Professional Documents
Culture Documents
Organizational learning has many virtues, virtues which recent writings in strategic
management have highlighted. Learning processes, however, are subject to some important
limitations. As is well-known, learning has to cope with confusing experience and the
complicated problem of balancing the competing goals of developing new knowledge (i.e.,
exploring) and exploiting current competencies in the face of dynamic tendencies to
emphasize one or the other. We examine the ways organizations approach these problems
through simplification and specialization and how those approaches contribute to three
forms of learning myopia, the tendency to overlook distant times, distant places, and
failures, and we identify some ways in which organizations sustain exploration in the face
of a tendency to overinvest in exploitation. We conclude that the imperfections of learning
are not so great as to require abandoning attempts to improve the learning capabilities of
organizations, but that those imperfections suggest a certain conservatism in expectations.
CCC 0143-2095/93/100095-18
© 1993 by John Wiley & Sons, Ltd.
96 D. A. Levinthal and J. G. March
Strategies for exploiting comparative advantage producing manufactured items decrease with the
and competitive opportunities were built on a cumulative number of items produced. It is
conception of calculated rationality. natural to attribute the improvements docu-
This vision of calculative rationality as the mented in studies of manufacturing to knowledge
basis of strategic management continues to be gained from experience. The lessons of experience
the dominant vision, though it has been modified are transferable from one operating unit to
on the basis of various criticisms of its assump- another (Argote, Beckman, and Epple, 1990).
tions, particularly those associated with the They may also spillover from one activity to
availability of information, the information pro- another (Udayagiri and Balakrishnan, 1993).
cessing capabilities of organizations, and the Such experiential-based knowledge can be an
preference axioms of rationality. Because stra- important basis of competitive advantage for a
tegic anticipatory rationality seems to demand firm, and for some students of organizations
both greater cognitive and calculative capabilities and strategy, learning has become a plausible
and more consistency and stability in preferences mechanism substituting for, or augmenting, calcu-
than can be reliably assumed, considerable effort lative rationality in the pursuit of intelligent
has been directed to improving the informational organizational action. The (re)discovery of learn-
and analytical basis for organizational action and ing has been stimulated by the current interest
to developing consistent, stable organizational among students of strategic management in
objectives. Modern decision-oriented information organizational capabilities and knowledge
systems and procedures for defining (or (Prahalad and Hamel, 1990). Successful organiza-
negotiating) goals reflect this spirit (Keen and tions are described as having capabilities for
Morton, 1978; Jones and MacLeod, 1986). learning—for responding to experience by mod-
ifying their technologies, forms, and practices
(Stalk et al., 1992). Executives are enjoined to
The discovery of learning
monitor their experience in order to learn from
At the same time as ideas of strategic calculative it and to organize to stimulate learning and
rationality have been modified and refined, the utilization of knowledge gained from the
learning has been 'discovered' by the world of experience of others as well as their own (Senge,
practice and the academic field of strategic 1990). These enthusiasms supplement—and to
management. As researchers have considered some extent replace—earlier enthusiasms for
the stability of differences in firm performance long-term planning and rational calculation as
in the face of changing business environments, bases for organizational prosperity and survival.
many have come to view the ability to learn as
an important, indeed in some accounts a unique,
source of sustainable competitive advantage Confusions of experience
(Burgelman, 1990; Senge, 1990). This is reflected Studies of reductions in the cost of production
both in the attention to learning and learning associated with the number of units produced do
organizations in management circles (Senge, not, in general, provide direct confirmation of
1990; Stalk, Evans, and Shulman, 1992) and in the processes by which those improvements
the exploration of learning models of adaptation have occurred, nor do they demonstrate that
by economists (Cross, 1983) and students of experiential learning processes inexorably lead
organizations (Argyris and Schon, 1978; Levitt to optimal practices. The limitations of experience
and March, 1988). as an instrument of intelligence are not esoteric.
They stem from relatively generic problems of
Learning from experience adaptive intelligence.
Experience is often a poor teacher, being
Organizations and the individuals in them often typically quite meager relative to the complex
improve their performance over repetitions of and changing nature of the world in which
the same task. Repetition-based improvements learning is taking place. Many of the same
in manufacturing performance have been docu- cognitive limits that constrain rationality also
mented in some detail in numerous studies of constrain learning. Learning from experience
learning curves (Yelle, 1979). The costs of involves inferences from information. It involves
The Myopia of Learning 97
memory. It involves pooling personal experience nor specialization is unique to learning processes.
with knowledge gained from the experiences of They are, however, particularly salient to dis-
others. The difficulties in learning effectively in cussions of the design of learning organizations.
the face of confusing experience are legendary.
Even highly capable individuals and organizations
Simplification and the construction of buffers
are confused by the difficulties of using small
samples of ambiguous experience to interpret Learning presumes interpretation of experience.
complex wodds (Brehmer, 1980; Fischhoff, 1980). Organizations code outcomes into successes and
The cognitive and inferential limitations of failures and develop ideas about causes for
individuals are accentuated by organizational them. Experience is clouded by the interactive
limitations. The interpretations of history are complexity of history, particularly by the way
political, reflecting efforts to assign and evade experience is shaped by many actors simul-
responsibility and to establish favorable historical taneously learning. If one's own actions are
stories (Sagan, 1993). Organizations record the embedded in an ecology of the actions of many
lessons of histories in the modification of rules others (who are also simultaneously learning and
and the elaboration of stories, but neither is a changing), it is not easy to understand what is
perfect instrument. Problems of memory, conflict, going on. The relationship between the actions
turnover, and decentralization make it difficult of individuals in the organization and overall
to extract lessons from experience and to retain organizational performance is confounded by
them (March, SprouU, and Tamuz, 1991). simultaneous learning of other actors. Particularly
in environments in which performance is a noisy
refiection of organizational decisions, highly
Self-limiting properties of learning interactive learning is likely to be unrewarding.
Recent examinations of learning as an adaptive For example, while isolated subunits can often
process have raised questions not only about the learn quite effectively (Cyert and March, 1992;
confusions of experience but also about the ways Lave and March, 1993), simultaneous learning
in which learning is self-limiting. The effectiveness by several interacting subunits in a noisy environ-
of learning in the short-run and in the near ment can be quite difficult (Lounamaa and
neighborhood of current experience interferes March, 1987).
with learning in the long run and at a distance. Organizations that want to disentangle the
Knowledge and the development of capabilities interactions introduced by multiple simultaneous
improve immediate performance, but they often learners have two general options: They can seek
simultaneously reduce incentives for and com- to generate enough experience so that they can
petence with new technologies or paradigms. fit relatively complicated models to the data. In
Learning has its own traps. practice, this is often not feasible. Alternatively,
In the next three sections, we consider the organizations can seek to control the effects of
major mechanisms organizations use to reduce interactions by preventing multiple simultaneous
experimental confusion, the problems of myopia adjustments. One means of increasing the effec-
they face, and the dynamic complications of tiveness of learning is to simplify natural experi-
balancing exploration and exploitation in learning. ence by inhibiting learning in one part of an
organization in order to make learning more
effective in another part (Lounamaa and March,
TWO MECHANISMS OF LEARNING 1987). Organizations seek to transform confusing,
interactive environments into less confusing, less
Organizations use two major mechanisms to interactive ones by decomposing domains and
facilitate learning from experience. The first is treating the resulting subdomains as autonomous.
simplification. Learning processes seek to simplify They create buffers. They enact environments.
experience, to minimize interactions and restrict
effects to the spatial and temporal neighborhood
Decomposition and organizational structures
of actions. The second mechanism is specializa-
tion. Learning processes tend to focus attention Departmentalization is, perhaps, the most basic
and narrow competence. Neither simplification mechanism to mitigate interaction effects in
98 D. A. Levinthal and J. G. March
learning within a complex organization. Although occasion for learning, in other elements of the
the transformation from functional to product system. Customer complaints are not merely
organizations has usually been justified as a absorbed by boundary spanning personnel but
means to enhance control and coordination become more broadly known. Problems in
(Chandler, 1962), it also is a way of segregating production are not masked by a buffer inventory
experience. Less prominent in the normative of partially assembled units or by postproduction
literature on strategy and organizations, but inspection and repair.
prominent in more descriptive accounts (Cyert The apparent discrepancy between the two
and March, 1992) is the sequential allocation of perspectives on buffers can, in part, be resolved
attention to divergent goals. While the sequential by considering the simultaneous difficulties of
allocation of attention is generally viewed as an oversight produced by the way buffers conceal
outcome of goal conflict and bounded rationality, signals of problems and difficulties of misinter-
it also results in a simplification of experiments pretation of signals produced by trying to
in organizational change. understand complex, interactive systems. The
The conception of buffers as enhancing organi- former difficulties suggest tight coupling in order
zational effectiveness has a long tradition in the to facilitate detection of signals; the latter
organizations literature (Thompson, 1967), and difficulties suggest loose coupling in order to
resource or other buffers between units can facilitate their interpretation. The basic argument
achieve substantial simplifications of the learning of those who advocate tight coupling is that many
environment. Buffers between units organizations have gone too far in attempting to
(departmentalization) and between goals segregate the problems they face. The arguments
(sequential attention) allow local consequences to are twofold. On the one hand, they are assertions
be examined. Marketing departments experiment that modern markets and technologies link things
with alternative marketing strategies and pro- together more than earlier ones did. On the
duction departments experiment with alternative other hand, they are also assertions that modern
production strategies, each evaluated in such a analytical and coordination techniques—perhaps
way that their effects on each other are ignored. due to new developments in information
Depending on an organization's structure, global technology—reduce the costs of centralized prob-
problems of poor performance are viewed as lem solving.
local problems of cost reduction or as local Tightly coupled systems are relatively good
problems of revenue enhancement. for system-wide error detection, but they are
It is noteworthy, however, that many contem- relatively poor for error diagnostics. Loosely
porary academics and practitioners advocate the coupled systems make diagnostics easier
tight coupling of organizations (Bower and Hout, (assuming that the system is, in fact,
1988). Perhaps the most prominent application decomposable) but localize error detection, there-
of ideas of tight coupling is the notion of lean by making more general awareness of problems
production systems and just-in-time inventory difficult. The appropriate balance between invest-
systems. The ideas extend beyond manufacturing, ments in error detection and in diagnostics
however. In some cases, the idea of making presumably depends on the frequency of errors
organizations tightly coupled is tied to ideas and the difficulty of diagnosis.
about the importance of linkages to customers
and has produced popular organizational slogans
about the desirability of being 'customer driven.' Decomposition and enactment
In other cases, the ideas have been extended to The conditions under which the buffers of
linkages within the organization (Schonberger, departmentalization or sequential attention lead
1990). to effective local learning are usually described
Advocates of these various mechanisms of in terms of the extent to which the problems
tight coupling suggest that an important virtue that are faced are decomposable, so that relatively
of such structures is that they enhance learning. few interactions occur across departmental bound-
Learning is enhanced because the problems that aries or across goals (Gulick and Urwick, 1937).
arise through ongoing operations in one part of Decomposability is usually treated as an inherent
the system become observable, and hence an property of a problem. Insofar as decomposability
The Myopia of Learning 99
particularly fast in adapting to their children's is an adaptive substitute for search or change in
needs reduce the pressure on the latter to be an activity.
adaptive, resulting in a lack of socialization
(manners) in children of highly adaptive parents.
Or, political/legal systems that meet special or Multiple responses: Exit, voice, and loyalty
changing circumstances through adaptation by The substitution principle has been used by
law enforcement agencies or by courts reduce Hirschman (1970) to account for some features
the pressure on legislatures for changes in the of the development of rail transport in Nigeria,
statutes. public school systems, and other systems that
have dissatisfied participants. In Hirschman's
Multiple mechanisms: Targets, search, and slack framework, participants who experience a decline
in quality of organizational services or products
In bounded rationality search models, an organi- have two possible responses: The first alternative
zation is seen as responding to success or failure is to exit from the unsatisfactory relationship and
by varying the intensity of search, the level of seek another. The second alternative is to try to
organizational slack, and the target (aspiration fix the existing relationship. The two are substi-
level) for performance (Cyert and March, 1992). tutes in the adaptive story of correcting declines
Success decreases search and increases slack in quality. From the vantage point of the
and targets, while failure increases search and dissatisfied participant, either of the alternatives
decreases slack and targets. Changes in search, is satisfactory in the sense that each has a
slack, and targets function effectively as substi- reasonable prospect of removing the difficulty.
tutes for each other. Adjustments in search Exit can substitute for voice, and vice versa.
substitute for adjustments in slack or aspirations, From the point of view of the organization,
and vice versa. The different responses have however, the two alternative responses have quite
equivalent effects from the point of view of different implications. If dissatisfied participants
restoring the aspiration/performance equilibrium, exit, they abandon the organization to less
but they are not necessarily equivalent from the demanding participants, thus condemning it to a
point of view of the organization and its learning. gradual degradation of capabilities. On the other
In particular, there may be substantial differ- hand, if dissatisfied participants exercise voice,
ences in the long run between a system that they encourage the organization to improve
adjusts aspirations slowly and slack rapidly and quality. The organizational problem is to slow
a system that adjusts aspirations rapidly and the exit of quality-conscious participants long
slack slowly. For example, the standard pygmalian enough to use their infiuence in improvement.
story is one in which aspiration adjustments to One solution is found through the encouragement
unsatisfactory performance are slowed by means of loyalty, a form of friction on exit. The
of rosy interpretations of that performance. Hirschman loyalty mechanism can be seen as a
The classic pygmalian complication is that rosy way of slowing search and adjustment of aspi-
interpretations of performance inhibit a down- ration levels in order to increase pressure on
ward adjustment of aspirations, but they also slack.
inhibit reduction of slack and an increase in search
(by underestimating the discrepancy between
aspirations and performance). Multiple, nested options
Preferences also adapt in response to experi- Learning experience is nested. That is, learning
ence (March, 1988). Tastes for opera, ballet, occurs at several different but interrelated levels
and baseball are developed at the same times as at the same time. An organization simultaneously
competencies at those activities, and they are learns which strategy to follow and how to operate
considerably affected by those competencies. within various alternative strategies (Herriott,
Similarly, preferences for particular technologies Levinthal, and March, 1985). An individual simul-
develop in tandem with competencies at them. taneously learns whether to think like an economist
Since propensities to reevaluate the wisdom of and how to think like an economist. An army
engaging in particular activities are reduced by learns which technology to use and how to use
gains in competence at them, preference change several alternative technologies. A business firm
The Myopia of Learning 101
learns which market to enter and how to function form of myopia is the tendency to ignore
effectively in several alternative markets. the larger picture. The near neighborhood is
When learning is nested, learning at one level privileged by organizational learning. As a
is effectively a substitute for learning at another. result, survival of more encompassing systems is
Refining an existing technology substitutes for sometimes endangered. The third form of myopia
recognizing a better one, and vice versa. Strength- is the tendency to overlook failures. The lessons
ening abilities within an existing paradigm substi- gained from success are privileged by organiza-
tutes for finding a new one that is better, and tional learning. As a result, the risks of failure
vice versa. Learning the nuances of an existing are likely to be underestimated.
relationshi:' substitutes for finding an alternative
that is better, and vice versa.
Overlooking distant times
The same thing happens within an organiza-
tional structure. Fast adaptation at one level in There is no guarantee that short-run and long-run
an organization leads to slow adaptation at survival are consistent. It is easy to imagine
other levels. Insofar as operating levels in an situations in which the only strategies that permit
organization make adjustments in implementing survival in the short run assure failure in the long
policies as conditions change, the pressure for run and vice versa. Thus, it is fairly easy to make
changes in policies is relieved. The operating an argument that any consideration of the future
managers of a firm in a changing competitive must accept survival in the short run as a constraint.
environment may adjust by discovering new Simplification and specialization, however, seem
markets for the firm's existing products. For exceptionally myopic with respect to the future.
instance, a defense manufacturer could respond We can illustrate this by looking at the erosion of
to the decline in the U.S. military budget by enactment with time, at the second order effects
pursuing foreign military markets for its wares. of learning substitution, and at some problems
Such adaptation, however, masks a higher associated with knowledge inventories.
level problem that the firm faces. Learning at
the operating level of an organization substitutes
Erosion of enactment
for learning at higher levels. Insofar as customers
adapt to the inadequacies of the products they Learning processes tend to enact environments
use, manufacturers are less likely to do so. that are sufficiently simple to permit inferences
Insofar as subordinates respond to individual and incremental gains. There is, however, a limit
customer complaints, bosses are less pressed to to enactment. The classic tension between social
do so. Lower-level adaptation is a sensible construction of reality and the interventions of
activity that tends to enhance an organization's other reality processes (for example, of nature)
position in its present environment. In the long is well-known. Learning creates a simplified
run, however, such first-order learning can not world and specializes an organization to it. Such
substitute for second-order learning of new models are more likely to capture the central
routines and strategies. elements of past environments than the contin-
gencies of current circumstances. Only the most
enthusiastic observers of enactment deny that
PROBLEMS OF MYOPIA the world of nature constrains social enactment
and sometimes forces reconsideration. Inexor-
By simplifying experience and specializing adap- ably, at some point a mental model becomes
tive responses, learning improves organizational unsustainable, and the organization's competen-
performance, on average. However, the same cies become irrelevant. The process is as familiar
mechanisms of learning that lead to the improve- to modern firms as it was to ancient systems of
ments also lead to limits to those improvements. magic, religion, warfare, and trade.
In particular, we will note three forms of learning
myopia: The first form of myopia is the tendency
Second-order effects of specialization
to ignore the long run. The short run is privileged
by organizational learning. As a result, long run Substitutions of learning in one part of an
survival is sometimes endangered. The second organization for learning in another part are
102 D. A. Levinthal and J. G. March
normally sensible forms of specialized adaptation. extraordinarily difficult and required shutting
They do, however, produce some dysfunctional down the manufacturing facility for a considerable
second-order effects in the form of disparities in period of time.
the development of adaptive capabilities. These
effects typically take longer than do the immediate Traps of power
effects of local learning, involving as they do the
development or decay of skills, procedures, and Organizational power is a short-run aset but
technologies of learning. A strategic problem is potentially a long-run liability. Power allows an
created by the fact that the learning that yields organization to change its environments rather
a comparative advantage in one domain is likely than adapt to them. Thus, firms with strong
to be rewarding in the short run, but it leads to market positions impose their policies, products,
a longer-run potential decay of adaptive capability and strategies on others, rather than learn
in other domains. to adapt to an exogenous environment. This
capability to define an environment—such as a
Traps of distinctive competence firm's capability to set industry standards—
provides an advantage to the organization since
An organization develops better skills in some it can organize around a specific plan without
parts of the organization, in some markets, in concern about contingencies. This advantage is
some technologies, and in some strategies than exploited and improved upon by refining the
in others. The mechanism is one of mutual skills of power.
positive feedback between experience and com- In the long run, however, the use of power to
petence. Organizations engage in activities at impose environments is likely to result in atrophy
which they are more competent with greater of capabilities to respond to change. An organiza-
frequency than they engage in activities at which tion becomes skilled at influencing its environ-
they are less competent. The differences in the ment, but not at responding to the environment
frequency with which different activities are (Deutsch, 1966: 111). Should its ability to
pursued translate into differences in the amount influence the environment be overwhelmed by
of experience at the various potential activities, economic, political, or demographic forces
which in turn translate into differences in beyond its control, the underdevelopment of
competence. These distinctive competencies adaptive skills will be exposed, and there may
invite utilization, which furthers their additional not be enough time to overcome the resulting
development. The self-reinforcing nature of disadvantage.
learning makes it attractive for an individual or
organization to sustain current focus. The result
is that distinctive competence is accentuated, and Knowledge inventories and the problem of
organizations become specialized to niches in timing
which their competencies yield immediate advan- The complications in balancing the long- and
tage. short run are also illustrated by the management
Learners become increasingly removed from of knowledge inventories. Organizations some-
other bases of experience and knowledge and times act by solving problems after they arrive.
more vulnerable to change in their environments They discover problems, diagnose their causes,
(David, 1985). Since the degree to which firms or experiment with solutions to them, and then
individuals learn about alternative opportunities is implement solutions that appear likely to yield
a function of their level of involvement in them favorable outcomes. Such a procedure is implied
(Cohen and Levinthal, forthcoming), knowledge in many theories of decision making and by the
about and use of old competencies inhibit efforts design of many decision support systems. Often,
to change capabilities. Abernathy and Wayne however, organizational action is better seen
(1974) provide a classic illustration of this as a programed exercise of prior capabilities
pathology when they describe Ford's pursuit of (Starbuck, 1983), or as the result of monitoring
efficient production of the Model T. While the environments and drawing appropriate responses
company was able to drive down the cost of the from a prior repertoire (March and Simon, 1993).
Model T, the transition to the Model A was The surveillance/response mode is particularly
The Myopia of Learning 103
likely when response times are short. The Overlooking distant places
time between the anticipation of a problem
and its arrival may not be adequate for an As has been observed often in the study of the
organization to identify and develop the evolution of nested systems, it is relatively
knowledge, or accumulate the experience, unusual for a strategy that maximizes the
required to respond effectively. As Dierickx prospects for survivial of the components of a
and Cool (1989) suggest, there are time system to be the same as a strategy that maximizes
compression diseconomies in building organi- the prospects for the survivial of the system as
zational capabilities. As a result, organizations a whole (March, forthcoming 1994). Strategies
build inventories of competencies (Feldman, of survival for organizations may be optimal
1989). The inventories are represented by neither for survival of the economies or social
storehouses of information and experience systems of which they are a part, nor for the
both within the organization and outside individuals and groups that form the organization.
it. Organizations develop contingency plans.
They stockpile knowledge about products,
technologies, markets, and social and political Selection among learners
contexts. They develop networks of contacts As we have argued earlier, learning gives
with consultants and colleagues. advantage to results in the spatial neighborhood
In a world in which there are only a of current action. Organizations that learn effec-
few possible situations and the appropriate tively become well-adapted to their environments,
responses are stable, maintaining appropriate even as their environments become well-adapted
knowledge inventories is relatively uncompli- to them. When the world changes exogenously,
cated. Normally, those inventories are rep- as inevitably it does, the matches between
resented by a small number of specialized organizations well-adapted to their previous
competencies maintained by the individuals and environments and the new environments are at
groups that make up the organization. Where risk. Existing organizations are likely to die and
situations or proper responses are numerous be replaced by new organizations which will, in
and shifting, it is harder to specify and realize turn, become specialized to the new environment.
optimal inventories of knowledge. By the time This threat to organizational survival is substan-
knowledge is needed, it is too late to gain it; tial, but the resulting cycle of specialization and
before knowledge is needed, it is hard to replacement may well be an efficient system for
specify precisely what knowledge might be the system as a whole, combining as it does the
required or useful. It is necessary to create advantages of learning at the organization level
inventories of competencies that might be used and the advantages of selection at the system
later without knowing precisely what future level. Thus, the 'self-destructive' properties of
demands will be. learning are properties that make the replacement
Determining the variety and depth of knowl- of obsolescent organizations easier. Rigidities in
edge to be added to the inventory is filled with one individual or organization serve to exploit
potential pitfalls. Knowledge that has clear, current knowledge and simultaneously make old
immediate uses is specialized to current techno- markets vulnerable to new entities with new
logies and markets. It is easily specified and capabilities (Hannan and Freeman, 1984). Sys-
has relatively early and local returns. Broader tematic advantages stemming from component
or deeper knowledge is less likely to have vulnerabihty have, of course, long been favorite
immediate pay-off but results in a greater ability topics of evolutionary theorists, and it should be
to adapt to changes. Moreover, knowledge no surprise that they arise here.
facilitates the use of other knowledge. Organiza- Learning is, however, not entirely benign in
tions that have some competence in an emerging its consequences for systems of organizations.
technological domain are better able to assess The fruits of successful exploration, whether
the potential importance of that domain and new technologies, product ideas, or modes of
to evaluate possible investments in new knowl- management, tend to diffuse over populations of
edge in that domain (Cohen and Levinthal, organizations. They are public goods. In contrast,
1990 and forthcoming). the risks and costs of exploration are private
104 D. A. Levinthal and J. G. March
predictor of future success. Consider, for exam- ration - the pursuit of new knowledge, of things
ple, using experiential learning to learn how to that might come to be known. And they engage
avoid or produce an extremely rare event—for in exploration - the use and development of
example, a major nuclear disaster or a major things already known. An organization that
scientific discovery. Experience rarely generates engages exclusively in exploration will ordinarily
a rare event. As a result, most people involved suffer from the fact that it never gains the returns
in nuclear safety are likely to come to believe of its knowledge. An organization that engages
they are more capable of producing a safe exclusively in exploitation will ordinarily suffer
environment than they actually are, and most from obsolescence. The basic problem con-
people involved in scientific discovery are likely fronting an organization is to engage in sufficient
to come to believe they are less likely to produce exploitation to ensure its current viability and,
a major scientific discovery than they actually at the same time, to devote enough energy to
are. Experience probably makes nuclear safety exploration to ensure its future viability. Survival
engineers over-confident and scientific requires a balance, and the precise mix of
researchers under-confident. exploitation and exploration that is optimal is
Research on individual attributions of causality hard to specify.
to events indicates that individuals are more
likely to attribute their successes to ability and
Problems in Maintaining a Balance
their failures to luck than they are to attribute
their successes to luck and their failures to ability Maintaining a balance between exploitation and
(Miller and Ross, 1975). Biases in the perception exploration is complicated not only by the
of the relative contributions of ability and luck difficulty of determining what the appropriate
to outcomes translate into biases in the estimation balance should be, but also by several ways in
of risk. Any incUnation to over attribute outcomes which learning itself contributes to imbalances.
to luck will be associated with overestimating Learning leads organizations into dynamics of
risk, thus with decreasing risk taking. Similarly, accelerating exploitaion or exploration, and learn-
any inclination to overattribute outcomes to ing makes negative as well as positive contri-
abiUty will be associated with underestimating butions to competitive position.
risk, thus with increasing risk taking. As a
result, persistent failure leads to a tendency to
overestimate the risks of actions, and persistent The traps of learning
success leads to a tendency to underestimate Organizations become trapped in one or more
those risks. Successful people have confidence in of several dynamics of learning that self-
their ability to beat the apparent odds. They destructively lead to excessive exploration or
tend to underestimate the risks of their actions excessive exploitation. These dynamic distor-
and overestimate their expected returns (March tions of the exploitation/exploration balance
and Shapira, 1987; Kahneman and Lovallo, are not perverse. They stem from the same
1993). Since organizations promote successful processes of adaptation that lead to effective
people to positions of power and authority, matching of organizational behavior with
rather than unsuccessful ones, it is the biases of environmental conditions (Hedberg, Nystrom
success that are particularly relevant to decision and Starbuck, 1976). They are processes that
making. involve short-term positive feedback on either
exploration or exploitation and thus upset a
balanced attention to both.
THE EXPLOITATION/EXPLORATION
BALANCE The failure trap
The elements of myopia detailed above are Sometimes exploration drives out exploitation.
embedded in a broader problem for adaptive Organizations are turned into frenzies of exper-
intelligence. Organizations divide attention and imentation, change, and innovation by a
other resources between two broad kinds of dynamic of failure. Failure leads to search and
activities (March, 1991). They engage in explo- change which leads to failure which leads to
106 D. A. Levinthal and J. G. March
more search, and so on. New ideas and standard, potentially self-destructive product of
technologies fail and are replaced by other new learning. The trap can be broken by rapid
ideas and technologies, which fail in turn. This upward adjustment of aspirations or by false
pathology is driven by three pervasive features feedback as to the high value of exploration, but
of organizational life: it forms a powerful consequence of learning
1. Most new ideas are bad ones, so most processes.
innovations are unrewarding.
2. The return from any particular innovation, Learning and competitive advantage
technology, or reform is partly a function
of an organization's experience with the There are two characteristic features of learning
new idea. Even successful innovations, that are important to competitive advantage. The
when first introduced, are likely to performfirst is that learning generally increases average
poorly until experience has been accumu- performance. More experienced and more exten-
lated in using them. sively trained individuals or groups will generally
3. Aspirations adjust downward more slowly do better than less experienced or less trained
than they adjust upward and exhibit a ones. The second feature of learning is that it
consistent optimistic bias (Lant, 1992). generally increases reliability. More experienced
These three features can trap an organization in and more extensively trained individuals and
an endless cycle of failure and unrewarding groups produce fewer surprises. Moreover,
change. The cycle of exploration and the failure organizations accumulate experience across indi-
trap can be broken by the introduction of an viduals. They use rules, procedures, and standard
exceptionally good alternative or the relatively practices to ensure that the experiences of earlier
rapid downward adjustment of aspirations, as individuals are transferred to newer members of
might occur in a situation in which all organiza- the organization. This process of routinization
tions experience similar histories of failure. is a powerful factor in converting collective
experience into improved average performance.
The success trap It is also a powerful influence on reliability and
reduces the average amount of deviation from
Sometimes exploitation drives out exploration. normative behavior as an individual or organiza-
The returns to exploitation are ordinarily more tion ages. Learning reduces variability.
certain, closer in time, and closer in space than Competitive advantage is clearly helped by the
are the returns to exploration (March, 1991). improved average performance that learning
Exploratory experiments with new procedures or ordinarily offers. Indeed, this feature of learning
forms are likely to lead to poorer results in the makes it a prime contributor to competitive
short run, and the returns to exploration are advantage. Improved reliability, on the other
likely to be greater for the organization, or hand, is a mixed blessing from the point of view
a population of organizations, than for an of competitive advantage. By increasing the
individual. reliability of individuals and organizations, learn-
Particularly with rapid rates of turnover of ing tends to reduce exploratory deviation. When
decision makers, the uncertain and distant returns we ask whether individuals and organizations
associated with exploration are likely to have that learn will be selected by a competitive
a high discount rate associated with them. environment, we find that the answer is compli-
Furthermore, past exploitation in a given domain cated. Competition can make reliability (and
makes future exploitation in the same domain therefore learning) a disadvantage.
even more efficient. As a result, organizations Consider the following simple model (March,
discover the short-term virtue of local refinement 1991): Assume that survival is based on compara-
and the folly of exploration (Levinthal and tive performance within a group of competitors.
March, 1981). As they develop greater and Each single performance is a draw from a
greater competence at a particular activity, they performance distribution specific to a particular
engage in that activity more, thus further individual or organization. The mean of the
increasing competence and the opportunity cost distribution reflects the individual's or organiza-
of exploration. This competency trap is a tion's ability level and the variance reflects
The Myopia of Learning 107