systemic knowledge is unstable and subject to frequent revision,

then the advantages of an interface based regime last for an even shorter duration. The

gains from interface-based coordination regimes are also smaller for organizations

that can build coordination competence rapidly. Finally, organizations that delay the

adoption of an interface-based regime are likely to outperform organizations with

unstructured coordination regimes for significant periods of time. Organizational

interfaces have been at the centre of our analysis, because they are the basis on which

the division of activity within and between firms is defined. Rather than simply view

such interfaces as a pre-existing alternative to an unstructured coordination regime,

we have proposed a dynamic model of how the systemic knowledge necessary to

define them builds over time and at the expense of developing competence at

unstructured coordination.

23

Our analysis of the dynamics of the two coordination regimes and their

interactions suggests several implications for organizational form, which we discuss

below. In keeping with standard theoretical treatments about firm boundaries, we

make two assumptions about the difference between intra-and inter-firm contexts:

first, that there is lesser incentive conflict between the two parties interacting through

an interface when both are part of the same firm than when they are not (Williamson,

1985; Williamson, 1991); and second, that costs of unstructured coordination between

the two parties is lower when both are part of the same firm than when they are not

(Kogut & Zander, 1992, 1996). These two assumptions characterize transaction cost

and the knowledge based perspective respectively, and a point of contention between

theorists is the sufficiency of each as an explanation of firm boundaries (Foss, 1996).

However, that is not our focus; we make both assumptions with a view to

understanding how interfaces evolve and confer benefits within the inter-firm as

opposed to the intra-firm context.

Implications for ownership and organizational boundaries

Our analysis suggests an endogenous explanation for why interface-based

coordination regimes are periodically discarded by organizations in favor of

unstructured coordination regimes. Organizations may merge activities separated by

departmental/divisional boundaries, re-integrate into vertical segments they have

exited, and periodically discard existing standards and rules not necessarily because

these become inappropriate due to some exogenous changes to the system, but simply

because their value relative to an unstructured coordination regime inevitably declines

over time. When systemic knowledge is in flux- the pattern of underlying

interdependencies connecting organizational sub-units is changing frequently - this

reduces the period in which interfaces confer coordination cost advantages over

24

unstructured coordination regimes, leading to interfaces being discarded more rapidly.

Note that this is not the same as saying that when systemic knowledge is outdated, the

interface that is based on this knowledge is less valuable - instead we are arguing that

the regime of coordination by interfaces in general becomes less valuable due to the

frequency with which interfaces must be re-defined- even if there were no costs to

creating each interface (as we have assumed).

Our analysis also suggests that the magnitude of the coordination cost

economies from interface-based regimes is diminished in the presence of rapid rates

of coordination competence building. This indicates that the adoption of an interface-

based coordination regime is less likely, the faster an organization can bring about

improvements in coordination competence through building on coordination

experiences. Thus organizational interfaces are more likely to be adopted in

established rather than de novo relationships, because of the greater rate at which

coordination competence can improve with experience in the latter. Similarly, all else

being equal, within firms, coordination competence may develop faster because of

common language, low-powered incentives, propinquity and identification- starting

from similar initial levels (Monteverde, 1995). Thus, we would expect that activities

performed by different organizational units within a firm are more likely to rely on

unstructured coordination, than similar activities coordinated across organizational

boundaries. Interface based coordination regimes should be more widespread between

rather than within firms – the lower levels of standardization of interactions between

units within a firm, compared to similar interactions between firms may in fact be a

sign of efficiency.

There are other important differences between organizational interfaces within

and between firms that we have not directly examined in our model. For instance, the

25

design of interfaces may involve conflicts of interest cannot be ignored in the context

of inter-firm relationships. While we have assumed that having knowledge of the

underlying interaction structure is all that is needed to design an interface, actually,

selecting the appropriate interface may be more like a coordination game- with

multiple equilibria, on which different people may have different preferences. Two

organizational units within a firm may have different preferences on which interface

to select, but agreement will eventually be reached through the intervention of higher

authority, if necessary (Williamson, 1991). By definition, such a final and effective

arbitrator is not available when two independent firms disagree on the interface that

they prefer. The implication is that it is harder to create interfaces between rather than

within firms; but if an interface exists, it is easier to do business with external

organizations.

Our results on the benefits of delaying interface definition till some systemic

knowledge has been built underlines two key elements of our theoretical framework –

first, that an unstructured coordination regime must precede an interface-based

coordination regime, and second that the benefits of interface specification is subject

to diminishing marginal returns. Delaying interface definition allows for the build-up

of systemic knowledge through unstructured coordination, while once an interface is

introduced, the rate at which systemic knowledge is built must diminish (because of

the decrease in the extent of unstructured coordination each period). The

standardization of coordination must not be rushed- reorganization or vertical

disintegration in the aftermath of technological innovations must allow time for

systemic knowledge about new patterns of interdependence to be built.

These results also suggest that in general, using organizational interfaces to

partition activity is a good idea in the longer term only if there are significant gains

26

from the use of interfaces. Our own analysis has only focused on the coordination

costs associated with coordination through interfaces (as opposed to unstructured

coordination), but the potential benefits of using interfaces are well known. The idea

that specialization gains can arise from controlling cognitive diseconomies of scope is

at least as old as Adam Smith’s original description of the pin factory. Because well-

specified interfaces allow focusing cognitive resources on a narrower set of actions

(production only as opposed to production + coordination), they can lead to static and

dynamic improvements in performance. Indeed the formation of well-specified

interfaces can be motivated by the gains from trade that can emerge when less

efficient in-house operations are abandoned and substituted by outside providers

(Jacobides & Hitt, 2005); or the gains from specialization which obtain when firms

focus where they can leverage and develop their competencies (Jacobides & Winter,

2005).2

Well-specified interfaces enhance accountability because they describe

precisely the acceptable outputs from each sub-unit, and lower measurement costs

(Barzel, 1982), enabling the use of sharp incentives that link rewards to outputs

(Zenger & Hesterly, 1997)Jacobides & Billinger; 2006). Since well-specified

interfaces provide clear descriptions of acceptable outputs, they can also result in

lower costs of selecting and finding trading partners, and contracting with them

(Baldwin and Clark; 2003). Well-specified interfaces also enable rapid

reconfiguration – “mix and match”. Individual organizational units (including

suppliers) can be replaced by others, while maintaining the overall pattern of

interdependence between organizational units if the organizational interfaces that

2 It is important to keep in mind that since we are discussing the incentives of an individual firm to invest in an interface, the gains from specialization must be large enough for it to be in the interests of individual firms to gain from interface specification and outsourcing, and not just efficiency enhancing at some global level, unless we assume that selection pressures are indeed highly effective.

27

connect them are well specified. This property lies behind the suitability of the M-

form organization for conglomerate diversification, and of policies for rapid re-

assignment of charters to divisions within companies (Galunic & Eisenhardt, 2001b).3

Interfaces deliver the benefits of rapid reconfiguration even within the inter-firm

setting as they can enable firms to switch trading partners rapidly, as Hoetker shows is

possible in the laptop industry (2002).4

Since the coordination cost advantages of interface-based regimes decline

relative to unstructured coordination regimes in the long run, their adoption is

economically rational only when the gains from interfaces- specialization,

accountability, rapid re-configuration- are relatively large. Put differently,

unstructured coordination regimes may be preferable unless the benefits from

interface specification not only exist but are significant.

Conclusion

We see this paper as making three basic contributions. First, technological

interfaces in products differ in important ways from organizational interfaces; while

the former must be well specified (but may be either thin or thick), the latter need not

be. In fact, from an organizational standpoint, thin interfaces are not necessary for

most of the gains from modularity. Conversely, variations in the degree of interface

specification are important from an organization design perspective, because of gains

3 Localized disruption is the other side of the same coin – in the limit some sub-units can be destroyed and completely replaced with another, with minimal implications for the activities in other units. This is at the heart of the fable of the two watchmakers, Tempus and Hora that Simon entertainingly recounts. Faced with unpredictable disruptions that destroy all interlinked components, the watchmaker who adopts a modular assembly strategy with well-specified interfaces minimizes his maximum rebuild costs. 4 However, it is important to distinguish specificity and specification- a well specified interface, if it is not an industry standard may require significant relationship specific investment and create hold-up hazards. Thus rapid reconfiguration benefits in the inter-firm context are only feasible when interfaces are not only well specified, but open and widely used.

28

from reduced coordination costs and enhanced accountability; these concepts have no

good analogues in product modularity.

Second, the differences between product and organizational modularity

become even more pronounced as we look closer at the process of interface

formation. We have argued systemic knowledge is essential for effective interface

definition. Since unstructured coordination is an important means by which systemic

knowledge is gained, an implication of our arguments is that for novel products, non-

modular organizations may be necessary to develop modular products. Thus, the link

between technological and organizational modularity may be stronger in mature

technological settings rather than in settings with considerable novelty; existing levels

of systemic knowledge may be a powerful moderating variable that helps to explain

when product and organizational modularity go together, as well as when they do not.

Third, our analysis suggests that variations in organizational attributes such as

the rate at which systemic knowledge and coordination competence is built, and

differences in time horizons over which the benefits of interface specification are

evaluated can generate considerable variations in the rates of adoption of

organizational modularity and the degree of interface specification, even for the same

basic technologies. Elaborating on this analytical framework may help us to provide a

sophisticated rebuttal to the critique of technological determinism that often hovers

over the literature on technological and organizational modularity. We have also

sketched out arguments for why ownership may remain integrated despite the

existence of organizational interfaces (because of the need for specific investments to

conform to an interface), and conversely why interfaces may never form despite

ownership already being separated (this can occur if coordination competence is high

29

without a correspondingly high level of systemic knowledge and benefits from

modularity).

From a broader theoretical perspective, our research helps address a seeming

inconsistency in the knowledge-based view of the firm. For instance, Nickerson and

Zenger (2004) recently argued that “the two fundamental arguments within the

literature that support the efficiency of firms relative to markets in knowledge

exchange are fully contradictory. One claims that hierarchies essentially exist to avoid

knowledge transfer, emphasizing the firm’s capacity to exercise authority in directing

others actions. The other view claims that hierarchies exist instead to facilitate

knowledge transfer, emphasizing the firm’s capacity to support the formation of

shared language and identity”. We believe that our discussion shows that far from

being contradictory, these two views are complementary: On the basis of the

evolutionary dynamics we described, extensive knowledge sharing enhances interface

specification which thereafter reduces the need for extensive knowledge sharing over

time. The apparent inconsistency highlighted by Nickerson and Zenger disappears if

we recognize that the two advantages of hierarchies apply at different levels of

systemic knowledge (Nickerson & Zenger, 2004).

We hope that the framework developed in this paper will help to understand

both the benefits and the costs of the two coordination regimes- interface-based and

unstructured - as well as their dynamic properties, and lead to a more balanced,

empirically grounded descriptive theory, as well as more robust prescriptions for

organizational and ownership design.

30

Well-specified Poorly specified

Organizational interfaces

Reliance on Rules & regulations, Formal information systems, Standards

Reliance on cross-unit teams, Committees, Integration managers

Instances Design for Manufacturing processes(Puranam, Singh& Zollo,2006); Financial reporting systems in conglomerates (Markides & Williamson, 1996)

Cross-functional teams in product development (Iansiti, 1998); collocation of engineers in procurement relationships (Dyer, 1996); structural integration in technology acquisitions (Puranam, 2001)

Table 1: Well and poorly specified organizational interfaces

31

Table 2a

Table 2b

Task Structure Matrix

1 2 3 4 5 6 7 8 9 101 Initial engineering design X2 Design packaging X X3 Market research X X4 Manufacture X X X 5 Pick creative team X6 Finalize media capmpaign X X X7 Select packaging material X X 8 Prototyping X X X9 Select ad agency X

10 Design media campaign X X X X

Modularized Task Structure Matrix

1 3 8 4 9 5 10 6 7 21Initial engineering design X X3Market research X X8Prototyping X X X4Manufacture X X X 9Select ad agency X5Pick creative team X

10Design media campaign X X X X6Finalize media capmpaign X X X7Select packaging material X X2Design packaging X X

32

Figure 1: Coordination costs in unstructured and interface-based coordination regimes

Figure 2: Coordination competence and effects of interface adoption

33

Figure 3: Systemic knowledge in unstructured and interface-based coordination regimes

Figure 4: Coordination costs with delayed interface adoption

34

35

Figure 5: Coordination competence and effects of delayed interface adoption

Figure 6: The impact of shocks to systemic knowledge on coordination costs

Shocks to systemic knowledge

0.0000

0.2000

0.4000

0.6000

0.8000

1.0000

1.2000

0 10 20 30 40 50

Periods

Coo

rdin

atio

n co

sts

Without Interface

With Interface

With interface & shocks

36

Figure 7: Rare vs. Frequent shocks to systemic knowledge

Shocks to Systemic Knowledge

0.0000

0.2000

0.4000

0.6000

0.8000

1.0000

1.2000

0 10 20 30 40

Periods

Coo

rdin

atio

n C

osts

No InterfaceWith InterfaceI-rare shocksI-frequent shocks

37

References

Baldwin, C. Y., & Clark, K. B. 2000. Design rules: The Power of Modularity. Cambridge, MA: Harvard Business School Press.

Barzel, Y. 1982. Measurement Cost and the Organization of Markets. Journal of Law & Economics, 25(1): 27.

Bradach, J. L., & Eccles, R. G. 1989. Price, Authority and Trust :From Ideal types to Plural Forms. Annual Review of Sociology, 15: 97-118.

Dyer, J. H. 1996. Does Governance Matter? Keiretsu Alliances and Asset Specificity as Sources of Japanese Competitive Advantage, Organization Science: A Journal of the Institute of Management Sciences, Vol. 7: 649.

Ethiraj, S., & Puranam, P. 2004. The Distribution of R&D Effort in Systemic Industries: Implications for Competitive Advantage, Advances in Strategic Management, Vol. 21: 225-253.

Ethiraj, S. K., & Levinthal, D. 2004. Modularity and Innovation in Complex Systems. Management Science, 50(2): 159.

Foss, N. J. 1996. More critical comments on Knowledge based theories of the firm. Organization Science, 7(5).

Galbraith, J. R. 1974. Organization Design: An Information Processing View. Interfaces, 4(3): 28.

Galbraith, J. R. 1977. Organization Design. California: Addison-Wesley. Galunic, D. C., & Eisenhardt, K. M. 2001a. Architectural Innovation and Modular

Corporate Forms. Academy of Management Journal, 44(6): 1229. Galunic, D. C., & Eisenhardt, K. M. 2001b. Architectural Innovation and Modular

Corporate Forms, Academy of Management Journal, Vol. 44: 1229. Grandori, A. 1997. An Organizational Assessment of Interfirm Coordination Modes.

Organization Studies (Walter de Gruyter GmbH & Co. KG.), 18(6): 897. Grossman, S. J., & Hart, O. D. 1986. The Costs and Benefits of Ownership: A Theory

of Vertical and Lateral Integration. Journal of Political Economy, 94(4): 691. Gulati, R., Lawrence, P. R., & Puranam, P. 2005. Adaptation in vertical relationships:

beyond incentive conflict. Strategic Management Journal, 26: 415-440. Heath, C., & Staudenmayer, N. 2000. Coordination Neglect: How Lay theories of

organizing complicate coordination in organizations. Research in Organizational Behavior, 22: 53-191.

Helfat, C., & Eisenhardt, K. M. 2000. Inter-temporal economies of scope, organizational modularity, and the dynamics of diversification. Working Paper.

Hennart, J.-F. 1993. Explaining the Swollen Middle: Why Most Transactions Are A Mix of "Market" and "Hierarchy". Organization Science: A Journal of the Institute of Management Sciences, 4(4): 529.

Hoetker, G. 2002. Do modular products lead to modular organizations?, College of Business Working papers, UI-UC. Urbana Champagne, Illinois.

Iansiti, M. 1998. Technology integration. Harvard, MA: Harvard Business School Press.

Kogut, B., & Zander, U. 1992. Knowledge of the Firm, Combinative Capabilities, and the Replication of Technology. Organization Science: A Journal of the Institute of Management Sciences, 3(3): 383.

Kogut, B., & Zander, U. 1996. What firms do? Coordination, identity, and learning. Organization Science, 7(5): 502-518.

38

Langlois, R. N. 2001. Modularity in Technology and Organization. Journal of Economic Behavior and Organization.

Markides, C. C., & Williamson, P. J. 1996. Corporate diversification and organizational structure: A resource-based view, Academy of Management Journal, Vol. 39: 340.

Monteverde, K. 1995. Technical dialog as an incentive for vertical integration in the semiconductor industry. Management Science, 41(10): 1624-1638.

Nickerson, J. A., & Zenger, T. R. 2004. A Knowledge-Based Theory of the Firm--The Problem-Solving Perspective, Organization Science: A Journal of the Institute of Management Sciences, Vol. 15: 617.

Puranam, P. 2001. Grafting innovation: The acquisition of entrepreneurial firms by established firms.,

University of Pennsylvania., Philadelphia. Puranam, P., Singh, H., & Zollo, M. Forthcoming. Organizing for Innovation:

Managing the coordination-autonomy dilemma in technology acquisitions, Academy of Management Journal.

Sanchez, R., & Mahoney, J. T. 1996. Modularity, flexibility, and knowledge management in product and organization design. Strategic Management Journal, 17(10): 63.

Schilling, M. A. 2000. Toward a General Modular Systems Theory and Its Application to Interfirm Product Modularity. Academy of Management Review, 25(2): 312.

Schilling, M. A., & Steensma, H. K. 2001. The Use of Modular Organizational Forms: An Industry-Level Analysis. Academy of Management Journal, 44(6): 1149.

Simon, H. A. 1962. The sciences of the artificial. MA: MIT Press. Thompson, J. D. 1967. Organizations in action. New York: McGraw Hill. Tushman, M. L., & Nadler, D. A. 1978. Information processing as an integrating

concept in organizational design. Academy of Management Review, 3(3): 613.

von Hippel, E. 1998. Economics of product development by users: The impact of `sticky' local information. Management Science, 44(5): 629.

Williamson, O. E. 1985. The Economic Institutions of Capitalism. New York: Free Press.

Williamson, O. E. 1991. The logic of Economic Organization. In O. E. Williamson, & S. Winter (Eds.), The Nature of The Firm.

Zenger, T. R., & Hesterly, W. S. 1997. The Dissaggregation of Corporations: Selective Intervention, High-powered Incentives, and Molecular Units. Organization Science: A Journal of the Institute of Management Sciences, 8(3): 209.