Understanding exploration and exploitation in changing ... - CiteSeerX

OPERATIONS AND TECHNOLOGY MANAGEMENT WORKING PAPER OTM 04-022 2004/09/01

Understanding exploration and exploitation in changing operating routines: the influence of industry and organizational traits

Andrea Masini Department of Operations and Technology Management London Business School email: [email protected] Murizio Zollo Strategy Department INSEAD email: [email protected] Luk van Wassenhove Technology Management Department INSEAD email: [email protected]

Submitted to The Academy of Management Journal

1 September 2004

ALL RIGHTS RESERVED. DO NOT COPY, TRANSMIT OR QUOTE WITHOUT PERMISSION

Abstract This paper analyses a sample of 69 ERP adopters to investigate the role of exploration and exploitation in the development of organizational routines. The results highlight that the relative value of these evolutionary processes vary with the firm strategic orientation, its industry clockspeed and the characteristics of its internal organization. The analysis of moderators yields surprising and unexpected results, which are in sharp contrast with some of the most recent theoretical developments in the field.

1 INTRODUCTION Since Abernathy’s observation that a firm’s excessive focus on productivity may inhibit flexibility and therefore pave the way for its own economic decline (Abernathy, 1978), management scholars have recognized that the degree of success of business organizations is linked not only to their ability to increase productivity, but also to focus on efficiency and innovation simultaneously (Hayes and Abernathy, 1980). Similarly, organization theorists and strategy researchers have noted that a firm’s ability to achieve sustained competitive advantage often resides in its ability to both explore and exploit, i.e. to integrate and build upon its current competencies while simultaneously developing totally new capabilities (Teece et al., 1997). The work of James March (March, 1991) shed further light on the processes of exploration and exploitation and provided inspiration for a stream of works that addressed the same trade off between these two activities (Brown and Eisenhardt, 1998) (Gavetti and Levinthal, 2000) (Benner and Tushman, 2003). Yet, in spite of these conceptualizations and of the large body

2

of research dedicated to the topic, several questions remain unexamined and indicate interesting avenues for further research. For instance, whereas at the theoretical level there is generalized support for the hypothesis that degree of success of a business organization depends on its ability to both exploit and explore, it is less clear how this objective can be achieved in practice, given that the two activities are clearly distinct and that they often require different type of investments and different management approaches. Second, scholars have traditionally dedicated attention to study the two evolutionary processes in relation to new product development or - more recently – in the context of process and operations management (Benner and Tushman, 2003). However, they have generally neglected the question of whether and how these activities affect the development of routines and organizational processes. This is an important gap, as there is ample evidence that the in several instances often linked to the adoption of a new technology firms do need to carefully mix exploration and exploitation activities to design and deploy effective business processes. For instance, the implementation of an enterprise system1 is one of such cases. Faced to a multitude of alternative process templates provided by the software library, which often depart significantly from the configuration currently in use, many firms struggle to find the appropriate balance between the option to explore radically new solutions and that to merely optimize the existing ones. Finally, the internal and external boundaries of both activities remain blurred: if the scarcity of resources renders it impossible to pursue exploration and exploitation simultaneously, under what circumstances should a firm prioritize one activity over the other? Is there any specific contingency factor that determines the relative effectiveness of these two activities?

1

Enterprise resource planning systems (ERP or ES) are large computer systems that – through a common

database and a library of process templates - integrate different application programs in many (possibly all) functions of the firm: (Jacobs and Whybark 2000). 3

If so, how should firms adjust the intensity of their exploration and exploitation efforts to maximize the returns from their investments in different settings? The purpose of this paper is to shed further light on the above questions and to investigate the role played by exploration and exploitation processes in the development of organizational routines and the achievement of improved operational performance. To this end we studied the organizational processes of 69 companies that adopted an enterprise systems from a major European vendor between 1996 and 2000. More specifically, we analyzed the extent to which changes occurred in a vector of key performance indicators after the implantation of the software could be explained by means of other changes occurred in some key process characteristics that reflect the presence of either exploratory or exploitative activities. Since the firms included in the analysis displayed radically different attitudes vis à vis the two evolutionary processes of interest, the chosen research setting represented an excellent testbed to address the research question highlighted above. Finally, drawing upon the manufacturing strategy paradigm that recognizes the importance of process-level competences as a source of competitive advantage (Skinner, 1974) (Clark, 1996) (Skinner, 1996) we deliberately decided to restrict the focus our investigation to measures of performance at the operational level. The results of our analysis provide ample support for the hypotheses that – even in the context of operating routines – companies that conduct exploration and exploitation activities in parallel have clear advantages over firms that chose to emphasize either one of the two processes. Conversely, the analysis of the industry and organizational boundaries yields surprising and unexpected results, which are in sharp contrast with some of the most recent theoretical developments in the field. The reminder of the paper is organized as follows. In sections 2 we discuss an evolutionary model of process effectiveness and we derive testable hypotheses. In section 3 we describe the research setting and the data collection strategy adopted. In section 4 we highlight 4

methodological issues. Finally in sections 5 and 6 we discuss the main results of the analysis and their managerial implications.

2

2.1

EXPLORATION AND EXPLOITATION IN CHANGING OPERATING ROUTINES

B ALANCING EXPLORATION AND EXPLOITATION

After two decades of research on the exploration-exploitation trade off, scholars have concluded that the degree of success of business organizations is linked to their ability to conduct these two activities simultaneously (Hayes and Abernathy, ) (March, 1991). However, whereas this hypothesis has been tested at the firm level making extensive use of measures of business performance, it is still unclear whether it also holds true at the operational and business process level. The resource-based view of the firm (Penrose, 1959) 1959; (Prahalad and Hamel, 1990); (Barney, 1991)) and theory of dynamic capabilities (Leonard-Barton, 1992) (Pisano, 1994) recognize that business success derive primarily from the firm’s “ability to integrate, build and reconfigure internal and external competencies to address rapidly changing environments” (Teece et al., 1997). However, researchers have also suggested that this ability does not arise spontaneously. It is the result of a knowledge evolution process that occurs through a “variation-selection-replication-retention” cycle and that is supported by deliberate investments in experience accumulation, knowledge articulation and knowledge codification (Zollo and Winter, 2001). Through this cycle the firm explores new operational routines (i.e. new tasks sequences and new resource allocation schemes), submits them to market evaluation and, finally, retains and improves the ones that prove to be most efficient. Although a trade-off does exist between these two activities (the exploration of new routines and their exploitation once adopted), it is crystal clear that they both contribute to support the evolutionary cycle through which organizations develop operating routines. Based on the 5

above discussion and in line with existing literature findings we suggest that the combined execution of exploration and exploitation activities have a positive impact even in the context of the development of operating routines. Therefore we propose: H1 An increase in process exploration activities conducted by a firm is positively associated with an increase in the firm’s operational performance H2 An increase in process exploitation activities conducted by a firm is positively associated with an increase in the firm’s operational performance 2.2

INDUSTRY AND FIRM-SPECIFIC BOUNDARIES

If to achieve superior performance firms need to combine both efficiency and innovation, it follows that they should simultaneously invest in exploration and exploitation activities. However, resource scarcity, strategic priorities, industry standards and the fact that both activities have intrinsic advantages and disadvantages often force organizations to favor one particular evolutionary process over the other. For instance, privileging exploitation requires investments in knowledge codification. These investments accelerate the understanding of cause-effect relationships and facilitate the actual implementation and the replication of the organizational procedures. However, they also have explicit disadvantages because they “increase the organizational inertia consequent to the formalization and structuration of task execution” (Zollo and Winter 2001, p. 343), thereby hampering the firm’s ability to promptly respond to environmental changes. The delicate balance between exploration and exploitation activities affects the capabilitybuilding mechanisms of the firm, the nature of these capabilities and, ultimately, the ability of the organization to generate operational improvements. As anecdotal evidence suggests that the relative contribution of these two evolutionary processes is not constant: it varies for firms that operate in different sectors or that exhibit organizational architectures. Accordingly in the following we provide an attempt to shed

6

further light on this relationship. To this end we discuss the role of three general factors that are likely to moderate the relationship between exploration, exploitation and performance. The first factor accounts for the strategic orientation of the firm. The second and the third consider two “metasystems” in which the exploration and exploitation activities occur. The first metasystem is the external environment in which the firm operates: by submitting the newly designed routines to an evaluation, the external environment ultimately determines whether they are appropriate to respond to the specific challenges posed. The second metasystem is the organization itself, i.e. the structured social system composed of the individuals who are called to face the consequences of the exploration and exploitation efforts conducted by the firm. 2.2.1 Strategy The theory of complementarity (Milgrom and Roberts, 1990) and the manufacturing strategy perspective (Skinner 1974; Clark 1996; Skinner 1996) suggest that organizations in pursuit of superior performance should configure their resources and design their operations so as to match their strategic choices. Based on this rationale, it is therefore reasonable to expect that firms would obtain the greatest benefits from the evolutionary process that best fits their strategic objectives. For purposes of conciseness we accounted for the firm’s strategic priorities by reconducting them to the stylized distinction between cost leadership and differentiation. As a first remark, one would expect that firms that prioritize differentiation would be particularly concerned by the threat of seeing their idiosyncratic business processes being imitated. Conversely, they would be presumably less affected if – in order to hedge against this risk, they had to face higher operating cost. As a result, one would expect that these firms would be naturally better off by adopting exploratory mechanisms when updating their

7

operating routines, because these would enable them to keep modifying their underlying architecture, thereby decreasing the risk of imitation. Based on this rationale we suggest: H3 The operational benefits from conducting additional exploration are larger for firms whose business strategy privileges differentiation over cost leadership. By the same token, we expect that organizations that choose to achieve competitive advantage through cost leadership should be less concerned by the risk of being imitated. Conversely, they would naturally prefer to limit the risks associated with the exploration of radically new process configurations, as the latter would typically require longer to be fine-tuned, with the consequence of entailing higher operational costs. Therefore we propose. H4 The operational benefits from conducting additional exploitation are larger for firms whose business strategy privileges cost leadership over differentiation. 2.2.2 Market dynamism We suggest that the external environment influences the evolutionary mechanisms that subsume the generation of a firm’s organizational routines, because it provides the firm with the evaluative feedback through which this assesses the effectiveness of its operational routines and its routine generation mechanisms. The degree of dynamism of an industry (Miller, 1987) - i.e. its clockspeed (Fine, 1998), is one of the environmental attributes that most contribute to determine the nature of this evaluative feedback and, hence to determine which routines and routine-generation mechanisms are more appropriate under given circumstances. Indeed researchers have verified the occurrence of “a positive association between the clockspeed of an industry segment and the speed of the internal clock that paces the internal operations of a business unit in that segment” (Mendelson and Pillai, 1999) p. 8). The resource-based view has also examined the impact of different levels of market dynamism on a firm’s capability generation process. Following (Eisenhardt and Martin, 2000) we portray two antithetical scenarios. In moderately dynamic markets changes occur at a

8

slow pace and along predictable paths. The industry structures are relatively stable, market boundaries are clearly defined and the major customers and competitors are quite well known. Hence, organizations that operate in these environments can obviously heavily rely on previous experience to optimize their operational routines, because the environmental conditions under which this knowledge was developed still hold. Investments in knowledge codification and process exploitation are thus expected to be highly valuable. Conversely, in faster business environments, changes occur at a higher pace and, especially, along paths that cannot be easily predicted. The industry structure is subject to continuous modifications, successful business models are fundamentally unclear and new players continuously replace old business partners. In these circumstances organizations cannot rely on existing knowledge to optimize their operational routines, as these were developed in an environment that has considerably changed afterward. The firm needs to update rapidly both its operational and learning routines. Also, any action that increases the inertia of the system can be intrinsically hazardous. Investments in knowledge codification and process exploitation are therefore expected to be less effective, if not even dangerous. Mindful of this characterization we suggest that the industry clockspeed moderates the relationship between operational performance and the evolutionary processes that subsume the generation of dynamic capabilities. More precisely we propose: H5

The operational benefits from conducting exploration activities are larger for firms that operate in high clockspeed industries than for firms that compete in lowclockspeed industries.

H6

The operational benefits from conducting exploitation activities are larger for firms that operate in low clockspeed industries than for firms that compete in highclockspeed industries.

9

2.2.3 Organizational boundaries Exploration and exploitation activities do not typically occur in a vacuum. They take place inside structured social system, composed of individuals with codified behaviors, working habits and tacit or explicit routines already shaped and often crystallized. For instance, the modification of these routines – which is typically required by any process improvement effort either based on exploration or exploitation, may generate cultural and organizational clashes that diminish the effectiveness of these efforts. However, whereas the fact that organizational traits play a critical role in facilitating or hampering the evolutionary mechanisms that subsume the development of a firm’s adaptive capacity is well acknowledged and substantiated by abundant anecdotal evidence, from a theoretical viewpoint it is less clear what specific organizational traits intervene in these process and why. Since long ago organization theorists have recognized that bureaucracies are a necessary evil. On the one hand they are useful to limit coordination costs and to increase task performance, particularly in a manufacturing environment (Deming, 1986) (Schonberger, 1986) as well as to reduce role ambiguity (Nicholson and Goh, 1983). As such, the deployment of exploitation activities aimed at increasing process efficiency should be facilitated by the existence of structured bureaucracies that rely on formalization, therefore we propose: H7 The operational benefits from conducting exploration activities are larger for firms with low degrees of formalization in their pre-existing organizational routines. H8 The operational benefits from conducting exploitation activities are larger for firms with high degrees of formalization in their pre-existing organizational routines. On the other hand bureaucracies also tend to de-skill employees, to stifle creativity and to decrease a firm’s predisposition to innovate (Bonjean and Grimes, 1970) (Kakabadse, 1986); (Arches, 1991). Hierarchy and excessive formalization may exert a particularly negative impact when they are embedded in a coercive logic (Adler and Borys 1996; Adler 1999). A 10

coercive bureaucracy uses task formalization to impose conformity to existing procedures and to prevent from deviation from standardized routines. Conversely, a bureaucracy that displays enabling characteristics (the opposite of coercive) uses the same task formalization as a tool to encourage employees to search for new and more effective solutions. In this environment formalization is meant to accelerate learning and to facilitate continuous improvement rather than conformity and compliance. As empowerment and the employees’ predisposition to innovate are a necessary condition to modify organizational routines and to increase process flexibility, we expect that exploration activities aimed at increasing process agility should be less effective if conducted in organizations with coercive bureaucracies. Thus we posit: H9 The operational benefits from conducting exploration activities are larger for firms with low degrees of coerciveness in their pre-existing organizational routines H10

The operational benefits from conducting exploitation activities are larger for firms with low degree of coerciveness in their pre-existing organizational routines.

3 THE RESEARCH SETTING

3.1

RESEARCH DESIGN AND SAMPLE SELECTION

The data necessary to perform the analysis were gathered by administering a questionnaire to a randomly selected sample of companies that recently adopted an enterprise resource planning system (ERP or ES). This choice was motivated by two main reasons. First and foremost it was suggested by the fact that the adoption of this type of software forces an organization to modify the mix of exploration and exploitation activities through which it pursues operational improvements. As such, it provides a perfect opportunity to study both the extent to which a change in these two activities affects performance and the internal and external contingencies that affect this process. Second, the choice was motivated by the relevance that the ERP phenomenon has for the business community. Despite the current 11

economic environment, Boston-based AMR Research still estimated a total turnover of $ 84 billion for the ERP ecosystem (software vendors, specialized consultants) and predicted that the demand for these applications to increase at an annual rate of around 30% in 2003. However, as suggested by academic journals and the popular press, which report both horror stories (Bancroft and others, 1998); (Laughlin. S.P., 1999)) and spectacular successes (Umble et al., 2003), there is still conflicting evidence about what ultimately determine an ERP project to succeed or fail, both from practical and theoretical standpoints. This suggests that there is a need for solid empirical research that shed further light on the reasons that lead these systems to succeed or fail. To control for possible confounding effects we required that the firms included in the sample satisfy the following criteria: i) to use the same software from the same ES vendor; ii) to belong to different (yet comparable) industry sectors with different degrees of market dynamism; iii) to have completed the implementation at least one year before they reported the results of their experience (so as to be able to soundly assess the impact produced by the ERP system on their operations after the initial phase of chaos that typically follows the migration); iv) to have experienced antithetical results (i.e. the sample should ideally contain both “champions” and also companies that faced problems either during or after the implementation). For purposes of consistency, we decided to focus exclusively on companies in three main industrial sectors (process industry, discrete manufacturing and consumer products) that adopted SAP R/3 between 1996 and 2000. We also decided to restrict the analysis to four representative European countries (France, Germany, Belgium and Italy) as well as to North America2. The selection of a final sample from the population of firms that satisfied the above criteria was achieved with the valuable assistance of three SAP regional

12

subsidiaries and of two local SAP User Groups (country-based associations of SAP clients completely independent of the software vendor). In each region of interest we asked these organizations to select a sample of around 100 R/3 clients from their population of customers or members (for the SAP User Groups) that met the inclusion criteria and to indicate a contact person in each organization. To limit selection biases, we particularly emphasized the fact that in order to obtain robust results the sample had to contain companies with antithetical implementation histories (i.e. both successes and failures). Both partners had a strong interest in obtaining unbiased results and agreed with this request. Our final sample contained 560 companies in the following countries: France, Germany, Italy, Belgium, US and Canada.

3.2

QUESTIONNAIRE DESIGN AND ADMINIS TRATION

After the selection of the final sample of potential respondents, the first step of our data collection strategy consisted in conducting a series of semi-structured interviews with executives from five European companies in different industrial sectors to i) submit our conceptual framework to a first empirical validation and ii) to define specific metrics for the measurement of the variables included in the models. Based on the feedback from these interviews, we developed a questionnaire in three different languages (English, French and Italian) and we personally pre-tested each version with representative SAP clients in the target countries. The revised version of the questionnaire was finally administered to the 560 companies in the sample via either e-mail or airmail. To limit possible biases we paid particular attention to give respondents strong incentives to provide accurate answers: i) we guaranteed that the information collected would remain

2

The choice was based on two criteria: the fact that the author could master the language of the country chosen

and the relevance of the particular market for SAP. 13

completely confidential (especially vis à vis SAP AG); ii) we agreed to distribute to each respondent a personalized feedback document where each company’s individual project was benchmarked against the overall sample of participants and iii) we agreed to share with respondents the final results of the study. The questionnaire was administered to a general manager who supervised or sponsored the project or who was ultimately involved in performance evaluation. To guarantee that each completed questionnaire could be used in our analysis as a single and representative data point we asked respondents to complete the survey on behalf of the part of the organization that was under their direct responsibility and to report this information. For small companies the unit of analysis typically coincided with the entire firm whereas for larger groups responses mainly referred to the strategic business unit under the direct responsibility of the respondent.

3.3

REPRESENTATIVENESS OF THE SAMPLE

We received a total of 82 answers with a total response rate of around 15%, which was comparable to that of other studies of this nature (Mabert et al., 2003) and judged acceptable given the time and effort required to complete the questionnaire. The sample retained for statistical analysis can be considered representative both of the companies to which the questionnaire was administered and of the entire population of SAP customers in the three industry meta-sectors retained (process industries, discrete and consumer industries), based on the following considerations. First, a comparison of SAP revenue breakdown by sector and of the sample of companies retained for the statistical analysis (Table 1) indicates that the sample of companies retained for the statistical analysis reflects quite closely the characteristics of the population of SAP customers in the three “meta-sectors” retained, with the only difference that consumer

14

industries seem to be slightly over-represented. However, this difference can be easily explained by the fact that companies involved in the manufacturing or distribution of consumer products are on average smaller (and therefore likely to have lower ERP spending) than firms in the process and discrete industries3. As a result, the contribution of these companies can be proportionally larger when measured in terms of number of firms than when assessed with respect to the revenue they generate for SAP. SAP customer population 2000a MEuro %

SAP customer population 2002a MEuro %

Research sample n. firms

%

Total (process, discrete, consumer)

3911

62%

4601

62%

Process industry

1366

35%

1537

33%

24

29%

Discrete manufacturing

1549

40%

1764

38%

29

35%

Consumer products

996

25%

1300

29%

29

35%

Other

2354

23%

2812

23%

82

TOTAL

6265

a

7413

SAP AG annual report

Table 1: SAP revenue breakdown and sample breakdown by sector

Process industry Discrete manufacturing Consumer products TOTAL

Population (Italy and France) 72 33.33% 79 36.57% 65 30.09% 216

Sample (Italy and France) 12 37.50% 10 31.25% 10 31.25% 32

Table 2: Population and sample breakdown by sector: Italy and France

To further assess the representativeness of our sample we analyzed more formally the subsample of Italian companies, for which we had precise demographic information (number of employees) that enabled us to carry out statistical tests. The population included 110 companies (38 in the process industry, 43 in the discrete and 29 in the consumer industry), 15 of which returned the questionnaire. The comparison between the sample and the population shows that no statistically significant differences can be found between the two

3

For instance, this difference is evident and statistically significant (at the 10% level) in the Italian sample, for

which we had more detailed information. 15

groups, at least with respect to the size of the companies. The F and the t-test reported in Table 3 suggest that neither the hypotheses of equal variances nor that of equal means could be rejected (t = 0.11 vs t0.05 = 1.97 and F = 1.08 vs F0.05 = 2.18, non significant at the at 5% level). After eliminating 6 outliers and 7 incomplete questionnaires from the original set of answers, we remained with 69 valid responses that were suitable for statistical analysis (45 companies were located in Europe whereas the remaining 24 were based in North America).

# of employees (st.deviation) Observations Df F t statistics

Population 1466 (2423) 110 109

Sample 1391 (2324) 15 14 1.09 0.11

Table 3: Characteristics of Italian sample

4 METHODS

4.1

OVERALL ANALYTICAL APPROACH

To address the main research questions discussed above we proceeded as follows. First and foremost we operationalized the variables included in our model. Second, we assessed the direct impact of the three learning mechanisms on performance. To this end, we estimated a series of regression models using the aggregated measure of performance derived above as a dependent variable. As none of the variables in the model posed collinearity problems (the largest correlation coefficient was far below 0.5) and as no major theoretical reasons could suggested the occurrence of non-linear phenomena, we decided to use a simple linear model and to estimate it by means of ordinary least square analysis at the benefit of higher parsimony and higher efficiency of the estimators. The analysis of the residuals of all the regressions reported below supported this decision, as none of the error scatterplots deviated significantly from the null plot, thereby confirming both the inherent linearity of the 16

phenomenon and the absence of heteroskedasticity. Finally we focused on the analysis of the internal and external contingency factors that were hypothesized to exert a moderating effect. To ascertain whether the hypothesized moderating variables were simple predictors, quasimoderators, pure moderators or homologizers4 we adopted a procedure suggested by ((Sharma et al., 1981); (Arnold, 1982)), and widely adopted in the management literature ((Yadong Luo, 2000); Kumar and Strandholm 2002; Slater and Narver 1984). For every moderating variable retained in our model, we first estimated a “base model” (hereafter named model “a”) in which the dependent variable Y is simply regressed against the variables that are hypothesized to be direct predictors Y = β 0 + β 1x 1 . Second we estimated an “extended version” of the base model, where the hypothesized moderator X2 is included in the model only as a direct predictor Y = β 0 + β 1 x 1 + β 2 x 2 (model “b”). Third, we estimated a moderated model, that includes both the hypothesized moderator and its cross products with the

direct

predictors

with

which

it

was

expected

to

have

an

interaction

Y = β 0 + β 1x 1 + β 2 x 2 + β 3 x 1 x 2 (model “c”). The difference between models was assessed by

computing the increments in R2 and by conducting an F test for the corresponding hierarchical F5 (Carte and Russell, 2003). If significant statistical difference6 occurred only

4

Pure moderators and quasi moderators affect the form rather than the strength of a relation and are best

detected by means of moderated regression analysis (MRA). However, whereas pure moderators have only an indirect impact on the dependent variable through their influence on the regression coefficient of the primary explanatory variable(s), quasi-moderators have both an indirect and a direct impact. Conversely, homologizers affect the strength rather than the form of the relation and are typically best detected by applying subgroup analysis (Sharma et al. 1981). The distinction has relevant implications: whereas the presence of a moderator may require the implementation of different strategies for different internal or external environments, the occurrence of a homologiser simply demand a change in the emphasis of the strategy adopted. 5

F = (∆R2 /∆K)(N-K2 -1)/(1-R2 2 ) where K2 and R2 2 are the number of predictors and the coefficient of

determination in the extended model. 17

between models “a” and “b”, the hypothesized moderating variable was retained as a pure predictor. If we observed significant statistical difference only between models “a” and “c” the moderating variable was considered as a pure moderator (i.e. it was considered to affect the dependent variable only through its effect on the coefficient of one of the direct predictors). If significant statistical difference occurred among all the three models, the variable was classified as a quasi-moderator (i.e. it was considered to affect the dependent variable both directly and indirectly). Finally, if none of the case above occurred, we conducted a subgroup analysis to ascertain whether the target variable was a homologizer. The difference between the coefficients of the direct predictors in the two subgroups was assessed by converting the betas to partial correlations and using Fisher’s Z-test (Hambrick, 1983)

4.2

OPERATIONALIZATION OF CONSTRUCTS

4.2.1 Independent variables Our first objective was to derive operational measures for the two constructs, exploration and exploitation. To this end we used the semi-structured interviews conducted with industry representatives to identify exploratory and exploitative activities in areas that were relevant to our research setting (adoption of and enterprise resource planning system) and whose intensity was likely to vary after the implementation of the software.

6

It is important to note that - contrary to what sometimes argued – moderated regression analysis (MRA) is an

appropriate technique to test for interaction effects, as it does not alter the test for significance of the interaction terms (Southwood, 1978). However, it is also important to recall that as it does alter the test for the other coefficients, the models that examine the impact of moderating variables must be evaluated only with respect to their overall significance and to the significance of the coefficient of the interaction terms.

18

Given the nature of the technology, the items retained for the analysis related to the general areas of information processing and business process re-engineering. These are: the accuracy, the timeliness and the homogeneity of information (1-3); the amount of time and resources necessary to execute tasks (4-5) and the ability of the organization to deal with unexpected events, the frequency with which it reallocates resources across functions and the frequency with which it modifies the configuration of business processes (6-8). To evaluate the degree of change of these variables, in the questionnaire we asked respondents to consider a representative business function in their unit that had been profoundly transformed by the software introduction and to evaluate on a 7-point likert scale the extent to which each of the above variables had changed after the live date (we considered a time interval of one year after the live date as a reference). To extract the underlying dimensions of change we applied a factor analysis to the eight items 7 (Table 4 displays the rotated factor patterns, the commonality and the proportion of variance explained by each factor for the pooled sample). The eight items loaded on three factors8 that, together, explained about 70% of the variance in the sample. The questions related to information processing loaded consistently on one factor, which we simply named “sensing” to reflect the fact that it refers to the extent to which the organization can gather information on the environment in which it operates. The remaining items loaded on two distinct factors. The items loading on the second factor refer to activities whose nature is clearly exploratory. Therefore, given that it reflects the tendency of an organization to achieve operational improvements by modifying the

7

Kaiser’s Measure of Sampling Adequacy was above .60 for each individual variable considered. Hence, all

items were retained for factor analysis (Kaiser 1970). 8

Throughout all the analysis we used the mineigen criterion to select factors (i.e. we retain only factors whose

eigenvalue was larger than 1). 19

configuration of its processes, possibly to handle non-routine events – we named this factor process exploration. Conversely, the items that load on the third factor assess the amount of time and resources that an organization requires to execute a task. In recognition of the fact that they relate to the tendency of an organization to achieve operational improvements by using its resources efficiently within a given business process configuration, we named this factor process exploitation. Variable Accuracy of information Timeliness of information Homogeneity of information Frequency of job rotation Frequency of process changes Ability to manage non-routine tasks Resources required to execute tasks Time required to execute tasks % of variance explained

Sensing

Exploration

0.79 0.73 0.72 -0.23 -0.10 0.05 -0.03 -0.14 35%

-0.12 0.03 -0.19 0.86 0.80 0.68 0.09 0.19 18%

Exploitation Communality -0.09 0.03 -0.14 -0.06 0.18 0.34 0.89 0.85 15%

0.65 0.54 0.58 0.80 0.69 0.59 0.81 0.78 68%

Table 4: Factor analysis for the effect of ERP adoption on process attributes

To validate our measures, we split the sample into two sub-samples based on the geographical location of the firm (Europe vs. North America) and we repeated the analysis on each of the two regional sub-samples separately. The factors obtained from the two subgroups are consistent with those of the pooled case and are not further discussed. Finally as we wanted to obtain clean measures that could be used in regression analysis, we retained for each dimension the item that had the highest loading on each factor in the pooled sample (Mukerjee 1998). 4.2.2 Moderators 4.2.2.1 Organizational characteristics To operationalize the constructs related to the degree and the type of formalization of the organization (Adler and Borys, 1996), we followed an approach similar to the one described above. As a first step, we used the literature on organizational theory to generate a set of 20

items generally related to the concept of bureaucracy. We then used the interviews that preceded the administration of our questionnaire to ask managers to identify the items that had the largest impact on the implementation of the software and, especially on its daily utilization. After this process we retained six specific items coded on a 7-point likert scale: i) the existence of a structured hierarchy with a clear separation of roles, ii) the extent to which tasks and responsibilities are clearly defined inside teams, iii) the extent to which manuals, written documents and other formal procedures are used to facilitate the execution of tasks, iv) the extent to which the use of cross-functional teams was common in the organization (which reflects the firm attitude to be process-oriented as opposed to function-oriented), v) the extent to which employees are encouraged to bypass formal rules to complete their jobs and vi) the extent to which decisions need to be approved by a supervisor before being implemented. In addition, as we feared that many respondents could overemphasize the degree of structuration of their organization, we added a fifth question that dealt with the extent to which salaries are dependent on the formal position held (which was expected to measure the previous construct in a less subjective fashion). We initially applied factor analysis to all seven questions. However, as the item that measured the frequency of use of cross-functional teams exhibited an unacceptable ( 0, i = exploration, exploitation) could not lead to the rejection of the null hypothesis 27

for either of the two variables, thus suggesting that the intensity of these activities is a deliberate choice of the firm and it is not univocally influenced by the technology adopted. Second, we tested for the relative differences between the magnitude of the changes in the intensity of the exploration and exploitation activities conducted after the ERP implementation. Based on the result of the test, we could not reject the null hypothesis that the two means are different (t = 2.54 with p = 0.01), although the absolute value of the difference is negligible. Finally, we conducted a one-tailed F test to test for the difference between the variances of the two variables (H0: σ2exploration = σ2exploitation, H1: σ2 exploration < σ2 exploitation ).

The test was not significant, therefore reinforcing the hypothesis that the intensity

of both the exploratory and the exploitative activities of the firm are not directly influenced by the type of technological choices operated.

5.2

EXPLORATORATION AND EXPLOITATION AS KEY ACTIVITIES FOR THE ACHIEVEMENT OF SUPERIOR OPERATIONAL PERFORMANCE

As a second step, we turned our attention to ascertain the impact exerted by our two main process variables, exploration and exploitation, on operational performance. Table 8 reports the unstandardized regression coefficients and the overall explanatory power (adjusted and non adjusted) of the main models retained together with the increase in R2, the F ratios and the corresponding significance levels observed after the inclusion of additional variables. In the basic models 1, 2 and 3 after controlling for location, the changes in operational performance observed after the ERP implementation are explained only by means of differences in the quality of project execution, by the duration and the size of the project. In models 4, 5 and 6 the three explanatory variables related to the degree of change in process

28

exploration, process exploitation and sensing were progressively added13. The first three models – which include only the control variables, somehow reflects the “null” hypothesis that - regardless of the changes that occur in the magnitude of the exploratory and exploitative efforts conducted by the firm, an extensive ERP implementation that is properly managed (on time and with no budget overruns) would guarantee performance improvements, regardless of the efforts spent by the adopter to implement new organizational processes (i.e. to conduct exploration) or to further exploit the existing ones. The first interesting observation that emerges from the analysis is that the “null models” have virtually no explanatory power and that the control variables are not significant (with the exception of project duration). Contrary to the wisdom prevailing in many IT circles, an effective project management and the successful implementation of new software cannot by themselves guarantee the achievement of performance improvements. To do so, they need to be accompanied by appropriate efforts aimed at facilitating the exploration of new process configurations or at optimizing the existing ones. This is quite evident from models 4, 5 and 6. Regardless of the order followed, the process variables related to exploration, exploitation and sensing increase the explanatory power of the models and are highly significant. The model explains a significant amount of the overall variance of the sample when the two process variables are introduced, both independently and jointly (R2 = .242, ∆R2 = .06 significant at the 1% level when process exploration is included; R2 = .485, ∆R2 = .243 significant at the 1% level when process exploitation is also added). As argued in hypotheses 1 and 2, the two process-related variables have a strong and positive

13

The three process variables were also entered into the models according to different sequences. For purposes

of conciseness the results are not reported here. In all the combinations tested the variables added caused a statistically significant increase of the model’s explanatory power and displayed statistically significant coefficients. 29

impact on operational performance, and they do so consistently for all the models tested (i.e. even after entering additional variables). It is also worth stressing that – although not sufficient by itself to account for a significant amount of variance - project duration has a positive and significant impact in all the models tested. These supports our initial conjecture that long implementations are likely to be associated with serious BPR efforts and, in turn, contribute to generate operational benefits. In accordance with earlier conceptualizations (March, 1991) the analysis suggests that companies in pursuit of operational excellence should conduct exploration and exploitation activities in parallel. To further validate this claim, we subdivided the sample into four subgroups based on magnitude of the changes observed in the intensity of the two activities after the implementation of the software and we tested for performance differences across subgroups. The comparison suggests that the performance improvements achieved by companies that increased the intensity of the two activities simultaneously are significantly greater than the performance improvements of companies that did not sufficiently engage in any of the activities or that provided exploratory efforts without parallel investments in exploitation (F = 5.04 significant at the .01 level based on a Scheffe test). As an additional remark, it is also interesting to note that - in sharp contrast with most of the common fads that identify budget overruns and project delays as the primary culprit for the failure of large IT projects, our analysis suggests that deviating from planned budget or schedule has virtually no direct influence on key performance indicators (coefficient never significant and always close to zero). Needless to say, this result does not imply that the management of an implementation has no influence at all. It simply indicates that – in order to be effective – these efforts should be focused on improving the true sources of operational excellence that emerge from the analysis above. Model 1

Model 2

Model 3

Model 4

Model 5

Model 6 30

Control Variables Firm size Location

0.00 (0.00) -0.10 (0.21)

Quality of project mgt

0.00 (0.00) -0.12 (0.21) 0.06 (0.08)

Project Duration

0.00 (0.00) -0.14 (0.21) 0.06 (0.08) 0.03** (0.01)

Dynamic capabilities Sensing

0.00 (0.00) -0.02 (0.21) 0.07 (0.08) 0.02 (0.01)

0.00 (0.00) -0.06 (0.20) 0.06 (0.07) 0.02* (0.01)

0.00 (0.00) -0.20 (0.16) 0.01 (0.06) 0.02** (0.01)

0.27*** (0.09)

0.22** (0.09) 0.14** (0.06)

0.18** (0.07) 0.13** (0.05) 0.30***

0.182 0.117 2.79** 0.024 0.107 8.23*** 69

0.242 0.168 3.29*** 0.007 0.060 5.23** 69

0.485 0.426 8.21***