Content uploaded by Paul R Carlile
Author content
All content in this area was uploaded by Paul R Carlile on Dec 12, 2013
Content may be subject to copyright.
Organization Science
Vol. 15, No. 5, September–October 2004, pp. 555–568
issn 1047-7039 eissn 1526-5455 04 1505 0555
inf
orms
®
doi 10.1287/orsc.1040.0094
© 2004 INFORMS
Transferring, Translating, and Transforming: An Integrative
Framework for Managing Knowledge Across Boundaries
Paul R. Carlile
Information Systems Department, School of Management, Boston University, 595 Commonwealth Avenue,
Boston, Massachusetts 02215, carlile@bu.edu
T
he paper examines managing knowledge across boundaries in settings where innovation is desired. Innovation is a
useful context because it allows us to explore the negative consequences of the path-dependent nature of knowledge.
A framework is developed that describes three progressively complex boundaries—syntactic, semantic, and pragmatic—
and three progressively complex processes—transfer, translation, and transformation. The framework is used to specify the
practical and political mismatches that occur when innovation is desired and how this relates to the common knowledge
that actors use to share and assess each other’s domain-specific knowledge. The development and use of a collaborative
engineering tool in the early stages of a vehicle’s development is presented to illustrate the conceptual and prescriptive
value of the framework. The implication of this framework on key topics in the organization theory and strategy literatures
is then discussed.
Key words: knowledge; innovation; path dependence; power; boundary management; product development
1. Introduction
Dorothy Leonard’s (1995) statement that most innova-
tion happens at the boundaries between disciplines or
specializations tells us that working across boundaries is
a key ingredient of competitive advantage, but also why
innovation proves so difficult to create and maintain. The
growing research on knowledge in organizations under-
scores this challenge by recognizing first the “knowl-
edge boundaries” (Brown and Duguid 2001) between
specialized domains and second that knowledge is “both
a source of and a barrier to innovation” (Carlile 2002,
p. 442). In this paper the topic of boundaries in orga-
nization theory will be reexamined from a perspective
of managing knowledge across boundaries in settings
where innovation is desired.
This focus on innovation will help to extend the
conversation of boundary management from its classic
starting point of information processing (Lawrence and
Lorsch 1967, Galbraith 1973) and beyond its more con-
temporary focus on coordination (Malone and Crowston
1994, Gittell 2001). This examination will provide ana-
lytic descriptions of the varying circumstances possible
at boundaries and the processes involved in managing
knowledge across them. This effort will help resolve
the incompatibility between three different perspectives
of boundaries: an information processing approach that
focuses on knowledge as a thing to store and retrieve, an
interpretive approach that emphasizes the importance of
a common meaning to share knowledge between actors,
and a political approach that acknowledges how different
interests impede knowledge sharing. The analytic effort
to integrate these often incompatible perspectives sets
up the practical purpose of the paper—to describe the
different processes required at each type of boundary to
effectively manage knowledge.
To do this, a framework is developed that describes
three progressively complex boundaries—syntactic,
semantic, and pragmatic (see Shannon and Weaver 1949)
and three progressively complex processes—transfer,
translation, and transformation. A focus on the effective-
ness of managing knowledge across boundaries clarifies
that the relationship between actors is one where they not
only share their knowledge, but also assess each other’s
knowledge. Implicit in the effectiveness of this effort is
the existence of a common knowledge
1
that actors use
to share and assess each other’s domain-specific knowl-
edge. Acknowledging both domain-specific knowledge
and common knowledge at a boundary provides a use-
ful distinction to better understand the challenges as
actors try to work across domains when innovation is
desired.
The empirical context used to illustrate this frame-
work and understand these challenges will be the bound-
aries that exist among specialized domains in a new
product-development setting. These boundaries are espe-
cially challenging in the early development stages where
the impact of new requirements is hard to determine.
The following description offers a glimpse into the chal-
lenges at this initial stage and introduces the case that
will be used to illustrate the conceptual and prescriptive
value of the framework.
Beta Motors, like any product-development firm, faces
significant challenges in the early design stage of a
new vehicle. The functional groups involved at this
early design stage are vehicle styling, engine/power train,
555
Carlile: Integrative Framework for Managing Knowledge Across Boundaries
556 Organization Science 15(5), pp. 555–568, © 2004 INFORMS
climate control, and safety. The early stage represents
a unique opportunity to specify up front the overall
design constraints of the vehicle. If done well, it results
in fewer downstream problems such as conflicts between
engineering groups, launch delays, costly rework on the
line, and even warranty issues.
Since the 1960s, the most common engineering tool used
to share and assess knowledge across these groups was
the use of a variety of “clay models” of the vehicle’s
design. However, when increasing quality and time-to-
market pressures occurred in the 1980s, conflicts between
groups, cost overruns, and launch delays also steadily
increased. By the early 1990s, Beta Motors spent sig-
nificant resources to develop better methods to man-
age knowledge at this early design stage. In 1995 the
deployment of a new simulation tool proved extremely
successful—yet on its fourth deployment, expensive
delays and quality problems once again arose.
The framework developed will help explain why the use
of the clay models and the later deployment of the simu-
lation tool failed as a common knowledge for the groups
(actors) involved to adequately share and assess each
other’s domain-specific knowledge.
The first section of the paper is primarily a concep-
tual effort to develop the framework. Three relational
properties of knowledge at a boundary are introduced to
describe the varying circumstances that can exist when
managing it. The article then describes three types of
boundaries and their connection to the existing literature
on knowledge and boundaries and how, together, they
address the increasingly complex circumstances possible
at a boundary. In the second section, the case is devel-
oped more fully, with the development of the modeling
tool, its improvement over the use of clay models, yet its
failure in its fourth deployment described. The third sec-
tion focuses on how the presence of novelty required for
innovation clashes with the path-dependent tendency of
knowledge, creating relations between actors that limit
their ability to effectively share and assess each other’s
knowledge. The article then discusses the implications
this integrative framework has on key topics in organi-
zation theory.
2. Describing the Relative Complexity of
a Boundary
To integrate different approaches to managing bound-
aries in organization theory, it is important to describe
the potential range of circumstances or the relative com-
plexity at a given boundary. To develop the frame-
work, the three following properties of knowledge at
a boundary will be discussed: difference, dependence,
and novelty (Carlile and Rebentisch 2003). Difference in
knowledge refers to a difference in the amount of knowl-
edge accumulated, for example, the novice-expert dis-
tinction found in Schank and Abelson (1977) and Hinds
(1999) and/or the difference in the type of domain-
specific knowledge accumulated, such as specialization
in different problem-solving domains found in Weber
(1924/1947) between actors. Creating a complex product
or service often requires differences in the amount and
type of knowledge. This in turn creates differences in
levels of experience, terminologies, tools, and incentives
that are unique to each specialized domain. At Beta
Motors actors in vehicle styling, engine/power train,
climate control, and safety groups specialize in different
kinds of engineering work that demand different types
of knowledge and responsibilities.
This specialization of knowledge goes deeper than just
the actor’s role or identity. The styling group strives to
create a distinctive vehicle in comparison to the com-
petition. The engine/power train group tries to develop
the most efficient but powerful engine possible. Climate
control needs to adequately cool the vehicle in summer
and heat it in winter. Safety must ensure that the vehicle
is as safe as possible for its occupants, as well as for
the occupants of another vehicle with which it may
collide. For this reason knowledge is not only local-
ized but also invested within a given practice. Because
knowledge takes investment—time and resources to
acquire—it should be seen as “at stake,” indicating the
significant costs associated with giving it up and acquir-
ing different knowledge (Carlile 2002). As difference in
the amount and/or type of domain-specific knowledge
increases between actors, the amount of effort required
to adequately share and assess each other’s knowledge
also increases.
The second relational property of knowledge at a
boundary is dependence—Without dependence, differ-
ence is of no consequence. Examples of dependence can
be seen in the relations between coauthors working on
a paper, employees on different stages of a product line,
or a design engineer and a manufacturing engineer in
a product-development setting. Dependence was defined
by Litwak and Hylton (1962) as a condition where two
entities must take each other into account if they are to
meet their goals. Victor and Blackburn (1987) stipulated
how the actions of actors determine their individual pay-
offs or success, specifying the consequential link between
the activities and goals of actors who are dependent on
each other. Extending these insights in the context of
coordination theory, Malone and Crowston (1994) define
coordination as the management of dependence among
activities (tasks) and resources (see also Crowston 1997).
Substituting domain-specific knowledge for resources,
we begin to see some of the complexity that is not always
recognized in a coordination theory inspired by artificial
intelligence—for if knowledge is different in kind, and
not just in degree, then managing dependencies requires
the capacity to develop an adequate common knowledge
as resources and tasks change.
While Thompson’s (1967) categorization of three
types of system-level interdependence (pooled, sequen-
tial, and reciprocal) remains an important perspective on
Carlile: Integrative Framework for Managing Knowledge Across Boundaries
Organization Science 15(5), pp. 555–568, © 2004 INFORMS 557
dependence, the focus of this paper is how this man-
agement is done. For example, even though sequential
interdependence is seen as less complex than reciprocal,
the concern is how such dependencies are managed
when new circumstances emerge. The approach devel-
oped here is more in keeping with Henderson and
Clark’s (1990) concern that one of the biggest chal-
lenges in a complex technology development setting is
making visible the architectural dependencies that are of
consequence.
At Beta Motors the dependence between vehicle
styling and engine/power train recognizes that a bigger
engine raises the level of the hood, while the desired
look and feel embodied in the drawings constrains the
size of the engine. Thus, as the number of dependencies
increase between actors, the complexity and the amount
of effort required to share and assess knowledge at a
boundary also increases. In settings where innovation is
required, the coordination of what is known as well as
making visible the consequences of dependence that are
not currently known are vitally important.
The third relational property of knowledge at a bound-
ary is how novel the circumstances are. In a new
product-development setting, the most obvious source
of novelty is new customer needs that generate new
requirements of the various actors in their specialized
domains. This suggests that the most challenging aspect
of the relational nature of knowledge at a boundary
is that for each actor there is novelty to share with
others and novelty to assess from others. A less-obvious
source of novelty comes when an actor is unfamiliar
with the common knowledge being used to represent the
differences and dependencies between domain-specific
knowledge. When novelty arises there is often a lack
of common knowledge to adequately share and assess
domain-specific knowledge at a boundary. A common
knowledge could be the use of the English language by
actors to communicate and collaborate or more specifi-
cally the use of a prototyping methodology. In any case,
the common knowledge functions as a boundary object
(Star 1989, Carlile 2002), which the actors use to com-
municate across domains. When novelty is present both
the capacity of the common knowledge to represent the
differences and dependencies now of consequence and
the ability of the actors involved to use it become impor-
tant issues.
The word novelty instead of “uncertainty” is delib-
erately used. Uncertainty is an external characterization
(Tsoukas 2001) that all is not known in a given envi-
ronment, whereas novelty suggests no external vantage
point. Novelty underscores the participatory and rela-
tional nature of what an actor needs to share and to
assess when all is not known. Further, unlike uncer-
tainty, novelty doesn’t allow us to take for granted that
what is new is easily recognized as something unknown.
Actors are susceptible to misrecognizing what is novel as
something that is already known (i.e., competency traps
as seen in Levitt and March 1988, Martins and Kambil
1999) or discarding what is novel as irrelevant (Perrow
1994, Arrow 1994). Camerer et al. (1989) have referred
to this tendency as the “curse of knowledge,” which
recognizes the difficulty that actors have in abandon-
ing previous knowledge (i.e., knowledge is “at stake”).
These issues highlight the challenges that actors face in
identifying what is of consequence when novel circum-
stances arise (Weick et al. 1999). For example, if a pre-
vious engine is reused in a vehicle’s redesign, there is
potentially a significant amount of common knowledge
that can be used to share and assess knowledge about
the impact of the engine across styling, engine/power
train, climate control, and safety. However, when a new
engine is used, the amount of common knowledge avail-
able across the groups decreases, limiting their ability to
represent the differences and dependencies that are now
of consequence. So, as novelty increases, the amount of
effort required to adequately share and assess knowledge
also increases.
We can represent these relational properties of knowl-
edge by imagining a boundary as a vector between at
least two actors. It starts at the origin where the differ-
ences and dependencies are known; as novelty increases
the vector spreads, scaling the increasing complexity
and the amount of effort required to manage the bound-
ary. When the relevant differences and dependencies are
known, the reuse of the common knowledge has posi-
tive effects and the path-dependent nature of knowledge
(March 1972) proves beneficial. However, when novelty
increases, the path-dependent nature of knowledge has
negative effects (Hargadon and Sutton 1997) because the
common knowledge used in the past may not have the
capacity to represent the novelties now present (Carlile
and Rebentisch 2003). A problematic scenario that often
arises is when a powerful actor reuses a common knowl-
edge (i.e., path dependency) that constrains the capac-
ity and ability of other actors to represent the novelty
they are facing. Such a mismatch at a boundary puts the
first actor in a more powerful position to represent his
or her domain-specific knowledge. However, without the
development of an adequate common knowledge (i.e.,
capacity) to represent the novelty now present and/or the
ability of the actors to use it, the outcome produced will
generate latent and costly problems in the long term.
3. Developing the Integrative Framework
In this section three different approaches to manag-
ing boundaries in the organization theory and product-
development literatures are discussed and integrated
in single framework. Using the image of the vector
described above, we scale the relative complexity of the
circumstances at a boundary using Shannon and Weaver’s
(1949) three levels of communication complexity: syn-
tactic, semantic, and pragmatic (see Figure 1). While
Carlile: Integrative Framework for Managing Knowledge Across Boundaries
558 Organization Science 15(5), pp. 555–568, © 2004 INFORMS
Figure 1 An Integrated/3-T Framework for Managing
Knowledge Across Boundaries
Increasing
Novelty
Known
PRAGMATIC
Transformation
SEMANTIC
Translation
SYNTACTIC
Transfer
Types of Boundaries
and Boundary
Capabilities
Increasing
Novelty
Known
Actor BActor A
these three terms also have significant history in the
field of linguistics (Cruse 2000), the focus here—on
the effectiveness of sharing and assessing knowledge
across boundaries—broadens the concern from just the
structure, meaning, or use of language, to Shannon and
Weaver’s practical concerns about what is required for
effective communication across domains. In comparison
to Shannon and Weaver, the difference in the use of
the three terms here is that although they acknowledged
all three levels, their mathematical information theory
focused almost entirely at the syntactic level.
3.1. A Syntactic or Information-Processing
Boundary: Transferring Knowledge
The most common label used to describe the movement
of knowledge in organizations is “knowledge trans-
fer” (Winter 1987, Szulanski 1996, Argote 1999). The
concept of “transfer” has its basis in the information-
processing approaches to boundaries in organization the-
ory (Lawrence and Lorsch 1967, Galbraith 1973), the
roots of which stem from Shannon and Weaver’s math-
ematical approach to communication and information
(1949). The practical strength of their approach is its
mathematical capacity to process a syntax (i.e., 0s and 1s
in the case of computer technologies) that adequately
define the relations between sender and receiver at a
boundary. When common lexicon sufficiently specifies
the differences and dependencies of consequence at the
boundary, the boundary proves “unproblematic”; the pri-
mary concern is one of “processing” or transferring
knowledge across it. What is not always acknowledged
from such a perspective are the stable conditions that
allowed a common lexicon to be created and to ade-
quately function as a common knowledge. This failure
often leads to underestimating the effort required when
those stable conditions change.
Given its technical roots, it is not surprising that an
information-processing approach is the dominant view
used to describe managing boundaries in product devel-
opment (Brown and Eisenhardt 1995) and organiza-
tion design more generally (Lawrence and Lorsh
1967, Tushman and Nadler 1978). Further, information-
processing assumptions are the basis of most technology-
based approaches to knowledge management, where the
primary focus has been on the storage and retrieval of
knowledge (Davenport and Prusak 1998). Simply trans-
ferring knowledge, however, proves problematic when
novelty arises because the current lexicon is no longer
sufficient to represent the differences and dependencies
now present. The limitation of an information-processing
approach occurs because the processing of a common
lexicon is assumed to be always a sufficient common
knowledge (Reddy 1979). So while a common lexicon
is always necessary, it is not always a sufficient type of
common knowledge to share and assess domain-specific
knowledge. In the case of Beta Motors, as time-to-market
and quality pressures grew, the clay model (the actor’s
common lexicon) no longer had the capacity to suffi-
ciently represent the differences and dependencies that
now impacted the design of the vehicle and needed to be
resolved at an early stage.
3.2. A Semantic or Interpretive Boundary:
Translating Knowledge
The transition from a syntactic to a semantic bound-
ary occurs when novelty makes some differences and
dependencies unclear or some meanings ambiguous.
When new requirements and/or new actors are present,
interpretive differences in what a word, measurement,
or outcome means limits the effective management of
knowledge between actors. Researchers who adopt an
interpretive approach recognize how different domains
(i.e., thought worlds) naturally generate interpretive dif-
ferences and so emphasize processes that help create
“shared meanings” (Dougherty 1992) or mechanisms
“to reconcile discrepancies in meaning” (Nonaka and
Takeuchi 1995, p. 67). Research in this vein empha-
sizes the role of cross-functional teams (Ancona and
Caldwell 1992), colocation, and the use of various
shared methodologies (i.e., CAD/CAM, cross-functional
problem-solving templates) to do this. Others have
focused on the role of particular individuals as brokers
and translators who enable the flow of knowledge (Allen
1977, Hargadon and Sutton 1997). Further, the com-
munities of practice literature (Lave and Wenger 1991,
Brown and Duguid 1991) show that as individuals par-
ticipate in similar activities, they develop shared mean-
ings (i.e., Orr’s 1996 study of Xerox repairmen).
Many researchers who focus on knowledge at the
project or firm level have paid attention to the dis-
tinction between tacit and explicit knowledge (Polyani
1966) to recognize the situated and interpretive chal-
lenges of moving knowledge across boundaries (Nonaka
and Takeuchi 1995, Spender 1996). von Hippel and
Carlile: Integrative Framework for Managing Knowledge Across Boundaries
Organization Science 15(5), pp. 555–568, © 2004 INFORMS 559
Tyre’s work (von Hippel 1994, Tyre and von Hippel
1997) calls attention to the stickiness of situated knowl-
edge when trying to move it across different domains.
Nonaka’s description of the process of “externalization,”
making tacit knowledge explicit, has been recognized
as one of the most critical processes organizations need
(Nonaka 1994). All of this research acknowledges the
importance of developing common meaning as a way to
address interpretive differences across boundaries.
Under some circumstances, however, it is not just a
matter of translating different meanings, but of nego-
tiating interests and making trade-offs between actors
(Wenger 1998, Brown and Duguid 2001). Given their
focus on meaning, perspectives that take an interpretive
approach frequently do not specify processes that deal
with different interests and their political consequences.
For example, Nonaka’s process of externalization does
not recognize that in the course of making one’s knowl-
edge explicit, different interests are often revealed that
create barriers to developing shared meanings. Under
these circumstances creating common meaning is not
possible; what is required is a process in which actors
negotiate and are willing to change the knowledge and
interests from their own domain.
At Beta Motors, before the simulation tool, clay
modeling techniques lacked a syntactic and semantic
capacity to represent and discuss the impact of the
engine/power train systems on crash-test outcomes.
Because of that, it did not allow these groups to identify
trade-offs, test potential outcomes, and transform the
vehicle’s design to avoid costly changes and delays
downstream. Today the new simulation tools provide an
opportunity for the engine/power train and safety groups
to agree about the importance of “crash-test data.” Yet
what remains is exploring the consequences those results
have on each other’s work and renegotiating the location
of engine and power train systems to improve the crash-
test results (i.e., an “acceptable” amount of intrusion into
passenger compartment). When different interests arise,
developing an adequate common knowledge is a political
process of negotiating and defining common interests.
3.3. A Pragmatic or Political Boundary:
Transforming Knowledge
The transition from a semantic to a pragmatic boundary
arises when the novelty presents results in different
interests among actors that have to be resolved. What
this recognizes is that when actors have different inter-
ests, the dependencies between them are not indif-
ferent (James 1907)—And these pragmatic differences
generate costs to the actors involved. Under these cir-
cumstances domain-specific knowledge, as well as the
common knowledge used, may need to be transformed
to effectively share and assess knowledge at the bound-
ary. A focus on pragmatic boundaries recognizes that
knowledge is invested in practice and so is “at stake”
for those actors who have developed it (Carlile 2002).
When interests are in conflict, the knowledge devel-
oped in one domain generates negative consequences in
another. Here the costs for any actor are not just the
costs of learning about what is new, but also the costs
of transforming “current” knowledge being used (i.e.,
common and domain-specific knowledge). These costs
negatively impact the willingness of an actor to make
such changes, which helps explain the path-dependent
tendency of actors’ knowledge despite the presence of
novelty.
Research that acknowledges these pragmatic differ-
ences frames knowledge processes as “creative abrasion”
(Leonard-Barton 1995) and focuses on the negotiation
of practice (Brown and Duguid 2001) and the trans-
formation of knowledge (Carlile 2002). In addition to
the importance of teams, this work also recognizes the
role that shared artifacts and methods play in providing
the capacity to negotiate interests and transform knowl-
edge. Boundary objects (Star 1989, Carlile 2002, Bechky
2003) such as drawings, prototypes (Wheelwright and
Clark 1995, Schrage 1999), and “trade-off” method-
ologies (Carlile and Lucas 2003) have proved effective
in providing a concrete means of representing different
functional interests and facilitating their negotiation and
transformation in product-development settings.
For example, in 1990 the engine/power train group at
Beta Motors wanted to place their newest, most power-
ful engine into the vehicle platform. This engine was a
breakthrough for the engine group (representing an out-
come of a sustained effort over several years) because
it produced significant horsepower while still achieving
“good” gas mileage. The problem, however, was that the
size and the shape of the engine caused the hood to go up
much higher than the styling group wanted. Unlike the
1980s when “bulky” trucks were the norm, in the 1990s
the market was demanding and competitors were cre-
ating increasingly “aerodynamic-looking” vehicles. For
the engine/power train group the novelty encountered at
this boundary was a smaller engine compartment. For
vehicle styling, it was a powerful engine that was too
big. Unfortunately the specific consequences of these
dependencies were not well understood in the early
stages of the vehicle’s development. Not only did this
generate costly design changes and delays downstream,
but the political fallout of these costs drove a large
wedge between these two powerful engineering groups.
It was this event the led to the final push to develop a
new engineering tool to address the costly problems of
sharing knowledge across domains.
At a pragmatic boundary actors must be able to repre-
sent current and more novel forms of knowledge, learn
about their consequences, and transform their domain-
specific knowledge accordingly. This knowledge is a
transformed mixture of the knowledge determined to still
be of value and the knowledge that has been determined
of consequence given the novelty present.
Carlile: Integrative Framework for Managing Knowledge Across Boundaries
560 Organization Science 15(5), pp. 555–568, © 2004 INFORMS
3.4. Integrating Different Approaches:
A Theoretical Framework
The purpose of discussing these different approaches
to boundaries in the product-development and organi-
zation theory literatures (see Table 1 for summary) has
been to embrace the accurate insights of each while
avoiding the incompatibilities that arise when changing
circumstances are not specified. In Figure 1 each type
of boundary is categorized, showing the relative com-
plexity of a boundary according to the relational prop-
erties of knowledge discussed earlier. The framework
also identifies different types of processes—transferring,
translating, and transforming—associated with each type
of boundary. At the bottom of the inverted triangle a
syntactic boundary is faced, and when knowledge is
transferred according to a common lexicon, domain-
specific knowledge can be efficiently managed across
the boundary. As novelty arises, new differences and
dependencies exist that need to be identified and their
consequences understood. At a semantic boundary, a
process of learning about and translating domain-specific
knowledge establishes common meanings that become
adequate for the actors involved to share and assess
their knowledge. If a semantic response does not resolve
the problem, then a pragmatic boundary is faced. What
is now required is negotiating and transforming both
the common knowledge and domain-specific knowledge
used in the past. Here common interests are developed
Table 1 Comparative Summary of Approaches to Sharing and Assessing Knowledge Across Boundaries
Syntactic boundary: A transfer or Semantic boundary: A translation Pragmatic boundary: A
information-processing approach or interpretive approach transformation or political approach
Circumstances Differences and dependencies Novelty generates some differences Novelty generates different interests
between actors are known. and dependencies that are unclear— between actors that impede their
A common lexicon is developed different interpretations exist. Common ability to share and assess
that is sufficient to share and meanings are developed to create knowledge. Common interests are
assess knowledge at a boundary. shared meanings and provide an developed to transform knowledge
adequate means of sharing and and interests and provide an
assessing knowledge at a boundary. adequate means of sharing and
assessing knowledge at a
boundary.
Solutions Theory: Information processing Theory: Learning (i.e., communities of Theory: “Creative abrasion” (Leonard-
(Shannon and Weaver 1949, practice)—creating shared meanings Barton 1992)—negotiating practice
Lawrence and Lorsch 1967)— (Dougherty 1992, Nonaka 1994), (Brown and Duguid 2001);
transferring knowledge translating knowledge transforming knowledge (Carlile
2002, Bechky 2003)
Techniques: Syntactic capacity, Techniques: Semantic capacity, Techniques: Pragmatic capacity,
taxonomies, storage and retrieval cross-functional interactions/teams, prototyping and other kinds of
technologies. boundary spanners/translators boundary objects that can be
jointly transformed
Challenges Increasing capacity to process Making tacit knowledge explicit (Polanyi Changing knowledge that is “at stake”
“more” information (Galbraith 1966, Nonaka 1994) (Bourdieu and Wacquant 1992,
1973) Carlile 2002)
A common lexicon is necessary To create common meanings to share To create common interests to
but not always sufficient to and assess knowledge often share and assess knowledge
share and assess knowledge requires creating new agreements. requires significant practical and
across a boundary. political effort.
that allow actors to address the consequences, differ-
ences, and dependencies of each other’s domain-specific
knowledge.
Although the line between each type of boundary is
clearly demarcated in Figure 1, the transition where one
ends and another begins is not often easily identified by
the actors involved. Further, the purpose of the hierar-
chical representation in Figure 1 is to recognize that as
we move up in complexity, the process or capacity at a
more complex boundary still requires the capacities of
those below it. For example, an effective transformation
process also requires the existence of a common lexi-
con and meaning. However, these different levels should
not be seen like “layers in a communication protocol”
as they are in network computing (i.e., the International
Standards Organization (ISO) model). In such a math-
ematically based view of communication the interfaces
between increasingly complex layers can be managed
through common standards and algorithmic processing.
However, this is what breaks down as we move up to
semantic and pragmatic levels—and is what concerned
Claude Shannon as his mathematic approach was applied
above the syntactic level.
At a practical level, this framework can describe the
“mismatches” that can occur between the type of bound-
ary faced and the type or capacity of the process used,
for example, if a syntactic boundary is faced but the
more costly translation and transformation processes are
Carlile: Integrative Framework for Managing Knowledge Across Boundaries
Organization Science 15(5), pp. 555–568, © 2004 INFORMS 561
developed. A second scenario is when a pragmatic or
semantic boundary is faced but only a transfer process is
used. This latter scenario is the most strategically dan-
gerous, because the novelty that is not recognized and
resolved proves very consequential over time (i.e., see
disruptive technologies; Christensen 1997). Implied in
this second scenario is the situation in which powerful
actors have the tendency to reuse their common knowl-
edge. For those actors the path dependency and per-
ceived efficiency of using a common knowledge means
they will not likely recognize the semantic and prag-
matic issues that exist at the boundary (i.e., the curse of
knowledge). Such path dependency is helpful only when
the circumstances at the boundary remain stable.
4. Empirical Case and Approach
This case was developed during a set of follow-up
visits to Beta Motors, where there was an ongoing
research program studying the utilization of knowledge
in product development. Beta Motors is one of the three
largest automobile manufacturers in the world. The case
was collected at one of its design centers in North
America. The focus of the case is the development of
a computational fluid dynamic (CFD) tool with a three-
dimensional modeling technique and its combined use as
a collaborative engineering tool in early design efforts.
The case identified is of particular value because it pro-
vided both an historical as well as a present-day win-
dow into the development of a process of sharing and
assessing knowledge at a pragmatic boundary. Further,
Beta Motors committed significant resources to care-
fully deploying the tool across each setting, so it pro-
vided a weak but naturalized experiment to understand
the particular challenges of establishing and maintaining
an adequate process at a pragmatic boundary.
The data for the case were collected over a two-month
period, including interviews with seven participants and
five follow-up telephone interviews to clarify particular
events with key informants. Participants were selected
from each of the four engineering groups involved and
included the primary developer of the tool and members
of his staff as well. The purpose of this case is not to test
the integrative framework, but to illustrate its value in
connecting the variety of literatures on knowledge and
boundaries and describing how the path dependency of
knowledge and power create mismatches in effectively
managing knowledge across boundaries.
To present the material from this case, I first describe
how the tool was developed and then “successfully” used
on the B-150 vehicle. Next, I discuss the characteris-
tics that define an effective boundary capability at a
pragmatic boundary by linking these characteristics to
the framework developed earlier. Finally, I discuss the
“failed” use of the tool on the B-100 vehicle and sum-
marize what can be learned from this in terms of how
mismatches arise at a boundary.
4.1. Developing and Using the CFD Tool
Given the launch delays and downstream cost overruns
that Beta Motors had been experiencing since the early
1980s, the head of the engineering group was looking
for a better way to manage the early stages of a vehi-
cle’s design. As a member of the advanced vehicle
design group, Bill Knox was asked to develop a tool
to make communication and problem solving more
effective across the four major groups—vehicle styling,
engine/power train, climate control, and safety. What
made this task particularly difficult was that the knowl-
edge developed by these different groups had to be
invested in meeting the requirements within their spe-
cialized domains—yet it was dependent on others to
generate an effective joint outcome. For example, styling
wants to create an aesthetically distinctive vehicle design
that differentiates itself from competitors’ current prod-
ucts, which in turn defines the shape of the overall look
and feel of the vehicle. The engine/power train group has
horsepower requirements and fuel economy constraints
that affect the weight of the vehicle and slope of the
hood. The climate control group has to make sure that,
given the engine size, the vehicle can stay cool in the
summer and warm in the winter with a grill size that
doesn’t compromise aesthetics. The safety group has
concerns about the placement of bumpers and the loca-
tion of the engine to limit collision damage to the car
and other vehicles involved. For Bill, “knowing the limi-
tations of a design is about understanding the interdepen-
dencies across the requirements of the groups involved.”
With the increasing time-to-market and quality pres-
sures in the 1980s, a clay model could not adequately
represent differences and dependencies that now needed
to be identified and resolved at this early stage. How-
ever, because clay models were the primary tools used
by the powerful vehicle styling group and were devel-
oped early in the redesign of a vehicle, they remained
the dominant tool used to shape the design of the car.
This was not only due to the engineering culture at Beta
Motors, but also reflects that clay models represented the
look and feel important to distinguish the design in rela-
tion to the look and feel of the competitor’s vehicle. For
Bill Knox, what was missing was a tool that the groups
could use to represent “the major constraints that would
go into defining the overall ‘design space’ of a vehi-
cle.” As a specialist in aerodynamics, Bill had used CFD
techniques to assess the aerodynamics and fuel economy
of many vehicles. His thought was to use these tech-
niques to represent these cross-disciplinary requirements
and “improve joint problem solving.” To do this, he also
added a three-dimensional modeling technique to repre-
sent the “overall skin” to assess the impact of different
components on the vehicle’s shape. At a technical level,
he believed that representing the “skin” of the vehicle
and measuring its “drag coefficient” trade-offs across the
different functional domains could be better managed.
Carlile: Integrative Framework for Managing Knowledge Across Boundaries
562 Organization Science 15(5), pp. 555–568, © 2004 INFORMS
Bill often made reference to a “firemen’s tarp” as
the collaborative image that he saw the engineering tool
playing. A firemen’s tarp was used in the early twenti-
eth century to catch individuals jumping from burning
buildings. Bill always stated that the effectiveness of the
tarp came from three things. First, it had to be made of a
strong material and fashioned in a way that allowed each
fireman to easily hold and use it. Second, it needed to be
held by several firemen pulling as hard as they could in
different directions to safely break an individual’s fall.
And third, the firemen had to constantly look up at the
individual and make adjustments to ensure the individ-
ual jumping landed safely in the middle of the tarp. Bill
used this image as a guide as he set out to create an
engineering tool that would establish a “shared way for
each group to pull hard in their own direction and still
make good trade-offs.”
Over a period of four months Bill and four other
engineers consulted each group to understand the design
requirements faced within each specialized domain. This
information provided the technical basis for how each
group’s requirements would be specified using the mod-
eling tool. They also created a format where data such
as size, shape, movement, material, heat, and other toler-
ances could be specified. Bill and his groups also spent
time meeting with each group to check if the model did
indeed reflect their technical requirements, as well as
whether it had the capacity to test their intentions as they
learned more. What was most difficult to identify and
measure were the dependencies across each engineer-
ing domain. For example, what are the consequences of
putting in a larger engine on the shape, weight, cooling
requirements, gas mileage, and crash-test outcomes? At
this early stage they used the “skin” of the car and its
“drag coefficient” as measures of dependence to repre-
sent and assess the knowledge being used to generate the
vehicle’s design. Although Bill and his group of engi-
neers were careful to use the language of each engineer-
ing group, they also had to develop a language that was
familiar to all the groups involved.
Once Bill and his group felt comfortable that they
were able to represent the design requirements of each
group, they spent the next two months teaching individ-
uals from each of the four groups how to use the tool,
developing a common language, and making changes
to the tool that reflected concerns. The biggest chal-
lenge they faced was not technical; it was getting all
of the groups involved to recognize the benefits of the
tool despite some of the control that each group felt
it was giving up (i.e., discretion, Barley 1986)—with
the historically more powerful groups feeling like they
were giving up the most control (i.e., vehicle styling,
engine/power train).
After seven months of developing and fine tuning, the
CFD modeling tool was given to the B-150 redesign
team. The effectiveness of the tool was significant, both
in terms of reducing engineering time by 30% and
costs by 50% at the early stage, but more importantly
the downstream implementation of the B-150’s redesign
avoided the major rework costs and launch delays that
had occurred on its last two redesign efforts. Over the
next year, the CFD modeling tool was used in three other
redesign efforts. Two had similar successes to the B-150;
the last, the B-100, had similar savings in up-front
engineering time and costs, but when the design was
implemented, significant problems were identified down-
stream. The outcome was that launch delays and costs
were as high as previous redesign efforts on the B-100.
4.2. The Characteristics of a Process at a
Pragmatic Boundary
To explain the different outcomes of the CFD in the
B-150 and the B-100, four characteristics describe the
capability required at a pragmatic boundary (see Fig-
ure 2). The first characteristic is the development of a
common lexicon that actors use as they share and assess
each other’s knowledge (see Figure 2, #1). When Bill
and his colleagues developed the modeling tool, they
were able to establish a base common language that they
could use to specify critical differences (i.e., size, shape,
geometry, weight, functionality, etc.) and dependencies
measured (i.e., drag coefficient and “skin” of the car). In
the case of the B-150, the lexicon developed and embod-
ied in the CFD tool served as a basis for jointly identify-
ing what was consequential. A climate control engineer
who worked on the B-150 summed this up nicely when
he said, “We disagree sooner and know what we are
disagreeing about more productively since we have a
shared way to compare our design parameters.” In Fig-
ure 2, the characteristic of a common lexicon (#1) is
placed at the bottom of the triangle to indicate its foun-
dational role in supporting the other three characteristics.
As indicated before, when novelty is present, a syntac-
tic capacity remains necessary, but is not an adequate
capacity to share and assess knowledge.
The second characteristic required to share and assess
knowledge across a pragmatic boundary is that the actors
involved need the ability to identify and learn about
new differences and dependencies between them when
novelty is present (See Figure 2, #2). For example, on
the B-150 redesign team, each of the four groups had
very different requirements and design preferences. The
vehicle styling group wanted a new “sleek” design; the
engine/power train group preferred a newly developed,
higher horsepower engine; the climate control group
wanted a larger grill to solve past problems; and the
safety group wanted to see a bumper design no higher
than 16.5 inches from the ground to minimize colli-
sion damage to other cars. With these design preferences
specified, the dependencies across them also need to
be identified and assessed. The CFD model represented
these dependencies and identified their consequences
Carlile: Integrative Framework for Managing Knowledge Across Boundaries
Organization Science 15(5), pp. 555–568, © 2004 INFORMS 563
Figure 2 3-T Framework and the Four Characteristics of a “Pragmatic” Boundary Capability
PRAGMATIC
Transformation
SEMANTIC
Translation
SYNTACTIC
Transfer
1
2
3
4
Supports an iterative approach where actors
get better at developing an adequate common
knowledge for sharing and assessing each
other’s knowledge.
3. A pragmatic capacity establishes common
interests for making trade-offs and
transforming domain-specific knowledge.
A semantic capacity develops common meanings
for identifying novel differences and
dependencies and translating domain-specific
knowledge.
A syntactic capacity requires the development of
a common lexicon for transferring domain-
specific knowledge.
4.
2.
1.
through a simulation and measurement of the vehicle’s
“skin” and drag coefficient. For example, representing
the consequences of a more powerful engine showed
how it affected the slope of the hood, increased the
vehicle’s weight and grill size, and changed the loca-
tion of the bumper in a concrete and rapid manner that
would have been impossible using the clay model. The
CFD tool provided a semantic capacity at the boundary
where each group could identify the novelty present and
translate their knowledge to develop common meaning
and agreements about what is of consequence. When
only a semantic boundary is faced, the groups involved
do not have to transform their domain-specific knowl-
edge; rather, the development of a common meaning is
enough for them to effectively share and assess their
knowledge.
The third characteristic required at a pragmatic bound-
ary is a transformation of actors’ domain-specific knowl-
edge so they can work effectively together (see Figure 2,
#3). Being able to propose, negotiate, and transform
knowledge lies at the heart of trial-and-error problem
solving at a pragmatic boundary. The groups involved
must be able to “try on” alternatives and make trade-
offs; new agreements can now be created to adequately
share and assess knowledge at a boundary. Like the
example of the firemen’s tarp, the ability of actors to
change their own and other’s knowledge only emerges
when there is a pragmatic capacity, a way of representing
the consequences of how the knowledge of one group
generates consequences on the knowledge of another
group, and then making changes accordingly. By using
the CFD modeling tool, each group could first represent
their various concerns, data points, and requirements,
then engage each other to identify, negotiate, transform,
and verify the knowledge that they would then use to
design the vehicle at this early stage. This assessment
criterion is usually a collectively defined measurement
of cost, quality, time, or, in this case, the particular drag
coefficient negotiated and the shape of “skin” eventually
reached. Although the existence of a common measure-
ment (as a lexicon) is required in using the tool, it is
important to realize that the process of developing com-
mon meaning and interests cannot be made as explicit.
The fourth and last characteristic of managing knowl-
edge at a pragmatic boundary is that it requires multiple
iterations (see Figure 2, #4). Addressing the conse-
quences cannot be resolved with one try, but requires
an iterative process of sharing and assessing knowl-
edge, creating new agreements, and making changes
where needed. As the actors participate in each iterative
stage, they get better at identifying what differences and
dependencies are of consequence at the boundary; they
improve at collectively developing more adequate com-
mon lexicon, meaning, and interests. Through this itera-
tive capacity the invested and path-dependent nature of
knowledge can be transformed. For example, on previ-
ous redesign efforts on the B-150, sufficient heating and
cooling power of the climate control system had always
been a significant source of expensive downstream
design rework. The requirements of the climate control
group had always been sandwiched between two pow-
erful engineering groups: vehicle styling, which defined
the size grill, and engine/power train, whose engine gen-
erates most of the heating and cooling requirements.
However, because of the iterative capacity of the CFD
tool to compare design requirements, identify conse-
quences, experiment, and make trade-offs, the B-150
redesign was the first time that the grill (and its heat-
ing and cooling power) on the production vehicle fol-
lowed the same overall specifications defined in the early
design stage. The representational capacity of the CFD
tool and the ability of the actors to use it helped reduce
costs and delays on the B-150.
Carlile: Integrative Framework for Managing Knowledge Across Boundaries
564 Organization Science 15(5), pp. 555–568, © 2004 INFORMS
4.3. Understanding the Failure of the CFD Tool
Bill became concerned when downstream delays and
costs rose on the B-100, despite the use of the CFD
tool. At first he thought that the tool had not been used
correctly because of training or resource limitations.
A resource audit was conducted to assess staffing and
training levels at the early design stage of the B-100, but
no glaring problems were identified. Bill then turned his
attention to the biggest source of the downstream costs
and delays faced on B-100—is failure on three out of
four primary crash tests. Given the increased safety con-
sciousness of the public in the 1990s, crash-test reports
had become a basic requirement in a competitive market.
This would indeed be an expensive problem to fix—But
what were the sources of this problem?
First, the B-100 is a smaller platform version of the
B-150—The gross design and layout of major compo-
nents were comparable. This fact was voiced by many
members of the B-100 team along the following lines:
“It was a similar vehicle, just on a smaller platform.”
One assumed similarity that was not questioned thor-
oughly enough was how a “shrunken” design from the
B-150 would perform on crash tests; therefore, some of
the differences in the size and layout of the B-150 and
the B-100 proved significant. Most critically, the crash
impact of the bumper in relation to large structures of
the vehicle, such as the frame, engine, and other large
components, was not reconsidered for a smaller plat-
form. In short, the “crash physics” and “trajectories of
force” of the B-150 design generated negative impacts
on the B-100 that were not identified in the early design
stage.
Second, because of the technical similarity between
the two vehicles, very early on in the use of the CFD
tool the deployment engineer identified a drag coeffi-
cient of 0.33 as a good target because it had been one
of the “successful” measures reached at the end of the
B-150 redesign. Given the design similarities between
the two platforms, the deployment engineer and others
on the B-100 redesign team believed that “the target
of 0.33 was technically a justifiable place to start for
the B-100 redesign.” There where other specifications
that were also similar to the B-150 that were easily
identified and quickly agreed on; together, these speci-
fications reinforced the assumed adequacy of 0.33 as a
critical design constraint. The consequences of specify-
ing this technical target at the beginning anchored and
overly constrained further efforts to identify and learn
about potential drawbacks of the current approach to the
design. This proved particularly consequential because
the drag coefficient was one of the two primary “mea-
surements” of the dependencies and the potential conse-
quences that each group placed on the other.
Third, although the safety engineer was experienced in
vehicle design, he was new to the B-100 platform. There
was a fair amount that he had to learn before he could
really determine what novelty he had to share and what
novelty was important for him to assess to determine any
negative consequence the current design had on safety
issues. This lack of experience worked against the safety
engineer in ways that he did not even realize because of
the constrained nature of the process described above.
Because a critical measure of dependence (0.33) was
determined so early, many of the differences and depen-
dencies that the safety engineer needed to represent to
others were either not identified or, if identified, not suf-
ficiently solved to identify all of the consequences that
eventually proved so costly downstream (i.e., crash-test
failures).
Even though the engineering tool used and the
numeric outcomes reached were the same (i.e., 0.33 drag
coefficient) across the B-150 and B-100, the process of
getting there was not. On the B-100 the iterative process
of representing, learning about what was most conse-
quential, and transforming the design was constrained
and truncated by the assumed similarities. The down-
stream costs that arose on the B-100 platform resulted
from reusing a key piece of knowledge developed on
another platform; conceptually the lexicon of 0.33 stood
in the way of recognizing and resolving the lack of com-
mon meaning and interests on the B-100. More prac-
tically this limited the capacity and the ability of the
actors—particularly the safety engineer—to represent
and resolve the consequences of the novelty present.
This set of features led to the “mismatch” at this prag-
matic boundary and the significant downstream costs
that resulted.
5. Discussion
The purpose of this paper has been to revisit the topic of
boundaries through an examination of managing knowl-
edge across specialized domains when innovation is a
desired outcome. To address the varying circumstances
that are possible at a boundary, the following relational
properties of knowledge at a boundary were discussed:
difference, dependence, and novelty. This provided a
language to talk about the increasing effort required
to adequately share and assess domain-specific knowl-
edge as circumstances at a boundary grow more com-
plex. The categories syntactic, semantic, and pragmatic
were used to scale the increasing complexity of man-
aging a boundary and were respectively linked to the
information-processing, interpretive, and political per-
spectives in organization theory. The framework devel-
oped (Figure 1) helped to recognize and resolve many
of the incompatibilities that these perspectives in orga-
nization theory often present.
Focusing on the effectiveness of managing knowl-
edge across boundaries clarified the distinction between
domain-specific knowledge and common knowledge at
a boundary. Acknowledging the importance of matching
Carlile: Integrative Framework for Managing Knowledge Across Boundaries
Organization Science 15(5), pp. 555–568, © 2004 INFORMS 565
the capacity of the common knowledge (common lexi-
con, meaning, and interests) with the type of boundary
faced, as well as making sure that the actors involved
have the ability to use that common knowledge or exper-
tise (Black et al. 2004), provided a “prescriptive” plat-
form to address the challenges of managing knowledge
across boundaries. The capability to manage knowledge
at a boundary can be framed as a combination of capac-
ity and ability—capacity times ability equals capability.
Complicating this rough formula is the recognition that
actors tend to reuse knowledge even when novelty is
present. This invested, path-dependent tendency is the
source of the competency traps and the “curse of knowl-
edge” that generate mismatches between actors.
The conceptual and prescriptive value of the frame-
work helps resolve such paradoxes as to why knowledge
is often sticky within the silos of a firm but leaks to
competitor firms (Brown and Duguid 2001). The frame-
work easily outlines a number of empirical questions as
to why this is the case: Across what type of boundary is
knowledge sticky and at which is it leaky? What is the
amount of difference, dependence, and novelty at each?
What is the existing capability (capacity times ability)
present at each boundary? Such a paradox could also be
seen in the case. In the 1960s the clay model provided an
adequate common knowledge to share and assess knowl-
edge at the early stage of a vehicle’s design, but with the
competitive changes in 1980s and 1990s, it no longer
had the capacity to share and assess the knowledge that
had to be identified and resolved. With the development
of the CFD tool and its application to the B-150, we
see a representational tool that had an adequate capac-
ity to represent the relational properties of knowledge at
this early stage and significantly reduce the delays and
cost overruns. The CFD tool used on the B-100 platform
still had sufficient representational capacity, but the team
assumed that a critical measure of dependence (0.33)
was an adequate starting point. As a common lexicon of
dependence, it constrained the process the actors used to
share and assess each other’s knowledge, limiting their
ability to identify the novelty present and understand its
consequences (i.e., crash-test physics on a smaller plat-
form). This constrained the identification of the conflicts
that would arise as well as the development of common
meaning and interests to resolve them.
Understanding the different capacities of common
knowledge (i.e., lexicon, meaning, and interests) and the
abilities of the actors involved to use them improves
our understanding of when an object can function as
a boundary object (Star 1989). The distinction between
types of boundaries reminds us that depending on the
type of boundary faced, boundary objects with different
capacities are required. For example, at an information-
processing boundary where the circumstances are stable,
a common lexicon (i.e., price, specifications, 0s and 1s)
is a sufficient boundary object, whereas the lexicon
of 0.33 for the B-100 was not. As seen in the case
discussed, when a semantic or pragmatic boundary is
faced, the ability of the actors to use the boundary object
can no longer be taken for granted. The majority of
empirical research on boundary objects only describes
their benefits (Bechky 2003, Henderson 1999, Ancona
et al. 2001), not how and why they function the way
they do. This framework explains why a given bound-
ary object is no “magic bullet” (Carlile 2002) when it is
used in a situation where its capacity as a type of com-
mon knowledge (lexicon, meaning, interests) and/or the
ability of the actors to use it is not well matched.
A relational understanding of knowledge at a bound-
ary also specifies at a very concrete level the relationship
between knowledge and power (Foucault 1980, Hardy
and Clegg 1996). Even when actors have equal ability
to use a common knowledge to effectively share and
assess each other’s domain-specific knowledge, power is
still being expressed. In these circumstances the “rela-
tive” power of each actor to represent differences and
dependencies to each other is roughly matched (Black
et al. 2004); thus, the knowledge and power each actor
uses does not generate negative consequences. However,
when abilities to use the common knowledge are not
equal or the common knowledge used does not have the
capacity to represent a particular actor’s knowledge and
interests, mismatches arise. Extending this, in complex
processes in our society (i.e., product development, pub-
lic policy development, etc.) specialized knowledge is
distributed across different domains and cannot always
be equally represented at the same time. This temporal
dimension of dependency means that the consequences
of downstream knowledge generally have a harder time
being represented earlier in the process, putting upstream
knowledge (i.e., designing a product or policy) in a polit-
ically stronger position relative to downstream knowl-
edge (i.e., building the product or implementing the
policy). Given this, we should not assume the actors
involved at a boundary occupy politically equal posi-
tions in representing their knowledge to each other. This
is why the clay model continued to be the dominant
method used at this complex boundary even into the
1990s; it represented what was at stake for the most
upstream and powerful group, vehicle styling and their
marketing champion. In the 1990s safety was the newest
group and was in the weakest position to represent its
domain-specific knowledge. This is why the effective-
ness of an outcome at a pragmatic boundary is based on
the capacity of the common knowledge and the ability
of the actors involved to use it.
As with any research there are limitations in this work;
three in particular require specific mention. First, the
case itself, although illustrative, does not in any way
test the framework being developed. However it should
be pointed out that the development of this frame-
work arises out of several empirical studies on sharing
Carlile: Integrative Framework for Managing Knowledge Across Boundaries
566 Organization Science 15(5), pp. 555–568, © 2004 INFORMS
and managing knowledge across specialized domains
(Carlile 2002, Carlile and Rebentisch 2003, Black et al.
2004); thus, the major concepts in the framework have
been empirically and conceptually refined. Second, this
work emphasized the desirability of convergence rather
than divergence at a boundary. Given the time horizons
and performance constraints that product-development
organizations face, convergence is a desired outcome.
However, in more loosely coupled systems such as sci-
ence and the arts, a certain amount of divergence gener-
ates more robust outcomes in the long term (i.e., physics;
see Galison 1999). Last, as with any work that attempts
to conceptualize and incorporate insights from across
different disciplinary domains, the effort is not without
its rough edges. To integrate different perspectives, the
framework simplifies some issues while stressing oth-
ers. This effort can be seen as a strength or weakness,
depending on how broad an approach to boundaries one
takes. The irony of this is that this work is itself an
example of how hard it is to work across boundaries—to
produce a common knowledge that can be of value in a
multidisciplinary field such as organization theory.
6. Implications
The fact that most innovation occurs at the boundaries
between specialized domains (Leonard-Barton 1995)
tells us that effectively managing knowledge across
the various types of boundaries in an organization is
what drives competitive advantage. Applying this frame-
work to strategic questions provides a concrete way
to describe core concepts such as dynamic capability
(Teece et al. 1997), where the stated concern has been
how to change old knowledge to create new knowledge
in a firm. The case described the capability of this pro-
cess at a pragmatic boundary and provided an example
of what is required (or lacking). Failures occurred when
the actors involved did not have sufficient capacity or
ability to manage the novelty that was present. At a firm
level, a dynamic capability can be thought of as a collec-
tion of different combinations of capacities and abilities
that can be used to share and assess knowledge across
the various types of boundaries. From this vantage point,
instead of seeing the firm as a bundle of resources
(Barney 1991), it can be more completely described as a
bundle of different types of boundaries where knowledge
must be shared and assessed. This framework provides
a concrete description of such boundary capabilities
(“capacity times ability”), a potentially fruitful ground to
begin linking a firm’s organizational and strategic views.
This work is also an addition to the knowledge-based
view of the firm (Grant 1996, Spender 1996) that has
also been suggested as a connection between organiza-
tional and strategic views. The benefits of this work is
that it provides conceptual specificity to knowledge as
both a thing (content) and a process, and describes the
consequences of path dependency and power that have
not been addressed by knowledge-based views.
More broadly, the potential conceptual value of the
framework is its application to a number of classic and
more contemporary topics such as differentiation and
integration (Lawrence and Lorsch 1967), boundary span-
ning (Allen 1977), organization design (Galbraith 1973,
Tushman and Nadler 1978), absorptive capacity (Cohen
and Levinthal 1990), and modularity (Baldwin and Clark
1999). For each, the question of boundary management
when novelty arises must be addressed; however, what is
generally left unspecified by these topics is what type of
boundaries are being faced, what is required to develop
an adequate common knowledge, and how the current
capacities and abilities might need to be changed to
address the novelty now present. The challenge for any
perspective that seeks to explain how systems adapt over
time remains understanding the capacities and the abil-
ities of actors to make the necessary trade-offs between
the knowledge that was used before and the novelty
present to create something new.
This has implications for a broad definition of actors,
whether they are individuals, groups, organizations, or
even computer-based agents. This challenge, however, is
magnified by the tendency of actors to reuse knowledge,
which limits their capacity and ability to represent dif-
ferences and dependencies when novelty is present. This
position of power—at least in the short term—whether
exercised consciously or not, makes it harder for novelty
to be represented and the consequences of it understood.
These practical and political challenges need to be better
recognized to explain how to generate innovation when
knowledge must be shared and assessed by actors across
different domains.
Acknowledgments
First, the author would like to thank Sim Sitkin and Ann
Clark at Organization Science and the two anonymous review-
ers for their expert help. Second, he would like to thank his
colleagues at MIT who helped him improve the ideas in this
paper: Lotte Bailyn, Rebecca Henderson, Wanda Orlikowski,
Eric Rebentisch, John Van Maanen, and Eleanor Westney.
Last, he would like to thank MIT’s Center for Innovation in
Production Development for help in funding this research.
Endnote
1
“Common knowledge” is used in the more technical sense
(i.e., communication theory) of referring to a shared body of
knowledge that allows for communication between actors (see
also “mutual knowledge,” Cramton 2001).
References
Allen, T. 1977. Managing the Flow of Technology. MIT Press,
Cambridge, MA.
Ancona, D., D. Caldwell. 1992. Bridging the boundary: External
activity and performance in organizational teams. Admin. Sci.
Quart. 37 634–665.
Carlile: Integrative Framework for Managing Knowledge Across Boundaries
Organization Science 15(5), pp. 555–568, © 2004 INFORMS 567
Ancona, D., G. Okhuyson, L. Perlow. 2001. Time-out: Taking time
to integrate temporal research. Acad. Management Rev. 26
512–529.
Argote, L. 1999. Organizational Learning: Creating, Retaining, and
Transferring Knowledge. Kluwer, Norwell, MA.
Arrow, K. 1994. Methodological individualism and social knowledge.
Amer. Econom. Rev. 84(2) 1–9.
Baldwin, C., K. Clark. 1999. Design Rules, Vol. 1: The Power of
Modularity. MIT Press, Cambridge, MA.
Barley, S. R. 1986. Technology as an occasion for structuring: Evi-
dence from observations of CT scanners and the social order of
radiology departments. Admin. Sci. Quart. 31 78–108.
Barney, J. B. 1991. Firm resources and sustained competitive advan-
tage. J. Management 17 99–120.
Bechky, B. 2003. Sharing meaning across occupational communities:
The transformation of understanding on the production floor.
Organ. Sci. 14 312–330.
Black, L., P. R. Carlile, N. Repenning. 2004. Expanding theoretical
insights from ethnographic evidence: Building on Barley’s study
of CT-scanning implementations. Pending publication.
Bourdieu, P. 1980. The Logic of Practice. Cambridge University
Press, Cambridge, U.K.
Bourdieu, P., L. Wacquant. 1992. An Invitation to Reflexive Sociology.
University of Chicago Press, Chicago, IL.
Brown, J. S., P. Duguid. 1991. Organizational learning and
communities-of-practice. Organ. Sci. 2 40–57.
Brown, J. S., P. Duguid. 2001. Knowledge and organization: A social-
practice perspective. Organ. Sci. 12 198–213.
Brown, S., K. Eisenhardt. 1995. Product development: Past research,
present findings, and future directions. Acad. Management Rev.
20 343–378.
Camerer, C., G. Lowenstein, M. Weber. 1989. The curse of knowl-
edge in economic settings: An experimental analysis. J. Political
Econom. 97 1232–1254.
Carlile, P. 2002. A pragmatic view of knowledge and boundaries:
Boundary objects in new product development. Organ. Sci. 13
442–455.
Carlile, P., W. Lucas. 2003. Taking care of complex boundaries:
Knowledge and boundary activities on technology development
teams. Working paper, Sloan, MIT, Cambridge, MA.
Carlile, P., E. Rebentisch. 2003. Into the black box: The knowledge
transformation cycle. Management Sci. 49 1180–1195.
Christensen, C. M. 1997. The Innovator’s Dilemma: When New Tech-
nologies Cause Great Firms to Fail. Harvard Business School
Press, Boston, MA.
Cohen, W., D. Levinthal. 1990. Absorptive capacity: A new perspec-
tive on learning and innovation. Admin. Sci. Quart. 35 128–152.
Cramton, C. 2001. The mutual knowledge problem. Organ. Sci. 12
346–371.
Crowston, K. 1997. A coordination theory approach to organizational
process design. Organ. Sci. 8 157–175.
Cruse, D. 2000. Meaning in Language: An Introduction to Semantics
and Pragmatics. Oxford University Press, New York.
Davenport, T., L. Prusak. 1998. Working Knowledge. Harvard Busi-
ness School Press, Boston, MA.
Dougherty, D. 1992. Interpretive barriers to successful product inno-
vation in large firms. Organ. Sci. 3 179–202.
Foucault, M. 1980. Power/Knowledge: Selected Interviews and Other
Writings 1972–1977. Harvester Press, Brighton, Sussex, U.K.
Galbraith, J. 1973. Designing Complex Organizations. Addison-
Wesley, Reading, MA.
Galison, P. 1999. Images and Logic: A Material Culture of Micro-
Physics. University of Chicago Press, Chicago, IL.
Gittell, J. 2001. Supervisory span, relational coordination and flight
departure performance: A reassessment of postbureaucracy the-
ory. Organ. Sci. 12 468–483.
Grant, R. 1996. Toward a knowledge-based theory of the firm. Strate-
gic Management J. 17 109–122.
Hardy, S., S. Clegg. 1996. Some dare call it power. S. Clegg,
C. Hardy, C. Nord, eds. Handbook of Organization Studies.
Sage, London, U.K.
Hargadon, A., R. Sutton. 1997. Technology brokering and innovation
in a product development firm. Admin. Sci. Quart. 42 716–749.
Henderson, K. 1999. On Line and on Paper: Visual Representations,
Visual Culture, and Computer Graphics in Design Engineering.
MIT Press, Cambridge, MA.
Henderson, R., K. Clark. 1990. Architectural innovation: The recon-
figuration of existing product technologies and the failure of
established firms. Admin. Sci. Quart. 44 83–111.
Hinds, P. 1999. The curse of expertise: The effects of expertise
and de-biasing methods on predictions of novice performance.
J. Experimental Psych. 5 205–221.
James, W. 1907. Pragmatism. The American Library, New York.
Lave, J. 1988. Cognition in Practice: Mind, Mathematics, and Culture
in Everyday Life. Cambridge University Press, New York.
Lave, J., E. Wenger. 1991. Situated Learning. Cambridge University
Press, Cambridge, U.K.
Lawrence, P., J. Lorsch. 1967. Organizations and Environments:
Managing Differentiation and Integration. Harvard Business
School Press, Cambridge, MA.
Levitt, B., J. G. March. 1988. Organizational learning. W. R. Scott,
J. Blake, eds. Annual Review of Sociology. Annual Reviews,
Palo Alto, CA, 319–340.
Leonard-Barton, D. 1992. Core capabilities and core rigidities:
A paradox in managing new product development. Strategic
Management J. 13 111–126.
Leonard-Barton, D. 1995. Well Springs of Knowledge: Building and
Sustaining the Sources of Innovation. Harvard Business School
Press, Boston, MA.
Litwak, E., L. F. Hylton. 1962. Interorganizational analysis: A hypoth-
esis on coordination agencies. Admin. Sci. Quart. 6 395–420.
Malone, T., K. Crowston. 1994. The interdisciplinary study of coor-
dination. ACM Comput. Surveys 26(March) 87–119.
March, J. 1972. Model bias in social action. Rev. Ed. Res. 44 413–429.
Martins, L., A. Kambil. 1999. Looking back and thinking ahead:
Effects of prior success on managers’ interpretations of new
information technologies. Acad. Management J. 42 652–661.
Nonaka, I. 1994. A dynamic theory of organizational knowledge cre-
ation. Organ. Sci. 5 14–37.
Nonaka, I., I. Takeuchi. 1995. The Knowledge-Creating Organization.
Oxford Press, Oxford, U.K.
Orr, J. 1996. Talking About Machine. Cornell University Press, Ithaca,
NY.
Perrow, C. 1994. Accidents in high risk systems. Tech. Stud. 1 1–38.
Carlile: Integrative Framework for Managing Knowledge Across Boundaries
568 Organization Science 15(5), pp. 555–568, © 2004 INFORMS
Polyani, M. 1966. The Tacit Dimension. Anchor Day Books,
New York.
Reddy, M. J. 1979. The conduit metaphor. A. Ortony, ed. Metaphor
and Thought. Cambridge University Press, Cambridge, U.K.
Schank, R. C., R. Abelson. 1977. Scripts, Plans, and Goals. Erlbaum,
Hillsdale, NJ.
Schrage, M. 1999. Serious Play: How the World’s Best Companies
Simulate to Innovate. Harvard Business School Press, Boston,
MA.
Shannon, C., W. Weaver. 1949. The Mathematical Theory of Commu-
nications. University of Illinois Press, Urbana, IL.
Spender, J. C. 1996. Making knowledge the basis of a dynamic theory
of the firm. Strategic Management J. 17 45–62.
Star, S. L. 1989. The structure of ill-structured solutions: Bound-
ary objects and heterogeneous distributed problem solving.
M. Huhns, L. Gasser, eds. Readings in Distributed Artificial
Intelligence. Morgan Kaufman, Menlo Park, CA.
Szulanski, G. 1996. Exploring external stickiness: Impediments to the
transfer of best practice within the firm. Strategic Management J.
17 27–43.
Teece, D. J., G. Pisano. A. Shuen. 1997. Dynamic capability and
strategic management. Strategic Management J. 18 509–533.
Thompson, J. 1967. Organizations in Action. McGraw-Hill, New York.
Tsoukas, H. 2001. Re-viewing organisation. Human Relations 54
7–12.
Tushman, M., D. Nadler. 1978. Information processing as an integrat-
ing concept in organization design. Acad. Management Rev. 3
613–624.
Tyre, M., E. von Hippel. 1997. The situated nature of adaptive learn-
ing in organizations. Organ. Sci. 8 71–83.
Victor, B., R. S. Blackburn. 1987. Interdependence: An alternative
conceptualization. Acad. Management Rev. 12 486–498.
von Hippel, E. 1994. The impact of sticky data on innovation and
problem solving. Management Sci. 40 429–439.
Weber, M. 1924/1947. The Theory of Social and Economic Organiza-
tion. A. H. Henderson, T. Parsons, eds. Free Press, Glencoe, IL.
Weick, K., K. Sutcliffe, D. Obstfeld. 1999. Organizing for high reli-
ability: Processes of collective mindfulness. B. Staw, R. Sutton,
eds. Research in Organizational Behavior. JAI Press Inc.,
Stamford, CT, 81–117.
Wenger, E. 1998. Communities of Practice: Learning, Meaning and
Identity. Cambridge University Press, Cambridge, U.K.
Wheelwright, S., K. Clark. 1995. Leading Product Development: The
Senior Manager’s Guide to Creating and Shaping the Enter-
prise. Free Press, New York.
Winter, S. 1987. Knowledge and competence as strategic assets.
D. Teece, ed. The Competitive Challenge. Ballinger, Cambridge,
MA.
