Industrial and Corporate Change, pp. 1 of 31

doi:10.1093/icc/dtm013

Exploration and exploitation in

product innovation

Henrich R. Greve

A central theoretical problem in organizational evolution is how organizations

acquire new capabilities. Organizational exploitation of current capabilities often

reduces exploration of new capabilities, resulting in a short-term bias in

organizational adaptation (March, 1991). In addition, problemistic search and

slack search have different consequences for exploration and exploitation because

exploration has greater risk and less routinization. Exploration and exploitation are

also affected by organizational momentum (Kelly and Amburgey, 1991) and direct

competition from exploitation to exploration (March, 1991). These propositions

are tested using data on innovations in shipbuilding between 1972 and 2000.

Exploration and exploitation are fundamental activities of organizations and other

adaptive systems (March, 1991). Organizational exploration is search for new

knowledge, use of unfamiliar technologies, and creation of products with unknown

demand. Because these activities do not reliably and quickly produce revenue,

exploration has uncertain and distant benefits. Exploitation is use and refinement of

existing knowledge, technologies, and products, and has more certain and proximate

benefits. Exploration and exploitation both draw resources, and thus resource

constraints require organizations to make tradeoffs between them (Levinthal and

March, 1993).

Organizations appear to have difficulty making these tradeoffs. Exploration and

exploitation rely on different organizational routines and capabilities (Lewin et al.,

1999; Galunic and Eisenhardt, 2001; Benner and Tushman, 2003), so it is easier to

specialize in one of them than to efficiently perform a mixture of both. In technology

development, exploitation of existing stocks of knowledge appears to reduce the

incentives for exploring new knowledge and possibly even the ability to do so

(Levinthal and March, 1981; Tushman and Anderson, 1986; Leonard-Barton, 1995;

Christensen and Bower, 1996). Despite the difficulties in combining exploration

and exploitation, theory and empirical evidence suggests that too little of either

exploration or exploitation reduces performance (Levinthal and March, 1993;

� The Author 2007. Published by Oxford University Press on behalf of Associazione ICC. All rights reserved.

Industrial and Corporate Change Advance Access published May 26, 2007

Katila and Ahuja, 2002; Fagiolo and Dosi, 2003; He and Wong, 2004). This indicates

a need to learn more about how organizations shift between exploration and

exploitation.

Some of the earliest theoretical work relevant to exploration and exploitation is

the behavioral theory of the firm (Cyert and March, 1963), which posits problemistic

search and slack search as drivers of organizational learning and change in general

and innovations specifically. Later work on organizational learning has added much

theory and evidence (see reviews by Levitt and March, 1988; Miner and Mezias, 1996;

Schulz, 2002; Argote and Greve, 2007), and has proposed additional ways in which

organizational learning may affect innovations. These include the propositions that

routinization causes repetition of organizational changes (Kelly and Amburgey,

1991) and that scarce resources and pursuit of proximate awards cause exploitation

to drive out exploration (March, 1991). These propositions are a good foundation,

but added theoretical and empirical work is needed to assess which mechanisms

affect organizational exploration and exploitation through making innovations.

First, these propositions are so general that they do not specify the type of

organizational change that is predicted, so it is easier to predict that low performance

will lead to some sort of change than to predict that it will lead to exploration.

For example, low performance leads to downsizing (Ahmadjian and Robinson,

2001), which seems contrary to exploration. Second, theory of routines will not

predict exploration well if exploration events are too unique to become routinized.

Currently only the proposition that exploitation competes with exploration gives a

direct prediction on how exploration and exploitation interact, but consideration of

whether the other theoretical mechanisms have unequal effects on exploration and

exploitation may yield additional predictions. One contribution of this article is to

develop theory on how performance, slack, and routines affect exploration and

exploitation, including the proposition that the greater risk and diversity of

exploration innovations cause them to be affected differently from exploitation

innovations.

Third, the evidence on extant propositions is scarce with respect to organizational

innovations, though there is evidence from other forms of organizational change.

There is little research examining performance effects on innovations (Bolton, 1993;

Greve, 2003), but on some other outcomes such as risk taking and strategic changes

(Bromiley, 1991; Lant et al., 1992; Nickel and Rodriguez, 2002; Miller and Chen,

2004). Most research on organizational slack has looked for effects on R&D or

patenting (Kamien and Schwartz, 1982; Antonelli, 1989), which are intermediate

steps in the process of developing innovations. The well-known difficulties of

getting innovations based on R&D projects selected for product launch (Burgelman

and Sayles, 1986; Dougherty and Heller, 1994) caution against drawing strong

conclusions on innovativeness from R&D evidence. Routinization of organizational

change has been investigated through mergers and acquisitions, alliances, and other

strategic changes (Amburgey and Miner, 1992; Amburgey et al., 1993; Lavie and

2 of 31 H. R. Greve

Rosenkopf, 2006), but not through innovations. A second contribution of this article

is to empirically test these propositions using innovation data.

1. Theory and Hypotheses

1.1 Innovations as exploration and exploitation

Many writers emphasize the uncertainty involved in developing innovative products

and launching them in the market (Burgelman and Sayles, 1986; Van de Ven et al.,

1999), suggesting that innovations are a form of organizational exploration. Much

product development hews closely to the current knowledge of the firm (Griffin,

1997), however, it is important to distinguish explorative innovations from those

that exploit current knowledge. One point of departure is the distinction between

incremental innovations, which advance existing technology, and radical innova-

tions, which develop new technology (Dewar and Dutton, 1986). Radical innovations

fit the definition of exploration because development of new technology is a form of

knowledge development (Rosenkopf and Nerkar, 2001; Benner and Tushman, 2003).

Despite its merits, the definition is not immediately applicable to firm-level research

on exploration and exploitation. First, it defines innovations as radical with respect

to the industry as a whole, which ignores differences in the knowledge held by a focal

firm with respect to a given innovation. Indeed, the finding that a firm’s depth of

knowledge predicts early adoption of both incremental and radical innovations

(Dewar and Dutton, 1986) can be explained by prior knowledge of an innovative

technology allowing early adoption. Second, it does not sufficiently take into account

the link from the knowledge held by the firm prior to developing the innovation and

knowledge that must be generated in order to develop it. An innovation is more

explorative when the firm had less advance knowledge to assess the probability of

successfully developing the innovation and launching it in the market. To make this

judgment, researchers have emphasized novelty in the technological and market

domain as a useful criterion (Rosenkopf and Nerkar, 2001; Benner and Tushman,

2002; 2003). This is because innovations that are technologically very different from

existing products have lengthy and unpredictable development durations

(McDonough, III, 1993), and innovations that address unfamiliar markets have

unpredictable market success (Christensen and Bower, 1996). Thus, we may define

the extent of exploration in an innovation launch as its technological and market

novelty for the focal firm.

Firms can choose not only the extent to which they seek to innovate, but also

whether to emphasize exploitation innovations or exploration innovations.

Exploitation innovations that apply known technology are frequent in firms that

design and customize products based on customer requests and in mass-producing

firms with product development processes geared towards the interests of

Exploration and exploitation in product innovation 3 of 31

customers (Burgelman and Sayles, 1986), but firms also develop exploration

innovations through learning new technologies (Rosenkopf and Nerkar, 2001). In the

shipbuilding industry, which is the context of this study, container ships without

hatch covers were an innovation that improved container handling in ports. They

were different from the earlier container ships with hatch covers, but did not require

new technology because the engineering problems in building them were familiar

from other types of ships without hatch covers, and were thus an exploitation

innovation. On the other hand, aluminum honeycomb hulls were designed as a

replacement of conventional steel hulls for smaller vessels, but design and

manufacture of aluminum honeycomb hull involves new skills for shipbuilding

firms, who are familiar with the manufacturing and seagoing properties of steel.

1.2 Balancing exploration and exploitation

Many have argued the benefits of balancing exploration and exploitation

(March, 1991; Levinthal and March, 1993; Benner and Tushman, 2003). A preference

for exploration results in excessive costs of failed experiments and insufficient

rewards from successful ones. A preference for exploitation may not be harmful in

the short run, or even in the long run if the environment is stable, but it reduces the

organization’s ability to discover opportunities and respond to environmental

changes. Thus, a common expectation is that a balance of exploration and

exploitation is preferable, though the costs of insufficient exploration may not be

apparent in the short run. Empirical work has supported this proposition (Katila and

Ahuja, 2002; He and Wong, 2004).

The arguments for why a balance might be beneficial stand in stark contrast to

arguments on the difficulty of maintaining a balance between exploration and

exploitation. Exploration in technological domains requires greater diversity of

knowledge than exploitation, and hence a different set of capabilities (Nonaka and

Takeuchi, 1995). Exploration calls for less attention to the current organizational

strategy, lower conformity to current organizational practices, and less emphasis

on leveraging current strengths (Burgelman, 1991; March, 1991; Dougherty and

Heller, 1994; Leonard-Barton, 1995). An emphasis on routine development and

refinement benefits exploitation but will suppress exploration (Benner and Tushman,

2002). Experienced top management teams favor exploitation over exploration

(Beckman, 2006).

These pressures towards specialization are sufficiently strong to raise the question

of how an organization might balance—or mix—exploration and exploitation.

If exploration and exploitation do not coexist easily, as these arguments suggest,

then some form of decoupling between them is needed. Two possible alter-

natives are organizational decoupling by having specialized subunits or temporal

decoupling through switching (Levinthal and March, 1993; Gupta et al., 2006).

Decoupling across subunits is possible at some cost, though the need for

4 of 31 H. R. Greve

separate management approaches in the same organization will introduce tensions

(Benner and Tushman, 2002). Decoupling over time requires the organization to be

capable of switching between two sets of behaviors, including one (exploration) that

is presumably used less often (Burgelman, 2002). Both approaches to mixing the two

are possible and will produce outcomes that appear similar, as it is difficult to tell

whether an observed mix of exploration and exploitation over time is a result of

temporal decoupling or of organizational decoupling with independent launches of

exploration and exploitation innovations.

There are scraps of evidence on organizations that mostly exploit but sometimes

explore as well. Organizations with insufficient capabilities to explore new

opportunities can obtain them from the outside (Cattani, 2005), with alliances a

likely approach when the necessary capabilities are dispersed across organizations

(Beckman et al., 2004; Lavie and Rosenkopf, 2006). Product development teams

with diverse capabilities are effective at both exploration and exploitation, suggesting

that the underlying capabilities are less important than the goal of exploring or

exploiting (Taylor and Greve, 2006). However, there is still modest evidence to

support either the balancing or the specialization hypothesis, so we are left with

the contention that a balance is possible but probably difficult to maintain.

The hypotheses derived here concern pressures towards exploration and exploitation,

and assume that balance or specialization is a result of the relative strength of the

countervailing forces.

Two features of exploration innovations are important for deriving the

hypotheses. First, exploration innovations are diverse. It is difficult to obtain new

technology in a routine and repeated manner, and even applying existing technology

to make qualitatively new products is a process that likely differs every time. As one

moves towards exploitation, the innovations become more homogeneous. For

example, the ease of loading and discharging ships without hatch covers or with

hatch covers that extend the entire width of the deck suggests the potential for

applying such designs to other ship types. As another example, a firm that has made

an engine of record-breaking fuel efficiency will find it natural to pursue the next

record-breaking engine through the same development process and to market it to

the same customers. This difference matters for theory of routine behavior. Second,

exploration innovations are riskier than exploitation innovations because they

require acquisition of new knowledge, which is a difference that matters for theory of

risk taking.

Keeping these two differences in mind, we can examine whether learning

rules have different consequences for exploration and exploitation innovations.

The learning rules that will be investigated are problemistic search (Cyert and March,

1963: 120–121), slack search (Cyert and March, 1963: 279), organizational

momentum (Kelly and Amburgey, 1991), and the exploitation bias (March, 1991).

These learning rules are fundamental and have proven successful in predicting

Exploration and exploitation in product innovation 5 of 31

organizational change, but their consequences for exploration and exploitation have

not been sufficiently investigated.

1.3 Problemistic search and innovations

Bounded rationality reasoning predicts that decision makers determine goal

variables and set aspiration levels for each goal variable so that they can use a

simple decision rule: search sequentially through the alternatives until finding one

that satisfies the aspiration levels (March and Simon, 1958). When managers

experience performance below the aspiration level, they initiate problemistic search

for remedial actions (Cyert and March, 1963: 120–123). As a result, performance

below the aspiration level predicts a wide range of actions that managers view as

consequential for performance, such as strategic reorientation (Lant et al., 1992),

market entry (Greve, 1998; McDonald and Westphal, 2003), and innovations

(Bolton, 1993; Greve, 2003).

The basic form of this argument applies equally to exploration and exploitation

innovations and can thus not predict changes in the balance of the two, but a simple

extension leads to such a prediction. It has been observed that many innovations that

organizations develop are not launched because of perceived lack of fit with the

current strategy (Dougherty, 1992; Dougherty and Heller, 1994). This observation

leads to two suggestions. First, lack of fit with the current strategy characterizes

exploration innovations, suggesting that organizations that do not explore in actual

product launches do explore in the development process. Second, the role of

managerial evaluations of strategy fit in halting launches of exploration innovations

suggest that organizations could turn sharply towards exploration if managers viewed

the existing strategy less favorably or accepted more risk. Actions that depart from

the current strategy will have less known consequences and thus be seen as risky, and

low performance causes increased propensity to take risky actions (Kahneman and

Tversky, 1979; Bromiley, 1991; March and Shapira, 1992), making exploration

innovations benefit disproportionately from low performance. Hence, performance

should have a greater effect on exploration innovations:

Hypothesis 1a: When performance relative to the aspiration level decreases,

the rates of launching exploration innovations and exploitation innovations

increase.

Hypothesis 1b: Performance relative to the aspiration level has a stronger

effect on exploration innovations than on exploitation innovations.

1.4 Slack search and innovations

Slack resources are argued to be an important determinant of innovations (Jelinek

and Schoonhoven, 1990; Schoonhoven et al., 1990). Organizations with excess

resources engage in slack search, which is search for “innovations that would not be

6 of 31 H. R. Greve

approved in the face of scarcity but have strong subunit support” (Cyert and March,

1963: 279). Unlike problemistic search, slack search is unrelated to immediate

problems and guided mainly by the interests of the individuals and groups engaged

in search. Slack search is thus closely associated with organizational exploration

(March, 1991). Organizations with slack resources have greater opportunities for

experimentation and laxer performance monitoring, both of which are needed to

make exploration innovations (Lounamaa and March, 1987).

Slack in the form of administrative resources beyond what is necessary for the

short-term operation and maintenance of the organization is called absorbed slack

(Singh, 1986). Facilities for R&D, staff specialized for development purposes, and

time for development activities among other staff are examples of absorbed slack

useful for developing innovations. Absorbed slack less useful for developing

innovations includes large administrations, costly facilities, and high wage levels,

so absorbed slack only increases the supply of innovations if some of it is directed to

innovation development.

Slack in the form of financial reserves is called unabsorbed slack (Singh, 1986).

Unabsorbed slack is not directly helpful in the development of innovations, but it

affects the decision to continue R&D projects because great financial resources lead

to laxer performance monitoring of uncertain projects. Strict performance

monitoring can cause new activities to be aborted before the organization has

accumulated enough experience to know whether they will eventually improve its

performance (Lounamaa and March, 1987), which can cause premature termination

of R&D projects.

As argued previously, exploration innovations have high risk of termination as a

result of perceived lack of fit with the current strategy (Dougherty, 1992) and high

risk (Singh, 1986; Howell and Higgins, 1990), so an innovation development process

geared towards risk reduction and strategy congruence may weed them out even

before the launch stage. Organizational slack can save explorative development

processes in the form of unauthorized research (Jelinek and Schoonhoven, 1990;

Burgelman, 1991), leading to the suggestion that slack increases the development of

exploration innovations. However, this prediction cannot be made for exploration

innovations only, as low levels of slack could lead to termination of projects to

develop exploitation innovations as well. Thus, we predict that slack search increases

the rates of both exploration and exploitation innovations, but that the effect on

exploration innovations is greater:

Hypothesis 2a: When absorbed and unabsorbed slack resources increase, the

rates of launching exploration innovations and exploitation innovations

increase.

Hypothesis 2b: Absorbed and unabsorbed slack resources have

a stronger effect on exploration innovations than on exploitation

innovations.

Exploration and exploitation in product innovation 7 of 31

1.5 Routinization of innovative direction

Organizations build up routines when activities are performed repeatedly (Nelson

and Winter, 1982), and this routinization offers productivity advantages such as

those seen in learning-curve effects on the unit cost of production (Argote, 1999),

causing actions performed recently to be more efficiently executed. Also, major

managerial actions are subject to interpretation and self-attribution that give actions

recently performed greater prominence in decision making (March et al., 1991).

These mechanisms make repetition more likely than novelty and give greater

short-term rewards to repeated actions than to new ones (Levinthal and March,

1981). The result is organizational momentum, or the tendency to repeat previous

actions (Miller and Friesen, 1982; Kelly and Amburgey, 1991).

Organizational momentum is seen as one of the processes favoring exploitation

over exploration because momentum is caused by positive feedback from repetition

of known actions (Kelly and Amburgey, 1991). The homogeneity of exploitation

innovations may result in greater ease of routinizing them, but some organizations

also develop routines for making exploration innovations (Jelinek and Schoonhoven,

1990). Thus, an organization is more likely to make an exploitation innovation the

more recently it has made an exploitation innovation, and is more likely to make an

exploration innovation the more recently it has made an exploration innovation.

Conversely, organizational routines atrophy with disuse (Argote, 1999), making it

less likely that organizations will innovate when they have not done so for a while.

A parsimonious way of formalizing this argument is to let the likelihood of an

exploitation innovation and an exploration innovation, respectively, depend on the

duration since the last time the same kind of innovation has been made (Amburgey

et al., 1993). The weakening of routines through disuse implies that the effect of

duration is negative; making an innovation less likely the more time has passed since

the previous one. These predictions can be made:

Hypothesis 3a: The rate of launching exploration innovations decreases with

the duration since the last exploration innovation.

Hypothesis 3b: The rate of launching exploitation innovations decreases

with the duration since the last exploitation innovation.

1.6 Exploitation bias and innovative direction

Exploration and exploitation compete for a limited pool of resources, leaving

managers with a choice of which to support more strongly (March, 1991). Moreover,

exploitation and exploration differ in attention patterns because an organization that

explores seeks to learn from experiments performed by subunits, while an

organization that exploits seeks to make each subunit conduct activities that draw

on the existing organizational knowledge (March, 1991). An important mechanism

to support exploitation is cancellation of R&D projects that diverge from the

8 of 31 H. R. Greve

organizational strategy and transfer of resources to strategy-congruent projects

(Burgelman, 1991; Dougherty and Heller, 1994).

How the tradeoff between exploration and exploitation is made differs among

organizations, and may systematically depend on their prior record of exploration

and exploitation, as hypothesized earlier. In addition to this mechanism, theoretical

work has proposed that the tradeoff tends to be made in favor of exploiting more

intensely because exploitation has greater and more certain benefits in the short run,

which is a feedback pattern that easily draws attention and is rewarded by managerial

incentives (Levinthal and March, 1993). Thus, the competition for resources is

asymmetric, with exploitation innovations harming exploration innovations but not

the other way around. The result is that performing exploitation not only improves

the organizational routines for exploitation and increases the likelihood that

exploitation will be performed again; it also reduces the resources available for

exploration. Thus the prediction is:

Hypothesis 4: The rate of launching exploration innovations decreases with

the proportion of recent exploitation innovations.

1.7 Integration and theoretical problems

Through its prediction that resources allow generation of innovations, the theory of

slack search seems to posit the opposite effect on innovations as that of problemistic

search. On closer examination, the theories differ because problemistic search is a

theory of perceived performance shortfalls, while slack is one of excess resources.

An organization with reserves built up over a period of time but with recent

performance below its managers’ aspiration levels would simultaneously have

resources necessary for innovating and a perceived need to innovate, and would be

predicted to have a high likelihood of launching innovations. An organization with

low reserves may perform above the expectations of its managers, and would be

predicted to have a very low likelihood of launching innovations. One may still ask

whether these mechanisms are different only in principle, but in reality are correlated

because slack resources diminish as a result of low performance and increase as a

result of high performance.

There are two reasons for slack and performance to have low correlation. The first

is that slack is a stock of resources that is incrementally changed by the resource flow

from the performance. If firm differences in stocks are large compared with the

temporal variation in resource flows, then performance and slack will not be highly

correlated. The second is that the evaluation of performance by comparing with an

aspiration level decouples subjective evaluations of success from resource flows

(Levinthal and March, 1981). While these relations both reduce the correlation of

slack and performance, it is an empirical question whether they are distinct in a given

study context.

Exploration and exploitation in product innovation 9 of 31

Problemistic search also has a potential conflict with the theory of threat rigidity,

which argues that low performance may be viewed as a threat, leading to risk aversion

and inability to act (Staw et al., 1981). Theoretical work has sought to resolve this

conflict by determining the conditions under which low performance is seen as a

threat to the existence of the organization or as a repairable problem (March and

Shapira, 1992; Mone et al., 1998). This resolution fits well with evidence that

problemistic search is usually found in organizations with low profitability, but threat

rigidity has been observed in organizations close to bankruptcy or with extremely

low profitability (Wiseman and Bromiley, 1996; Ketchen and Palmer, 1999; Ferrier

et al., 2002; Miller and Chen, 2004). Hence, the prediction that innovations

become more likely as the performance declines should hold for most performance

levels, but may be counteracted by threat rigidity for very low performance.

Finally, an unresolved theoretical question is whether the increased rate of change

in low-performing organizations is a result of problemistic search (Cyert and March,

1963), risk taking (Singh, 1986; Bromiley, 1991), or a combination of the two

(Greve, 1998). The causes are difficult to distinguish because problemistic search and

risk taking make the same prediction as long as changing is more risky than not

changing. Investigation of exploration and exploitation innovations gives leverage for

discovering the role of risk, as the prediction that the effect on exploration

innovations is greater relies solely on theory of risk taking. Hence, if risk taking is

an important explanation for the reactions to low performance, the findings from

analyzing exploration innovation and exploitation innovations should diverge. If risk

taking is not important, and thus problemistic search alone accounts for the effect,

then the findings for exploration innovations and exploitation innovations should be

similar. Clearly, these hypotheses can supply evidence relevant to important

theoretical disputes.

2. Methods

2.1 Japanese shipbuilding

This study uses data on the Japanese shipbuilding industry from 1971 to 2000.

The Japanese shipbuilding industry built its competitiveness in the 1950s and early

60s through a combination of government financing of work-in-process inventories,

firms’ attention to production efficiency, and anticipation of demand changes such

as the tanker boom (Blumental, 1976; Chida and Davies, 1990). Modern production

facilities gave Japanese shipbuilders unrivaled labor productivity, and large

docks allowed them to dominate the supertanker market. Rapid delivery times

were also a source of competitive advantage, and were achieved through

effective production management and an avoidance of large order reserves

(Chida and Davies, 1990). Finally, innovations were used to increase the competitive

10 of 31 H. R. Greve

strength and create new market niches of large and high technology vessels

(Chida and Davies, 1990).

Ships are investment goods used in the production of transportation services.

They have highly flexible life-times: 20 years is often used as a rough estimate of the

lifetime of ships, but they can be scrapped earlier if the shipping firm does not see

prospects for transportation rates high enough to justify continued operations,

and can continue operating for at least a decade more if high rates justify the higher

insurance, maintenance, and repair costs of old ships. The result of these product

characteristics is an extremely variable demand, as shown in Figure 1. Because the

Japanese shipbuilding industry was the world’s largest when the oil shock of 1972

caused the shipbuilding market to collapse, it (along with Western Europe) took

most of the production cut. Western European production never recovered to past

levels as a result of large capacity cuts.

Japanese shipbuilders took a different approach to this crisis. Shipbuilding became

the target of Ministry of Technology and Industry policies for structural adjustment

of declining industries in the late 1970s, which included plans to cut capacity and to

merge some shipbuilders. The firms were able to negotiate joint capacity cuts, but the

ministry’s merger plan failed because the target firms resisted (Strath, 1994). After

the oil crisis passed, the shipbuilders were left to fend for themselves, as policy

attention turned to industries thought to have greater growth potential (Patrick,

1986). The shipbuilders kept their production capacity roughly constant after the

initial round of cuts and used ship repairs and non-shipbuilding business to

maintain activity. The average annual growth of production facilities in Japan fell

from 11% during the most intense expansion from 1967 to 1973 to 4% in the decade

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

1964

1966

1968

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

20

0020

0220

04

Year

1000

GT

JapanKoreaChinaAmericasW. EuropeE. EuropeOthers

Figure 1 World ship deliveries by nation or region of production.

Exploration and exploitation in product innovation 11 of 31

following the oil crisis and 1% in the next decade.1 However, employment levels were

cut throughout this period. As a result of these decisions, Japanese shipbuilding firms

captured much of the growth in the brief recovery in the first half of the 1980s and

the sustained recovery thereafter, but also took most of the production cuts in the

intervening recession. Possibly as a result of the order cancellations after the oil

shock, Japanese firms built multi-year order reserves when the shipbuilding market

recovered, just as their foreign competitors did. Table 1 shows how selected firm

characteristics changed in 6-year intervals starting in 1973.

Throughout this period, the Japanese shipbuilders could do little to counter

competition from South Korea. Korean labor rates were one-third of the Japanese

until 1990, and South Korean shipbuilders had modern facilities, high productivity,

Table 1 Selected firm characteristics over time

Firms Mean SD Minimum Maximum

1973

Employees 13 19,511 22,180 851 75,330

Sales 13 115,927 135,173 7407 385,155

Order Res. 13 419,289 463,689 21,326 1,606,831

ROA 13 0.046 0.055 0.007 0.205

1979

Employees 12 17,010 19,218 947 63,771

Sales 12 374,471 446,380 16,842 1,274,862

Order Res. 12 461,546 625,944 14,182 2,050,309

ROA 12 �0.061 0.148 �0.443 0.118

1985

Employees 11 16,012 16,357 914 51,763

Sales 11 566,591 652,820 19,393 1,999,745

Order Res. 11 713,695 925,378 16,430 3,120,241

ROA 11 0.036 0.045 �0.057 0.135

1991

Employees 11 10,973 13,329 615 44,272

Sales 11 570,163 719,633 23,492 2,327,067

Order Res. 11 799,005 1,174,480 30,644 4,051,316

ROA 11 0.052 0.037 �0.038 0.094

1997

Employees 11 9698 12,029 547 40,829

Sales 11 653,215 805,666 25,363 2,733,765

Order Res. 11 986,165 1,544,477 29,770 5,342,736

ROA 11 0.045 0.031 �0.003 0.093

1Author’s calculation from establishment-level data on production facility value.

12 of 31 H. R. Greve

and government support. Recently, Chinese shipbuilders, which have yet-lower labor

rates, have expanded greatly, but are technologically behind the Japanese and South

Korean ones. The addition of shipbuilding capacity in China and elsewhere has kept

prices of most ship categories constant or even slightly declining after 19952 despite

the great increase in demand, but the shipbuilders have still been generally profitable

in this period as a result of the high capacity utilization. The idea of consolidation

through mergers resurfaced in year 2000 as two different groups of Japanese

shipbuilders announced that they were in talks to form alliances with a final goal of

merging (Hitachi-NKK and IHI-Sumitomo-Kawasaki-Mitsui; with industry leader

Mitsubishi notably absent from both groups). The first merger to be realized was IHI

and Sumitomo in October 2002.

2.2 Data

The data contain 13 firms, of which 11 were active throughout the study period,

while two disappeared through mergers in 1978 and 1985, respectively. These are all

the Japanese shipbuilding firms listed on the Tokyo and Osaka stock exchanges, and

all are present in the international market for oceangoing ships. The firms range in

size from the giant Mitsubishi Heavy Industries with more than 70,000 employees to

Hashihama Shipbuilding with about 800 employees.

Earlier work found that the Japanese shipbuilders launched innovations in

response to performance below the aspiration level, but did not find effects of slack

resources (Greve, 2003). That work did not distinguish between exploration

innovations and exploitation innovations, however, and estimated Poisson models of

the number of innovations per year, which does not allow testing of how the rate of

making innovations depends on the duration since the last innovation. Thus, the

hypotheses presented here have not seen previous testing.

2.3 Dependent variables

Innovations were identified by reading the monthly journals New Technology

Japan and Techno Japan from 1971 through 2000. The use of technology journals as

data sources reflects a tradeoff between the characteristics of alternative data

sources on innovations. Innovations are listed in the journals, when they meet

the industry’s definition of a technological innovation and are thus subject to

external review. The innovations in these journals are launched as products, unlike

patents, which are sometimes made just to record and protect a technological

advance not yet viewed as ready for use in a product. Because risk taking and

change of routines is important in the theory predicting exploration, mere recording

of a technological advance is a less appropriate dependent variable than the launch

2E.g., as reflected in the ship prices shown in Table 1.6.2 in Shipping Statistics Yearbook by the

Institute of Shipping Economics and Logistics.

Exploration and exploitation in product innovation 13 of 31

of a product innovation. Finally, firm sources such as annual reports lack

external review of the innovativeness of new technologies, so they are vulnerable

to self-presentation concerns.

Classification of the innovations was based on the written description of each

innovation as follows: (i) Exploration innovations were innovations that the sources

described as involving development of new technology or application of existing

technology not earlier used by the focal firm. Thus, Mitsubishi developing a semi-

submerged catamaran hull for high-speed passenger traffic for the first time in the

world and Ishikawajima-Harima developing a navigational simulator for the first

time in Japan were exploration innovations. This coding identified 70 exploration

innovations. (ii) All remaining innovations, 203 in total, were exploitation

innovations that did not involve the firm learning or developing new technology.

The frequency of exploitation innovations in the data reflects the ability of

shipbuilders to reconfigure existing technologies to provide new products, as in the

container ship example mentioned earlier and new marine engine designs by various

shipbuilders.

Figure 2 shows innovation event plots in which each innovation is placed in

a time line, and exploration and exploitation innovations are separated. The plots

are shown for three shipbuilders in the data: Mitsubishi, Mitsui, and Hitachi.

These have more than the average number of innovations, but this helps

demonstrate two features of their innovation plots that are typical of the firms in

the data. First, the rate of innovation is sensitive to the economic conditions,

as there are many more innovations in the troubled early period than in the

high-demand later period. The analysis will show that this difference is accounted

for by the effect of firm profitability on innovativeness. Second, there is

serial correlation in the type of innovation launched. Exploitation innovations

come in batches, and to some extent this seems to occur for exploration innovations

as well.

The latter observation is important for the theoretical arguments made earlier,

because it supports the idea of temporal specialization in exploration or exploitation

rather than mixing of the two. To provide a formal test, Table 2 shows a simple

analysis of the likelihood of switching innovation type. Each innovation is an

observation, and the dependent variable is whether the innovation is of a different

type (exploration or exploitation) than the previous. Firms are less likely to switch

from an exploitation innovation to an exploration innovation than the other way

around, as one would expect if exploitation drives out exploration. However, the

relative scarcity of exploration innovations can also explain this finding. There is not

a significant effect of the duration since the last innovation, though the positive sign

suggests that switching is least likely shortly after making an innovation. This is

similar to the tendency of self-reinforcing exploration and exploitation observed in

alliance formation (Lavie and Rosenkopf, 2006). Later analysis will examine the firm-

level determinant of each type of innovation.

14 of 31 H. R. Greve

2.4 Independent variables

Nikkei corporation’s NEEDS database of corporate accounts was used to download

accounting measures of performance and slack. Performance was measured through

return on assets (ROA), defined as the operating income divided by the total assets.

Return on assets is theoretically appealing because it is an estimate of economic rents

Exp

lora

tion

Exp

loita

tion

01jan1972 01jan1980 01jan1988

Mitsubishi

Exp

lora

tion

Exp

loita

tion

Inno

vatio

n

Exp

lora

tion

Exp

loita

tion

Inno

vatio

nIn

nova

tion

Hitachi

Mitsui

Date

Date

01jan1996

01jan1972 01jan1980 01jan1988 01jan1996

Date

01jan1972 01jan1980 01jan1988 01jan1996

Figure 2 Innovation event plots.

Exploration and exploitation in product innovation 15 of 31

and is salient in managerial performance evaluation. It is the preferred measure in

work on performance effects on organizational change and risk taking (Bromiley,

1991; Lant et al., 1992; Wiseman and Bromiley, 1996; Miller and Chen, 2004). Return

on equity is less preferred because it blends profitability from operations with

the debt–equity mix. Other measures that have seen less use are return on sales

(Audia et al., 2000) and various industry-specific performance measures (Miller and

Chen, 1994; Greve, 1998).

Following Cyert and March (1963: 123), the performance (P) was compared with

an aspiration level (A) computed as a mixture of a social and a historical aspiration

level. The social aspiration (SA) is the average of the other firms’ performance,

calculated as the average ROA of all shipbuilders in the data except the focal firm.

The historical aspiration (HA) level is a weighted average of the previous-year

historical aspiration level and the previous-year performance of the same firm.

Assuming a1 and a2 be weights, the formulae are:

Ati ¼ a1SAti þ ð1� a1ÞHAti

SAti ¼X

j 6¼iPt;j

� �=ðN� 1Þ

HAti ¼ a2HAt�1;i þ ð1� a2ÞPt�1;i

Here, t is a time subscript and i/j are firm subscripts. The weights were estimated

by searching over all parameter values in increments of 0.1 and taking the

combination giving the highest model log likelihood. This procedure yielded a1¼ 0.8

and a2¼ 0.2, which means that the industry average of performance had a weight of

0.8, the firm performance in the previous year had a weight of 0.16, and the historical

aspiration level in the previous year had a weight of 0.04. The performance was split

Table 2 Logit model of switching innovation type

Variable Estimate Odds Ratio

Previous innovation is exploitation �2.133** 0.118

(0.319)

Time since previous innovation 0.239 1.270

(0.157)

Constant 0.648

(0.368)

Log likelihood �146.598

Likelihood ratio test 53.1**

Standard errors are shown in parentheses. 258 observations.

**P50.01; two-sided tests.

16 of 31 H. R. Greve

into the two variables performance above the aspiration level and performance below

the aspiration level to allow a greater effect of performance above the aspiration level,

as predicted by performance feedback theory (Greve, 1998).

Absorbed slack is measured as the ratio of selling, general, and administrative

expenses to Sales, and unabsorbed slack is measured as the ratio of quick assets

(cash and marketable securities) to liabilities (Bourgeois, 1981; Singh, 1986).

These measures use accounting data to capture real and financial assets in

excess of what is needed for regular operations, and thus available for search

(Bourgeois, 1981). The theoretical fit and easy replicability across different

samples have made them leading measures in research on slack (Bromiley, 1991;

Nohria and Gulati, 1996; Wiseman and Catanach, 1997).3 These slack measures

are easiest to interpret when the organizations are involved in similar forms

of business, as they measure excess resources without adjusting for the normal

resource requirement for a given business. Because innovations are developed

over a period of time, the slack measures are expressed as averages over the

four preceding years.

Organizational momentum was captured by specifying a hazard rate model

in which the rate depends on the duration since the previous event of the same

type (Amburgey et al., 1993). The prediction is negative duration dependence,

which means that the event is more likely shortly after the previous occurrence.

In addition, for each type of innovation, an indicator variable was set to one until the

first time a firm was observed making an innovation of that type in the study period,

and was set to zero thereafter. This variable is an adjustment for the left-censored

observations, which have unknown values of the duration since the last innovation.

In preliminary analyses, an alternative approach of deleting all left-censored

observations was tried, and found to yield the same results. The models entering

an indicator for left-censored observations are displayed because they use more of

the available data.

The proportion of all firm innovations in the four most recent years that were

exploitation innovations was included to model whether exploitation drives out

exploration. In other not to lose four full years of observations in the start, the

observations from 1972 through 1974 used all available years of innovations instead

of the full four years to compute this proportion. The data for 1971 were only used to

initialize the innovation variables. To ensure that the findings were not vulnerable

to changes in the specification, other specifications were also tried, but these (6- and

8-year proportions, a log ratio, and the average of time-discounted innovation

counts) did not provide better fit or different results.

3Alternative measures of slack exist, along with some overlapping terminology. For example, current

assets divided by current liabilities is called available slack, and sales and general administrative

expenses divided by sales is also known as recoverable slack (Bromiley, 1991).

Exploration and exploitation in product innovation 17 of 31

Firms engaged in technological races may launch innovations as a response to

innovations made by other firms. To capture this effect, the number of innovations

made by other Japanese shipbuilding firms in the previous year is entered. To capture

the effect of firm size on innovation capability, the logarithm of the number of

workers was entered. Asset value was also considered as a size variable, but correlated

highly with the number of workers. To control for the effect of economic trends in

shipbuilding on R&D and innovation, a number of industry measures were entered

in exploratory runs, such as the finished tonnage completed by the Japanese

shipbuilders, the growth in shipping income, and various freight rates. Of these, the

Gulf–Europe freight rate proved to be the best predictor and was retained in the

analysis. Freight rates predict shipbuilding because they are high when there is

scarcity of shipping capacity. Finally, large customers may induce the shipbuilder to

make innovations as a result of their greater product development capabilities and

more specialized needs (Dyer and Singh, 1998; Darr and Talmud, 2003; Ingram and

Lifschitz, 2006). To control for this effect, the average customer size, specified as the

logarithm of the average number of ships owned by customers to whom it delivered

ships, was entered as a control variable. The independent variables were lagged

by 1 year.

Table 3 Descriptive statistics and correlation coefficients

Variable Mean SD 1 2 3 4 5 6 7 8 9

Innovations in

industry

13.298 9.576 1.00

Log employees 8.770 1.458 0.17 1.00

Freight rates 10.565 3.743 �0.32 �0.10 1.00

Average customer

size

3.382 0.944 �0.03 0.14 �0.04 1.00

Performance,

over aspiration

0.013 0.028 �0.06 �0.06 0.09 �0.12 1.00

Performance,

under aspiration

�0.016 0.035 0.02 0.16 �0.10 0.14 0.21 1.00

Absorbed slack 0.072 0.028 �0.41 0.48 0.24 0.12 �0.04 0.08 1.00

Unabsorbed slack 0.600 0.096 0.52 �0.23 �0.22 �0.05 �0.21 �0.05 �0.46 1.00

Exploitation

indicator

0.206 0.405 0.04 �0.62 �0.09 �0.03 0.09 �0.04 �0.39 0.26 1.00

Exploration

indicator

0.423 0.494 0.09 �0.83 �0.08 �0.12 0.10 �0.13 �0.76 0.36 0.58 1.00

Exploitation

proportion

0.846 0.239 �0.11 �0.47 �0.01 0.01 �0.00 �0.09 �0.41 0.09 0.30 0.55

4061 observations.

18 of 31 H. R. Greve

Table 3 displays the mean, SD, and correlation matrix of the independent

variables. Some variables have medium to high correlation, but preliminary analyses

with subsets of these variables indicated that the estimates were stable. In addition to

the correlations reported in the table, two correlations coefficients of interest are

those of the return on assets (not adjusted for aspiration levels) with absorbed slack

and unabsorbed slack. These are 0.05 and �0.23, respectively, showing that the

returns in a given year are practically unrelated to the organizational slack. Thus,

performance and slack are both conceptually and empirically distinct.

2.5 Statistical model

The analysis was performed as a continuous-time event history model with two

events—exploration innovation and exploitation innovation. These were modeled as

a competing risk model with a Gompertz specification of the rate and duration since

the last innovation of the same type as the time variable. The Gompertz model

specifies the rate (r) as a function of time (t) and covariates (X) with associated

coefficients (�) as follows:

rt ¼ e�X e�t

The Gompertz model can be fit to a rate that declines over time (which would

support H3a and H3b), one that is constant, or one that increases over time (which

would contradict H3a and H3b) depending on whether the duration parameter � is

negative, zero, or positive. Seemingly unrelated regression was used to obtain

significance tests of coefficient differences across the outcomes, as required by

hypotheses 1b and 2b.4 The data file split spells monthly, in order to update the

covariates, and the event times were precise to the nearest month. The data contained

three pairs of tied event times (two innovations by the same firm in the same

month), which were separated by letting them occur at a half-month interval. The

estimation was done with Stata 9.0.

3. Results

Table 4 presents the results of the analysis. It contains three sets of estimates. Model 1

omits the duration dependence and the proportion of exploitation innovations, and

is thus similar to the model specification in Greve (2003) except that it is a hazard

rate model instead of a count model, and it separates exploration innovations and

exploitation innovations instead of combining them.5 For exploitation innovations,

performance relative to the aspiration level has a negative and significant effect as

4The simpler test of mean equality of two normal distributions is incorrect because the coefficient

estimates are correlated with each other.

Exploration and exploitation in product innovation 19 of 31

Table 4 Event history models of innovations

(1) (2) (3)

Model Outcome Exploit Explore Exploit Explore Exploit Explore

Log employees 1.047** 1.120** 0.862** 0.881** 0.836** 0.879

(0.120) (0.214) (0.139) (0.279) (0.170) (0.280)

Log age �0.681 �2.310* �0.891 �2.210** �0.900 �2.206**

(0.777) (0.922) (0.848) (0.763) (0.841) (0.730)

Freight rates 0.015 0.104** 0.012 0.101** 0.012 0.101**

(0.024) (0.025) (0.022) (0.023) (0.023) (0.024)

Innovations in industry 0.007 0.023 0.003 0.016 0.001 0.016

(0.012) (0.015) (0.012) (0.012) (0.012) (0.013)

Average customer size 0.091 �0.187y 0.076 �0.158 0.088 �0.158

(0.094) (0.096) (0.098) (0.109) (0.112) (0.106)

Performance – aspiration level �10.047* �34.041** �9.394* �31.985** �9.733* �31.909**

(when over aspiration level) (4.384) (10.475) (4.756) (8.635) (4.720) (8.389)

Performance – aspiration level �2.399 26.634** �3.358 28.159** �3.459 28.167**

(when under aspiration level) (3.087) (7.809) (2.888) (6.369) (2.804) (6.559)

Absorbed slack 3.462 10.726y 2.461 8.303 1.616 8.301

(4.366) (6.525) (3.205) (6.561) (3.368) (6.560)

5Th

ism

od

elis

estimated

usin

gth

eexp

on

ential

distrib

utio

n,

wh

ichis

the

same

asa

Go

mp

ertzm

od

el

with

the

du

ration

dep

end

ence

con

strained

tob

ezero

.

20

of

31

H.

R.

Greve

Unabsorbed slack 4.329** 1.894 3.813** 0.876 3.849** 0.871

(1.254) (1.936) (1.173) (1.601) (1.198) (1.621)

Exploitation indicator 0.748 0.781

(1.118) (1.090)

Exploration indicator 0.816 0.820

(0.889) (0.910)

Duration since last �0.0104** �0.0109**

exploitation innovation (0.0037) (0.0039)

Duration since last �0.0092 �0.0093

exploration innovation (0.0059) (0.0065)

Exploitation proportion �0.408 0.024

(0.539) (0.449)

Constant �13.457** �7.009 �9.950* �3.875 �9.299* �3.877

(4.209) (4.867) (4.684) (4.997) (4.598) (4.987)

Events 195 67 195 67 195 67

Log pseudolikelihood �253.233 �101.981 �249.028 �99.727 �248.451 �99.727

Likelihood ratio Chi-square 252.228** 116.4469** 260.638** 120.9549** 261.792** 120.9549**

Degrees of freedom 9 9 11 11 12 12

Robust standard errors adjusted for clustering by firm are in parentheses. 4061 observations.

yP50.10; *P50.105; **P50.01; two-sided tests.

Exp

loratio

nan

dexp

loitatio

nin

pro

du

ctin

no

vation

21

of

31

predicted, but only above the aspiration level. For exploration innovations,

performance relative to the aspiration has a negative and significant effect above

the aspiration level, and a coefficient twice the size of that for exploitation

innovations, and a positive and significant effect below the aspiration level. The

results for exploitation innovations are exactly as those in Greve (2003), but the

positive effect of performance below the aspiration level on exploration innovations

is a new finding. It suggests that the greater risk of exploration innovations makes

these less likely for firms with very low performance, giving the greatest likelihood of

exploration innovations for firms with performance near the aspiration level. For

exploitation innovations, unabsorbed slack has a positive and significant effect. This

finding is different from Greve (2003), who found no effect of slack. The difference in

results suggests that financial resources are more important for exploitation

innovations.

Model 2 introduces the duration dependence variables, and is used to test

hypotheses 1 through 3. Hypothesis 1a is tested by the coefficient estimates for

performance minus aspiration levels, which show the same results as in model 1. The

effects on exploitation are both negative as predicted, and the coefficient above the

aspiration level is significant, in support of hypothesis 1a. The effect on exploration is

negative and significant above the aspiration level, in support of hypothesis 1a. Below

the aspiration level, the effect on exploration is positive and significant, which

contradicts hypothesis 1a and shows that this more risky outcome is less likely in

firms with very low performance. Hypothesis 1b is tested by seemingly unrelated

regression Chi-square tests of equality of the coefficients for performance across the

exploration and exploitation. Such tests were performed, but were not significant for

performance above the aspiration level despite the difference in effect magnitude.

The coefficients of performance below the aspiration level were significantly different

at the 5% level, consistent with hypothesis 1b.

Hypothesis 2a is tested by the coefficients for slack. The findings show lack of

support for an effect of absorbed slack on exploitation innovations, as before. The

coefficient for unabsorbed slack is positive and significant for exploitation,

supporting hypothesis 2a and indicating that financial resources help organizations

make exploitation innovations. For exploration innovations, neither slack variable

has significant effect. The Chi-square test for difference of coefficients used for

assessing hypothesis 2b found no significant difference of the effect of slack on the

two outcomes, so despite the different patterns of significance we cannot conclude

that slack affects exploration and exploitation differently.

Hypothesis 3a is tested by the duration coefficient of exploitation, and is

supported by the negative and significant negative coefficient estimates. Hypothesis

3b is tested by the duration coefficient of exploration, and fails to receive support

because the coefficient is insignificant. The exploitation coefficient shows the

predicted momentum from previous exploitation innovations, resulting in higher

rates of making exploitation innovations soon after a previous one than after some

22 of 31 H. R. Greve

time has passed. The duration coefficient is –0.0104, which implies that a firm that

lets 12 months lapse since the last exploitation innovation will have a rate of

innovating that drops to exp(12��0.0104)¼ 88.3% of the original rate. This is an

appreciable decline, but it is smaller than losses from production stoppages, which

have been estimated to as much as 90% in a year (Argote et al., 1990).

In model 3, the proportion of exploitation innovations has been added to capture

the effect of competition between exploration and exploitation. Hypothesis 4

predicts a negative effect for exploration, and also implies that no significant effect be

seen for exploitation. However, both coefficient estimates are insignificant, and fail

to support the hypothesis. Compared with model 2, model 3 has unchanged findings

on the variables testing hypotheses 1 through 3. This is important with respect to the

findings on duration, which could potentially be affected by entering the new

variable of exploitation proportion.

Various models were estimated to test the sensitivity of the estimates to changes in

the model specification. First, replacing absorbed slack with research and develop-

ment intensity gave no effect of R&D intensity on innovations and unchanged results

for the other variables. Second, estimating performance and slack variables separately

(but with all other variables entered) led to unchanged results. Performance and

slack appear to explain different parts of the variance in the rate of innovation.

Third, estimating models with frailty effects of firms (similar to random effects)

showed that the frailty effects were not significant, and showed only minor changes

in parameter estimates.

4. Discussion

The vulnerability of exploration is fundamental issue in organizational adaptation.

“Compared to returns from exploitation, returns from exploration are systematically

less certain, more remote in time, and organizationally more distant from the locus

of action and adaptation” (March, 1991: 73). Organizational routines such as local

search naturally lead to exploitation (Stuart and Podolny, 1996), and managerial

practices such as the search for production efficiency (Benner and Tushman, 2002)

and strategic consistency (Dougherty and Heller, 1994) have the same effect. Current

theory of exploration and exploitation raises the question of how organizations

manage to explore at all.

The theory developed here uses the lower risk and greater routinization of

exploitation to predict that exploration innovations are more sensitive to

performance and slack and less affected by momentum. Although the findings

suggested some differences between exploitation and exploration innovations,

the similarities are more salient. Exploration innovations were scarcer than

exploitation innovations in the data, but were generated by similar processes.

This is different from current theory, which specifies that low performance

Exploration and exploitation in product innovation 23 of 31

initially triggers local search and thus exploitation, but persistent problems cause

expansion of search and potentially lead to exploration (Cyert and March, 1963;

Grinyer and McKiernan, 1990; Katila and Ahuja, 2002). The findings of this study

suggest that reductions in performance significantly increased the rate of making

exploration innovations as well as that of exploitation innovations, and at the same

time. Managers solving problems do turn to exploitation as a solution, but also try

exploration. Problemistic search thus offers an explanation for why organizations

that usually exploit will sometimes try exploration.

The analyses showed that unabsorbed (financial) slack affected exploitation

innovations. This finding is perhaps surprising, and suggests that financial reserves

affects managerial decision making so strongly that they increase innovation launch

rates. Money in the bank obviously does not create innovations, but it does influence

managerial perceptions of risk. This finding is consistent with work showing that risk

taking is increased by unabsorbed slack, but not by absorbed slack (Singh, 1986), and

suggests a role of risk preferences in the launching of innovations.

The findings suggest that some organizations build up routines for exploration

that lead them to be overall more innovative than other organizations (Miller and

Friesen, 1982). The analyses showed decay in the rate of exploitation innovations,

when the duration since the last one increased, as routinization of innovations would

predict. The rates of decay for exploration innovations were approximately the same

as those for exploitation innovations, but not significant, so it is difficult to tell

whether these innovation types are routinized differently. Routinization theory

predicts repetition of innovation patterns, which explains both cross-sectional

differences of firm innovation rates and temporal clustering of individual firms’

exploration and exploitation innovations. The analysis did not show that

exploitation reduces exploration rates, and hence failed to support the direct

tradeoff between exploration and exploitation that theory has suggested (March,

1991). Overall the findings show that organizational learning theory explains how

organizations adjust the rates of exploration and exploitation innovations, but it does

better in predicting the overall level of each one than in predicting shifts between

them.

The study has some limitations that suggest a need for further work. First, the data

set is not sufficiently large that we can be sure that the failure to find differential

effects on exploration and exploitation is because the effects are not there, or because

the effects are too weak to be distinguished in the data. This problem is intrinsic to

the study of innovations, which are often rare events. A larger study might find

differential effects between these determinants of exploration and exploitation

innovations, and would give better indication of whether direct competition from

exploitation to exploration occurs.

Second, the single-industry research design leaves doubts about the general-

izability of the findings. There are industries with a higher baseline pace of

innovation because they use technologies that are in rapid development, as in

24 of 31 H. R. Greve

computing and telecommunications. Without further investigation we cannot be

sure that the organizational search processes seen in such industries are the same as

those observed here, so investigation of these hypotheses in industries with fast-

paced technological change should be a high priority.

Finally, exploration and exploitation are concepts that extend to the market

behavior of firms as well as to their technology behaviors. It is not clear that studies

can generalize across these two outcomes, although from a theoretical viewpoint,

symmetric findings are expected. When investigating exploration and exploitation in

markets, operationalizing the degree of exploration in a given action (such as a

market entry) is probably at least as difficult as when investigating exploration and

exploitation in technology. However, it would be highly valuable to investigate the

determinants of exploration and exploitation in markets.

There are clear opportunities for further research to extend our knowledge of the

drivers of exploration and exploitation. It would be useful to expand the search for

variables that shift the balance between exploration and exploitation, and to perform

similar tests of their joint effects on exploration innovation and exploitation

innovations. For example, an important topic in current research is how the network

positions of firms affect their innovativeness (Baum et al., 2000; Tsai, 2001; Ruef,

2002). It seems likely that network position affects the type of innovation firms make

as well as the overall level of innovativeness. Firms with diverse networks through

structural holes or diverse contacts may more easily obtain the knowledge necessary

to make exploration innovations (Burt, 2004).

However, innovation rates are not just determined by knowledge entering

organizations, but also by how knowledge is turned into innovations that are

launched as products (Fiol, 1996). Much work has sought to identify internal

organizational routines that efficiently generate innovations, as well as factors that

prevent successful innovation development (Burgelman and Sayles, 1986; Leonard-

Barton, 1995; Van de Ven et al., 1999). This work should be extended to also identify

factors that lead to more or less explorative innovations. The present findings showed

rather slow decay of innovation rates, implying that the ability to innovate is

embedded in relatively stable organizational structures and routines. Discovering

exactly what these structures and routines are requires direct examination of the

process.

The findings suggested a large influence of managerial decision making on

organizational innovations, as earlier work has done (Dougherty, 1992; Dougherty

and Heller, 1994). Specifically, managerial problem solving and risk taking seem to

account for the effects of performance relative to aspirations and unabsorbed slack.

This suggests that examination of additional managerial characteristics that may

influence organizational innovativeness will be fruitful. Clearly, characteristics that

predict willingness to accept applications of new knowledge would be particularly

important for exploration innovations. One conjecture is that management teams

with greater cognitive diversity will have greater willingness to launch exploration

Exploration and exploitation in product innovation 25 of 31

innovations because a diverse base of knowledge and interpretation improves the

ability to analyze such innovations (De Dreu and West, 2001).

Though the concepts of exploration and exploitation have deep roots in

organizational theory, empirical research distinguishing these types of innovations is

still in its infancy. This study and other recent empirical work have shown that

organizational theory can give testable predictions on how organizations choose

between exploration and exploitation, and that these predictions have some

empirical success. It seems likely that further investigation of organizational

exploration and exploitation will help develop the theory of organizational

adaptation to uncertain environments.

Acknowledgements

I am grateful for financial support from Japan’s Ministry of Education and for

helpful comments from Wesley Cohen, Jerker Denrell, Tai-Young Kim, Arie Lewin,

Torben Pedersen, Sim Sitkin, Bilian N. Sullivan, Anand Swaminathan, Alva Taylor,

Udo Zander, Xueguang Zhou, and seminar participants at Copenhagen Business

School, Duke University, Hong Kong University of Science and Technology, and the

Prince Bertil Symposium in Stockholm.

Address for correspondence

Henrich R. Greve, Norwegian School of Management BI, 0442 Oslo, Norway.

e-mail: henrich.greve@bi.no

References

Ahmadjian, C. L. and P. Robinson (2001), ‘Safety in numbers: downsizing and the

deinstitutionalization of permanent employment in Japan,’ Administrative Science Quarterly,

46, 622–654.

Amburgey, T. L., D. Kelly and W. P. Barnett (1993), ‘Resetting the clock: the dynamics of

organizational change and failure,’ Administrative Science Quarterly, 38, 51–73.

Amburgey, T. L. and A. S. Miner (1992), ‘Strategic Momentum: the effects of repetitive,

positional and contextual momentum on merger activity,’ Strategic Management Journal,

13, 335–348.

Antonelli, C. (1989), ‘A failure-inducement model of research and development expenditure:

Italian evidence from the early 1980s,’ Journal of Economic Behavior and Organization, 12,

159–180.

Argote, L. (1999), Organizational Learning: Creating, Retaining, and Transferring Knowledge.

Kluwer Academic Publishers: Boston.

26 of 31 H. R. Greve

Argote, L., S. L. Beckman and D. Epple (1990), ‘The persistence and transfer of learning in

industrial settings,’ Management Science, 36, 140–154.

Argote, L. and H. R. Greve (2007), ‘A behavioral theory of the firm–40 years and counting:

introduction and impact,’ Organization Science, 18, in print.

Audia, P. G., E. A. Locke and K. G. Smith (2000), ‘The paradox of success: an archival and a

laboratory study of strategic persistence following a radical environmental change,’ Academy

of Management Journal, 43, 837–853.

Baum, J. A. C., T. Calabrese and B. S. Silverman (2000), ‘Don’t go it alone: alliance network

composition and startups’ performance in Canadian biotechnology,’ Strategic Management

Journal, 21, 267–294.

Beckman, C. M. (2006), ‘The influence of founding team company affiliations on firm

behavior,’ Academy of Management Journal, 49, 741–758.

Beckman, C. M., P. R. Haunschild and D. J. Phillips (2004), ‘Friends or strangers? Firm-

specific uncertainty, market uncertainty, and network partner selection,’ Organization

Science, 15, 259–275.

Benner, M. J. and M. L. Tushman (2002), ‘Process management and technological innovation:

a longitudinal study of the photography and paint industries,’ Administrative Science

Quarterly, 47, 676–706.

Benner, M. J. and M. L. Tushman (2003), ‘Exploitation, exploration, and process

management: the productivity dilemma revisited,’ Academy of Management Review, 28,

238–256.

Blumental, T. (1976), ‘The Japanese shipbuilding industry,’ in H. Patrick (ed.),

Japanese Industrialization and its Social Consequences. University of California Press:

Berkeley.

Bolton, M. K. (1993), ‘Organizational innovation and substandard performance: when is

necessity the mother of innovation,’ Organization Science, 4, 57–75.

Bourgeois, L. J. (1981), ‘On the measurement of organizational slack,’ Academy of

Management Review, 6, 29–39.

Bromiley, P. (1991), ‘Testing a causal model of corporate risk taking and performance,’

Academy of Management Journal, 34, 37–59.

Burgelman, R. A. (1991), ‘Intraorganizational ecology of strategy making and organizational

adaptation: theory and field research,’ Organization Science, 2, 239–262.

Burgelman, R. A. (2002), ‘Strategy as vector and the inertia of coevolutionary lock-in,’

Administrative Science Quarterly, 47, 325–357.

Burgelman, R. A. and L. R. Sayles (1986), Inside Corporate Innovation: Strategy, Structure, and

Managerial Skills. Free Press: New York.

Burt, R. S. (2004), ‘Structural holes and good ideas,’ American Journal of Sociology, 110,

349–399.

Cattani, G. (2005), ‘Preadaptation, firm heterogeneity, and technological performance: a study

on the evolution of fiber optics, 1970–1995,’ Organization Science, 16, 563–580.

Exploration and exploitation in product innovation 27 of 31

Chida, T. and P. N. Davies (1990), The Japanese Shipping and Shipbuilding Industries: A History

of Their Modern Growth. Athlone press: London.

Christensen, C. M. and J. L. Bower (1996), ‘Customer power, strategic investment, and the

failure of leading firms,’ Strategic Management Journal, 17, 197–218.

Cyert, R. M. and J. G. March (1963), A Behavioral Theory of the Firm. Prentice-Hall:

Englewood Cliffs, NJ.

Darr, A. and I. Talmud (2003), ‘The structure of knowledge and seller-buyer networks in

markets for emergent technologies,’ Organization Studies, 24, 435–453.

De Dreu, C. K. W. and M. A. West (2001), ‘Minority dissent and team innovation: the

importance of participation in decision making,’ Journal of Applied Psychology, 86,

1191–1201.

Dewar, R. D. and J. E. Dutton (1986), ‘The adoption of radical and incremental innovations:

an empirical analysis,’ Management Science, 32, 1422–1433.

Dougherty, D. (1992), ‘Interpretive barriers to successful product innovation in large firms,’

Organization Science, 3, 179–202.

Dougherty, D. and T. Heller (1994), ‘The illegitimacy of successful product innovation in

established firms,’ Organization Science, 5, 200–218.

Dyer, J. H. and H. Singh (1998), ‘The relational view: cooperative strategy and sources of

interorganizational competitive advantage,’ Academy of Management Review, 23, 660–679.

Fagiolo, G. and G. Dosi (2003), ‘Exploitation, exploration and innovation in a model of

endogenous growth with locally interacting agents,’ Structural Change and Economic

Dynamics, 14, 237–273.

Ferrier, W. J., C. M. Fhionnlaoich, K. G. Smith and C. M. Grimm (2002), ‘The impact of

performance distress on aggressive competitive behavior: a reconciliation of conflicting

views,’ Managerial and Decision Economics, 23, 301–316.

Fiol, C. M. (1996), ‘Squeezing harder doesn’t always work: continuing the search for

consistency in innovation research,’ Academy of Management Review, 21, 1012–1021.

Galunic, D. C. and K. M. Eisenhardt (2001), ‘Architectural innovation and modular corporate

forms,’ Academy of Management Journal, 44, 1229–1249.

Greve, H. R. (1998), ‘Performance, aspirations, and risky organizational change,’

Administrative Science Quarterly, 44, 58–86.

Greve, H. R. (2003), ‘A behavioral theory of R&D expenditures and innovation: evidence from

shipbuilding,’ Academy of Management Journal, 46, 685–702.

Griffin, A. (1997), ‘The effect of project and process characteristics on product development

cycle time,’ Journal of Marketing Research, 34, 24–35.

Grinyer, P. and P. McKiernan (1990), ‘Generating major change in stagnating companies,’

Strategic Management Journal, 11, 131–146.

Gupta, A. K., K. G. Smith and C. E. Shalley (2006), ‘The interplay between exploration and

exploitation,’ Academy of Management Journal, 49, 693–706.

28 of 31 H. R. Greve

He, Z.-L. and P.-K. Wong (2004), ‘Exploration vs. exploitation: an empirical test of the

ambidexterity hypothesis,’ Organization Science, 15, 481–494.

Howell, J. M. and C. A. Higgins (1990), ‘Champions of technological innovation,’

Administrative Science Quarterly, 35, 317–341.

Ingram, P. and A. Lifschitz (2006), ‘Kinship in the shadow of the corporation: the interbuilder

network in Clyde River shipbuilding, 1711–1990,’ American Sociological Review, 71,

334–352.

Jelinek, M. and C. B. Schoonhoven (1990), The Innovation Marathon: Lessons From High

Technology Firms. B. Blackwell: Oxford, UK.

Kahneman, D. and A. Tversky (1979), ‘Prospect theory: an analysis of decision under risk,’

Econometrica, 47, 263–291.

Kamien, M. I. and N. L. Schwartz (1982), Market Structure and Innovation. Cambridge

University Press: Cambridge, MA.

Katila, R. and G. Ahuja (2002), ‘Something old, something new: a longitudinal study of

search behavior and new product introduction,’ Academy of Management Journal, 45,

1183–1194.

Kelly, D. and T. L. Amburgey (1991), ‘Organizational inertia and momentum: a dynamic

model of strategic change,’ Academy of Management Journal, 34, 591–612.

Ketchen, D. J., Jr. and T. B. Palmer (1999), ‘Strategic responses to poor organizational

performance: a test of competing perspectives,’ Journal of Management, 25, 683–706.

Lant, T. K., F. J. Milliken and B. Batra (1992), ‘The role of managerial learning and

interpretation in strategic persistence and reorientation: an empirical exploration,’ Strategic

Management Journal, 13, 585–608.

Lavie, D. and L. Rosenkopf (2006), ‘Balancing exploration and exploitation in alliance

formation,’ Academy of Management Journal, 49, 797–818.

Leonard-Barton, D. (1995), Wellsprings of Knowledge: Building and Sustaining the Sources of

Innovation. Harvard Business School Press: Boston, MA.

Levinthal, D. A. and J. G. March (1981), ‘A model of adaptive organizational search,’ Journal of

Economic Behavior and Organization, 2, 307–333.

Levinthal, D. A. and J. G. March (1993), ‘The myopia of learning,’ Strategic Management

Journal, 14, 95–112.

Levitt, B. and J. G. March (1988), ‘Organizational learning,’ in W.R. Scott and J. Blake (eds),

Annual Review of Sociology, 14. Annual Reviews: Palo Alto, CA.

Lewin, A. Y., C. P. Long and T. N. Carroll (1999), ‘The coevolution of new organizational

forms,’ Organization Science, 10, 535–550.

Lounamaa, P. H. and J. G. March (1987), ‘Adaptive coordination of a learning team,’

Management Science, 33, 107–123.

March, J. G. (1991), ‘Exploration and exploitation in organizational learning,’ Organization

Science, 2, 71–87.

Exploration and exploitation in product innovation 29 of 31

March, J. G. and Z. Shapira (1992), ‘Variable risk preferences and the focus of attention,’

Psychological Review, 99, 172–183.

March, J. G. and H. Simon (1958), Organizations. New York: Wiley.

March, J. G., L. S. Sproull and M. Tamuz (1991), ‘Learning from samples of one or fewer,’

Organization Science, 2, 1–13.

McDonald, M. L. and J. D. Westphal (2003), ‘Getting by with the advice of their friends:

CEOs’ advice networks and firms’ strategic responses to poor performance,’ Administrative

Science Quarterly, 48, 1–32.

McDonough, E., III (1993), ‘Faster new product development: investigating the effects of

technology and characteristics of the project leader and team,’ Journal of Product Innovation

Management, 10, 241–250.

Miller, D. and M.-J. Chen (1994), ‘Sources and consequences of competitive inertia: a study of

the U.S. airline industry,’ Administrative Science Quarterly, 39, 1–23.

Miller, D. and P. H. Friesen (1982), ‘Innovation in conservative and entrepreneurial firms: two

models of strategic momentum,’ Strategic Management Journal, 3, 1–25.

Miller, K. D. and W.-R. Chen (2004), ‘Variable organizational risk preferences: tests of the

March-Shapira model,’ Academy of Management Journal, 47, 105–115.

Miner, A. S. and S. J. Mezias (1996), ‘Ugly duckling no more: pasts and futures of

organizational learning research,’ Organization Science, 7, 88–99.

Mone, M. A., W. McKinley and V. L. Barker (1998), ‘Organizational decline and innovation:

a contingency framework,’ Academy of Management Review, 23, 115–132.

Nelson, R. R. and S. G. Winter (1982), An Evolutionary Theory of Economic Change. Belknap:

Boston.

Nickel, M. N. and M. C. Rodriguez (2002), ‘A review of research on the negative accounting

relationship between risk and return: Bowman’s paradox,’ Omega, 30, 1–18.

Nohria, N. and R. Gulati (1996), ‘Is slack good or bad for innovation?,’ Academy of

Management Journal, 39, 1245–1264.

Nonaka, I. and H. Takeuchi (1995), The Knowledge-Creating Company: How Japanese

Companies Create the Dynamics of Innovation. Oxford University Press: New York.

Patrick, H. (1986), ‘Japanese high technology industrial policy in comparative context,’

in H. Patrick (ed.), Japan’s High Technology Industries: Lessons and Limitations of Industrial

Policy. University of Washington Press: Seattle.

Rosenkopf, L. and A. Nerkar (2001), ‘Beyond local search: boundary-spanning, exploration,

and impact in the optical disc industry,’ Strategic Management Journal, 22, 287–306.

Ruef, M. (2002), ‘Strong ties, weak ties and islands: structural and cultural predictors of

organizational innovation,’ Industrial and Corporate Change, 11, 427–449.

Schoonhoven, C. B., K. M. Eisenhardt and K. Lyman (1990), ‘Speeding products to market:

waiting time to first product introduction in new firms,’ Administrative Science Quarterly,

35, 177–207.

30 of 31 H. R. Greve

Schulz, M. (2002), ‘Organizational learning,’ in J.A.C. Baum (ed.), Companion to

Organizations. Blackwell: Oxford, UK.

Singh, J. V. (1986), ‘Performance, slack, and risk taking in organizational decision making,’

Academy of Management Journal, 29, 562–585.

Staw, B. M., L. E. Sandelands and J. E. Dutton (1981), ‘Threat-rigidity effects in organizational

behavior: a multi-level analysis,’ Administrative Science Quarterly, 26, 501–524.

Strath, B. (1994), ‘Modes of governance in the shipbuilding industry in Germany, Sweden, and

Japan,’ in J.R. Hollingsworth, P.C. Schmitter and W. Streeck (eds), Governing Capitalist

Economies: Performance and Control of Economic Sectors. Oxford University Press:

New York.

Stuart, T. E. and J. M. Podolny (1996), ‘Local search and the evolution of technological

capabilities,’ Strategic Management Journal, 17, 21–38.

Taylor, A. and H. R. Greve (2006), ‘Superman or the fantastic four? Knowledge combination

and experience in innovative teams,’ Academy of Management Journal, 49, 723–740.

Tsai, W. (2001), ‘Knowledge transfer in intraorganizational networks: effects of network

position and absorptive capacity on business unit innovation and performance,’ Academy of

Management Journal, 44, 996–1004.

Tushman, M. L. and P. Anderson (1986), ‘Technological discontinuities and organizational

environments,’ Administrative Science Quarterly, 31, 439–465.

Van de Ven, A., D. E. Polley, R. Garud and S. Venkataraman (1999), The Innovation Journey.

Oxford University Press: New York.

Wiseman, R. M. and P. Bromiley (1996), ‘Toward a model of risk in declining organizations:

an empirical examination of risk, performance and decline,’ Organization Science, 7,

524–543.

Wiseman, R. M. and A. H. Catanach (1997), ‘A longitudinal disaggregation of operational risk

under changing regulations: evidence from the Savings and Loan industry,’ Academy of

Management Journal, 40, 799–830.

Exploration and exploitation in product innovation 31 of 31