Journal of Marketing ResearchVol. XLI (August 2004), 293305293

*Werner Reinartz is Associate Professor of Marketing, INSEAD (e-mail: werner.reinartz@insead.edu). Manfred Krafft is Professor of Market-ing and Director of the Institute of Marketing, Westphalian Wilhelms Uni-versity Mnster (e-mail: mkrafft@uni-muenster.de). Wayne D. Hoyer isJames L. Bayless/William S. Farish Fund Chair for Free Enterprise andChair, Department of Marketing, McCombs School of Business, Univer-sity of Texas at Austin (e-mail: wayne.hoyer@mccombs.utexas.edu). Theauthors gratefully acknowledge financial support from the Teradata Centerfor Customer Relationship Management at Duke University. Additionalfunding was provided by the INSEAD Research and Development Depart-ment, the Center for Customer Insight (McCombs School of Business,University of Texas at Austin), and the Center for Market-Oriented Man-agement (WHU-Koblenz). The authors thank Rick Staelin and WagnerKamakura from the Teradata Center at Duke University, Christoph van denBulte, Hubert Gatignon, Frenkel ter Hofstede, and participants of the 2003Marketing Science Conference for comments on a previous version of thisarticle. They also thank Heiko Mller for valuable research assistance.

WERNER REINARTZ, MANFRED KRAFFT, and WAYNE D. HOYER*

An understanding of how to manage relationships with customerseffectively has become an important topic for both academicians andpractitioners in recent years. However, the existing academic literatureand the practical applications of customer relationship management(CRM) strategies do not provide a clear indication of what specificallyconstitutes CRM processes. In this study, the authors (1) conceptualize aconstruct of the CRM process and its dimensions, (2) operationalize andvalidate the construct, and (3) empirically investigate the organizationalperformance consequences of implementing CRM processes. Theirresearch questions are addressed in two cross-sectional studies acrossfour different industries and three countries. The first key outcome is atheoretically sound CRM process measure that outlines three keystages: initiation, maintenance, and termination. The second key result isthat the implementation of CRM processes has a moderately positive

association with both perceptual and objective company performance.

The Customer Relationship ManagementProcess: Its Measurement and Impact onPerformance

An understanding of how to manage customer relation-ships effectively has become an important topic for bothacademicians and practitioners in recent years. Organiza-tions are realizing that customers have different economicvalue to the company, and they are subsequently adaptingtheir customer offerings and communications strategyaccordingly. Thus, organizations are, in essence, movingaway from product- or brand-centric marketing toward acustomer-centric approach.

Nevertheless, some key problems need to be addressed.Although the conceptual underpinnings of a customer rela-

tionship management (CRM) strategy are hardly ques-tioned, the implementation challenges appear to be enor-mous, as evidenced by commercial market research studies.These studies provide some convergent validity that approx-imately 70% of CRM projects result in either losses or nobottom-line improvement in company performance (Gart-ner Group 2003).

Previous studies have focused on components of CRMstrategy, such as the link between satisfaction and businessperformance (Kamakura et al. 2002), the link between cus-tomer loyalty and profitability (Reinartz and Kumar 2000),customer profitability heterogeneity (Niraj, Gupta, andNarasimhan 2001), and customer loyalty programs (Verhoef2003). However, there is a severe lack of research that takesa broader, strategic focus across firms. There is no clear evi-dence regarding either the characteristics of successfulCRM approaches or the reasons CRM may potentially fail.Furthermore, the existing academic literature and practicalapplications of CRM do not provide a clear indication ofwhat specifically constitutes the implementation of CRMprocesses. Some companies view CRM primarily as invest-ments in technology and software, whereas others treatCRM more expansively and are aggressive in developingsound and productive relationships with customers. In addi-tion, some companies have implemented CRM processes toa greater degree than others. Therefore, it is important toidentify the types of CRM activities that companies can

294 JOURNAL OF MARKETING RESEARCH, AUGUST 2004

employ and to explore how these relate to company per-formance and profitability.

Given this situation, the key contribution of this article isto conceptualize and operationalize a measure of the degreeto which CRM processes have been implemented. In partic-ular, we examine the functional and organizational compe-tencies that are necessary to conduct effective and profitableCRM activities. Developing such a measure or index ofCRM processes will enable us to determine whether higherlevels of CRM implementation are associated withimproved economic performance. We further examine someconditions in which CRM processes are associated withsuperior performance outcomes (i.e., moderators of thisrelationship).THEORETICAL FOUNDATION OF THE CRM PROCESS

A challenge of defining CRM is that any definition iscontingent on the level at which CRM is practiced in anorganization or, for that matter, what the researcher or man-ager believes about the correct level of CRM. There arethree different possible levels: (1) functional, (2) customer-facing, and (3) companywide.

In this article, we focus on the CRM process on thecustomer-facing level. This perspective includes the build-ing of a single view of the customer across all contact chan-nels and the distribution of customer intelligence to allcustomer-facing functions. This view stresses the impor-tance of coordinating information across time and contactchannels to manage the entire customer relationship sys-tematically. For example, a bank customer who has both aloan product and a savings product might interact with thebank through various channels and different types of inter-actions (e.g., transaction, information request, complaint),which may change over time. A CRM process on thecustomer-facing level would capture these interactions and,on the basis of the generated intelligence, would result incoordinated and well-defined actions through differentfunctions.

A key question is, How should the CRM process be con-ceptualized at the customer-facing level? The literature sug-gests that companies should recognize four distinct factors:(1) Building and managing ongoing customer relationshipsdelivers the essence of the marketing concept (Morgan andHunt 1994; Webster 1992), (2) relationships evolve withdistinct phases (Dwyer, Schurr, and Oh 1987), (3) firmsinteract with customers and manage relationships at eachstage (Srivastava, Shervani, and Fahey 1998), and (4) thedistribution of relationship value to the firm is not homoge-neous (Mulhern 1999; Niraj, Gupta, and Narasimhan 2001).

A key theoretical basis for CRM research is the relation-ship marketing literature. First, in this area it is theoreticallyheld that the building and managing of ongoing customerrelationships delivers the essence of the marketing concept(Morgan and Hunt 1994; Webster 1992). The new institu-tional economics approach uses economic theory to explainthe development and breakdown of customerfirm relation-ships. For example, transaction cost theory (Rindfleisch andHeide 1997) focuses on minimizing the cost of structuringand managing relationships and maximizing the returnsfrom them. Common to all theoretical approaches in therelationship marketing literature is that managing relation-ships is beneficial for the firm. This perspective has

received preliminary support from Reichheld and Teal(1996). However, the observations have been tempered byempirical evidence (e.g., Niraj, Gupta, and Narasimhan2001; Reinartz and Kumar 2000) that stresses the impor-tance of moderating effects. Thus, it is probably not truethat more relationship building is always better; rather,building the right type of relationship (which depends onsituational factors) is critical. In other words, facilitatorssuch as organizational design, adequate incentive schemes,and information technology resources, as well as industry,company, or customer structures, may affect the perform-ance of relationship marketing activities.

The second aspect of our conceptualization is that theCRM process should acknowledge that relationships evolvewith distinct phases (Dwyer, Schurr, and Oh 1987). Thus,relationships cannot be viewed as multiple independenttransactions; rather, the interdependency of the transactionscreates its own dynamic over time. In other words, CRMprocesses are longitudinal phenomena. The process of rela-tionship evolution can be subject to termination at any pointthrough customer causes (ceasing of category consump-tion), competitive causes, or internally unintended (attritionthrough service problems) or internally intended (customerfiring) causes.

The third aspect is that the recognition of relationshipevolution has implications for the organization: Firmsshould interact with customers and manage relationshipsdifferently at each stage (Srivastava, Shervani, and Fahey1998). For example, Jap and Ganesan (2000) find that theeffect of transaction-specific investments on relationshipcommitment in manufacturerretailer relationships is posi-tive in the exploration and the decline phases. A goal ofCRM is to manage the various stages of the relationshipsystematically and proactively. For example, companiessystematically attempt to mature relationships by cross-selling and up-selling products with high purchase likeli-hood (Kamakura et al. 2002).

The fourth aspect is the recognition that the distributionof relationship value to the firm is not homogeneous (Mulh-ern 1999; Niraj, Gupta, and Narasimhan 2001). This is aconsequence of the increasing adoption of recent account-ing practices, especially activity-based costing. The keyadvantage of activity-based costing is that firms are able tomake profitability statements along customer relationshiplines, not only along product lines. This enables firms toinvestigate resource allocations that are made against thecustomer relationship profitability distribution. A commonfinding is that best customers do not receive their fair shareof attention and that some companies overspend on mar-ginal customers. In a CRM paradigm, a key goal is to definedifferent resource allocations for different tiers of cus-tomers, where the customers tier membership depends onthe economic value of that customer or segment to the firm(Zeithaml, Rust, and Lemon 2001).

The continuous balance of CRM activities at each stage(i.e., customer acquisition, retention, and relationship termi-nation) should be guided by the attempt to maximize thevalue of the set of concurrent customer relationships andthus should be associated with better overall company per-formance. Therefore, we define the CRM process at thecustomer-facing level as a systematic process to managecustomer relationship initiation, maintenance, and termina-

Customer Relationship Management Process 295

tion across all customer contact points to maximize thevalue of the relationship portfolio.

Thus, our view of the CRM process entails the systematicand proactive management of relationships as they movefrom beginning (initiation) to end (termination), with exe-cution across the various customer-facing contact channels.This necessitates both information generation through theanalysis of customer and prospect needs and behavior andaction on this information, contingent on the customersvalue and life-cycle stage. We attempt to capture the multi-dimensional components (life-cycle stage, customer evalua-tion, and interaction) in a multilevel model.

Similar to other multilevel models in the literature (Bradyand Cronin 2001), our model suggests that each of the pri-mary dimensions of the CRM process (relationship initia-tion, maintenance, and termination) has distinct subdimen-sions. Customer evaluation is the first subdimension of eachprimary dimension. The subsequent subdimensions areacquisition and recovery management for the initiationstage; retention, up-selling/cross-selling, and referral man-agement for the maintenance stage; and exit managementfor the termination stage. These nine subdimensions pro-vide a structure for different CRM-related activities andserve as the basis for a conceptual framework for the CRMprocess construct. We consider the nine subdimensionsformative (i.e., consisting of explanatory combinations ofindicators that cover the distinct activities involved).

Our conceptualization is intended to measure how sys-tematic firms are in practicing the various activities of theCRM process. We believe that it is important to capture thesystematic aspects of the process, particularly if the processis practiced on a large scale, such as in a business-to-consumer environment. If firms formalize their CRMefforts, they become more consistent in execution acrosscontact channels, employees, and the portfolio of cus-tomers. It is important to note that we do not mean formal-ization in terms of rigidity but in terms of conformance tospecification. For example, firms want to avoid the mistakeof not identifying a good customer and subsequently notrewarding the customer accordingly (Type I error). Firmsalso want to prevent wrongful classification of low-valuecustomers as high-value customers and subsequent over-spending of resources (Type II error). The development ofand reliance on a systematic approach that aids in the meas-urement of customer value and in the interaction with theseheterogeneous customers decreases these errors.

It is important to compare our approach with otherframeworks that address similar issues, including the serv-ice profit chain (Heskett et al. 1994), return on quality(Rust, Zahorik, and Keiningham 1995), customer assetmanagement (CAM; Berger et al. 2002), and customerequity (CE; Blattberg, Getz, and Thomas 2001; Rust,Lemon, and Zeithaml 2001). All four approaches arecustomer-centric, and customer knowledge (e.g., customerdatabases, surveys) is critical to their implementation. How-ever, whereas the service profit chain and return on qualityapproaches address service quality issues, the CAM and CEapproaches, as well as our measure of CRM processes,focus more on companies identifying profitable customersand treating them adequately. The CAM and CE approachesdeal more with the application of traditional marketingtechniques to manage customer assets in terms of homoge-

neous customer segments. In contrast, CRM expands onthis approach by supplementing traditional marketing tech-niques with other relationship management activities (e.g.,systems to regain lost customers, up-selling and cross-selling, referral management) at the clearly identified stagesof the customer relationship (i.e., initiation, maintenance,and termination). Finally, CAM and CE focus on customersegments as assets, whereas our CRM process frameworkcenters on individual customer relationships. Thus, ourCRM approach supplements the important principles thatemanate from these other frameworks. In addition, it isimportant to note that the key concept of customer satisfac-tion is a central foundation across all these approaches(Oliver 1999).

A MODEL OF THE PERFORMANCE OUTCOMES OFCRM IMPLEMENTATION

Adoption and implementation of the CRM process isonly the initial part of the story. It is also critical to establishwhether CRM is a good thing for the company. Given thedearth of sound empirical findings in the domain and thatevidence now suggests that CRM strategies may not per-form as well as many people had expected, an investigationof the CRM processeconomic performance link should beof great interest to managers and academics. Thus, a secondgoal of this article is to conceptualize and test a model ofhow the three primary CRM dimensions are associated withorganizational performance. Figure 1 presents an overviewof the theoretical model, which has two key components.First, we investigate the main effect of the CRM process oneconomic performance. Second, we examine moderatingeffects, which may serve to establish some contingencyconditions.

First, in terms of performance outcomes, we relate thethree CRM dimensions to two types of performance meas-ures: perceptual and objective. Although most research inmarketing strategy assesses the impact of the focal con-

Figure 1A MODEL OF THE PERFORMANCE OUTCOMES OF THE CRM

PROCESS

H1

Economic Performance

Perceptual

Relationship initiation

ControlIndustry

Moderators

CRM-compatibleorganizational alignment

CRM technology

ObjectiveRelationship termination

Relationship maintenance

CRM Process

H2 H3

296 JOURNAL OF MARKETING RESEARCH, AUGUST 2004

struct on perceived performance (e.g., Bharadwaj, Varadara-jan, and Fahy 1993; Kohli and Jaworski 1990), we alsoassess the association with a measure of objective economicperformance (Varadarajan and Jayachandran 1999).

Second, regarding the contingencies of the CRMprocesseconomic performance link, we examine severalimportant moderating variables that are of interest to man-agers and that may either enhance or weaken the focal link(Bharadwaj, Varadarajan, and Fahy 1993; Holmstrm1979). Supply-side characteristics include a CRM-compatible organizational alignment (i.e., training proce-dures, employee incentives, and organizational structure)and CRM technology (e.g., investments in CRM technol-ogy, one-to-one communication capabilities). Finally, thespecification of our model controls for the types of indus-tries investigated.

HYPOTHESIS DEVELOPMENTEffects on Economic Performance

As we mentioned previously, our CRM process constructcaptures the degree of formalization of how to manage cus-tomer relationship initiation, maintenance, and termination.If companies are able to understand more effectively thevalue of the customer to the firm, they will perform betteron these three primary dimensions. Companies will then bebetter able to manage individual customer relationships andto determine more effectively the contribution of these rela-tionships to the profitability of the unit and/or the firm.

A high degree of CRM process implementation meansthat firms are able to adjust their interactions according tothe life-cycle stages of their customers and that they may beable to influence the stages actively (e.g., by maturing orextending relationships; Zeithaml, Rust, and Lemon 2001).The goal of these activities is to align the resources spent oncustomers with the revenues or profits derived from thesame customers (Mulhern 1999). Firms will spend a dispro-portionate amount of resources on highly profitable cus-tomers or ones that are worth the resource allocationbecause they are high potentials. Furthermore, firms willeconomize on unprofitable or marginally profitable cus-tomers, who then either may leave the relationship or maybuild up their relationship with the focal firm. Therefore,we expect a significant and positive association between thedegree of a business units customer management practiceswith regard to relationship initiation, maintenance, and ter-mination and the business units economic performance.

H1: Higher economic performance is associated with greaterimplementation of CRM processes at the stage of relation-ship (a) initiation, (b) maintenance, and (c) termination.

It should be noted that though all three subsections of thehypothesis are in the same direction, the possibility existsthat the magnitude of the effect across the three stagesvaries. Therefore, there is a question of whether the effec-tiveness of the different stages can differentially contributeto economic performance. Unfortunately, prior researchdoes not provide guidance to enable the development ofspecific hypotheses. However, we perform an exploratoryanalysis to address this important issue.

As we mentioned previously, there are several factors thatmay moderate the relationship between the implementationof the CRM process and economic performance. We exam-

ine two moderators that have been identified as havingstrong theoretical and/or managerial relevance and impact:CRM-compatible organizational alignment and CRMtechnology.

CRM-Compatible Organizational AlignmentDay (1992) argues that the various corporate functional

units have become more marketing oriented because mar-keting is becoming more important. Likewise, the view thatthe marketing function is distinct and nonoverlapping withother corporate functions has become mostly obsolete(Webster 1992). Therefore, as firms become able to aligntheir organizations and structures more effectively withtheir market goals, it is expected that they would be moresuccessful in that market because they can adapt more read-ily to the needs of customers. To address these needs, thereis an imperative to bring customer knowledge and orienta-tion deeper into the organization (Day and Montgomery1999; Kohli and Jaworski 1990).

A critical determinant of an organizations ability toinfluence CRM-compatible activities and processes is thedevelopment of appropriate compensation schemes andorganizational structures. For example, agency theoryargues that the design of incentive-compatible contractswith employees that realign company goals and the employ-ees utility is necessary to maximize company profit (Holm-strm 1979). Consistent with this argument, contingencytheory hypothesizes that company profit will be maximizedif appropriate organizational structures are depicted (Blackand Boal 1994; Miller 1996). The more that these aspectssupport specific CRM-compatible behavior, the stronger theCRM processeconomic performance link should be. Inothers words, if companies stress to employees that CRMactivities are important, structure their organizations tofacilitate these activities, and reward employees for engag-ing in CRM-related activities, the companies are more likelyto stress these activities in their interactions with customers.

H2: The greater the level of CRM-compatible organizationalalignment, the stronger is the positive link between eco-nomic performance and relationship (a) initiation, (b) main-tenance, and (c) termination.

CRM Technology

Another critical moderator of the CRM processeconomic performance link may be the degree to which afirm uses supporting information technology. In this con-text, CRM technology is the information technology that isdeployed for the specific purpose of better initiating, main-taining, and/or terminating customer relationships. Thepotential for information technology to constitute a sustain-able competitive advantage has been amply discussed(Bharadwaj, Varadarajan, and Fahy 1993). The key point isthat CRM technology plays a critical role in the context ofleveraging CRM-related activities and thus contributes toimproved organizational performance in the market; indeed,CRM technology is often (incorrectly) equated with CRM.Therefore, we would expect that, ceteris paribus, CRMtechnology functions as a facilitator of CRM activities andcontribute to better performance in the market.

Nevertheless, this strong conceptual support should betempered in light of evidence from practitioner and com-

Customer Relationship Management Process 297

mercial market research reports that investments in CRM-related technology may be associated with lower economicperformance. Day (2000) echoes this view by suggestingthat though the cost aspects of CRM investment are evident,the revenue enhancing aspects are much less obvious. Fur-thermore, there is anecdotal evidence that a large proportionof CRM technology deployments do not perform to expec-tations (Gartner Group 2003).

If this is so, there are likely to be multiple reasons, suchas lack of defining objectives or lack of appropriate trainingprocedures, for this disappointing result (Reinartz andChugh 2002). However, this does not necessarily mean thatthe technology is at fault per se. It is also important to pointout that investments in technology represent a direct short-term financial investment that may have a negative effect onthe bottom line in the short run. The payoffs for theseinvestments are more likely to be realized over a longerperiod.

Taken together, it is clear that there are conflicting argu-ments about the direction of the effect of CRM technologyon firm or economic performance. However, because thereseems to be more evidence on the positive side, we stillhypothesize a positive moderating effect for CRM technol-ogy. Thus:

H3: The greater the level of CRM technology, the stronger is thepositive link between economic performance and relation-ship (a) initiation, (b) maintenance, and (c) termination.

Control Variable

To control for the possibility of variance across differentindustries, we entered the type of industry as a control. Thisenables us to account for mean differences of economic per-formance across industries.

METHODOLOGYTo test our framework, we collected data from both pri-

mary and secondary sources. First, a cross-sectional surveywas conducted in the fall of 2001 in three countries: Aus-tria, Germany, and Switzerland. We limited our investiga-tion to consumer markets because business-to-business rela-tionships are characterized by small numbers of customersand a strong reliance on salespeople as major means ofcommunication between firms and clients. In our initialempirical work on CRM, we wanted to target a morediverse environment of multiple customer contact points,which is characteristic of consumer markets. Using litera-ture reviews and pretest interviews, we selected industrieson the basis of characteristics such as a large customer base,intensive use of various channels, professionalism in CRMactivities, and market pressure to differentiate from compe-tition. On the basis of these characteristics, we identified thefollowing industries as adequate targets: financial services,hospitality, online retailing, and power utilities.

A pretest of the questionnaire was given to a small sam-ple of marketing managers and CRM experts. Anotherpretest of the questionnaire was administered to assess thevalidity of the scales. We obtained the data from a large-scale mail survey. The final questionnaire was sent to asample of 1015 companies, which we derived from industryassociations member lists. A personalized mailing was sentto the executives identified in premailing telephone calls as

responsible for CRM operations. Whenever possible, weasked potential respondents to provide their e-mail addressand to fill out an electronic version of the questionnaire. In72% of the cases, we received digital responses rather thantraditional mail responses. To increase the response rate, weconducted follow-up telephone calls three weeks after theinitial mailing. This resulted in an effective response rate of21.1%. We consider this rate satisfactory, given that averagetop management survey-response rates are in the range of15% to 20% (Menon et al. 1999). Altogether, we obtained214 responses, of which 211 were usable.

In more than 75% of cases, senior executives such asmarketing or sales executives filled out the questionnaires.The executives were knowledgeable key informants aboutinformation pertaining to CRM design; they direct entitiesthat, in most cases, are responsible for CRM activities. Theunit of analysis was the strategic business unit (SBU).

To strengthen the insight and veracity of our analysis, wealso collected objective performance measures for the exist-ing set of firms. This is particularly important for empiricalsurvey research in which a reliance on subjective perform-ance measures may be a limitation (Jaworski and Kohli1996). Our goal was to assess the degree to which the sub-jective and the objective performance measures converge inorder to lend greater credibility to our survey results (Han,Kim, and Srivastava 1998). Because our sample consists ofpublic and nonpublic firms from different industries, wecould not rely on absolute performance measures; rather,we needed measures of relative performance. As in previousstudies, we assessed performance in terms of profitability(Han, Kim, and Srivastava 1998; McKee, Varadarajan, andPride 1989). We obtained the information on profitabilityfrom company reports for public companies and from sec-ondary sources for nonpublic companies. We chose thereturn on assets (ROA) performance measure, which is con-sistent with previous studies (Han, Kim, and Srivastava1998). In total, we were able to collect the objective per-formance measures for 98 companies (81 public, 17 non-public). The ROA measure that entered our analysis was theaverage ROA of the years 2001 and 2002. It is more appro-priate for us to use the average because it is more realistic toexpect a longer-term impact of the CRM process rather thana short-term spike.

A possible concern in single-informant studies is that aninformant may not necessarily possess a totally accurate orunbiased view of the entire organization. Relatedly, the reli-ability of the subjective performance indicators used in thestudy could be questioned (i.e., they could be artificiallyrelated to the other indicators measured). Therefore, tocross-validate the analysis and to counter a possiblecommon-method bias, we collected a second set of primarydata from a different set of respondents in the same firmsample (hereafter, Sample 2). The objective was to assessthe robustness of the Sample 1 findings with a separatesample of respondents (Deshpand, Farley, and Webster1993). The sampling frame was the 211 companies thatresponded in the first round of data collection. We collectedthe second set of data as soon as possible after we con-cluded Sample 1 to minimize any temporal biases. In Sam-ple 2, we obtained 95 valid responses (45% response rate)from the same group of target respondents (senior execu-tives, sales managers, marketing managers). Because a sub-

298 JOURNAL OF MARKETING RESEARCH, AUGUST 2004

stantial percentage of participants in our study were smalland medium-sized enterprises, it was extremely difficult toidentify a second knowledgeable informant in such compa-nies. Many respondents from Sample 1 were also reluctantto name second informants because they did not appreciatethe cross-validation procedure. To assess potential differ-ences in sample respondents, we compared Sample 1 andSample 2 respondents on several descriptive variables; how-ever, no differences between the groups were found.

Item Measurement and Index ConstructionAs mentioned previously, the key goal of our study was

to develop a comprehensive operationalization of the threeprimary dimensions of CRM process implementation (i.e.,relationship initiation, maintenance, and termination). Toachieve this goal, scales and measurement items for thestudy were developed as follows: All our constructs reflecta composite of individual indicators across different, uniquesources and are therefore operationalized effectively in aformative rather than reflective way (Bagozzi 1994). There-fore, we followed the guidelines for constructing indexesbased on formative indicators, as proposed by Diaman-topoulos and Winklhofer (2001). They identify four issuesthat are critical to successful construction of indexes withformative indicators: (1) content specification, (2) indicatorspecification, (3) indicator collinearity, and (4) externalvalidity. Our focal independent variables are the three pri-mary dimensions of the CRM process. To exemplify howwe proceeded to construct valid indexes with formativeindicators, we refer to these key constructs.

Content specification. We developed a new formative,multi-item scale of CRM processes at the customer-facinglevel that captures the three lifetime stages of customer rela-tionships. More precisely, on the construct level, the domainof CRM process implementation was defined as coveringthe activities of acquisition management and regain man-agement at the initiation stage; retention management, up-sell/cross-sell management, and referral management atthe maintenance stage; and termination management at thefinal stage of the customer relationship. On the constructlevel, we also captured activities of customer evaluation ateach of the three stages, which led to nine subdimensions ofCRM process implementation. The subdimensions repre-sent latent constructs that reflect the presence or absence ofCRM activities. We also established higher-level indexesthat express the total degree of CRM activities at the threestages of customer life cycles. Thus, in our contentspecification, we sought to capture major facets of evalua-tion and management activities along customercompanyrelationships.

Indicator specification. Critical for the design of validindexes with formative indicators is the choice of items,because the indicators must capture the entire scope of thelatent construct as described previously. On the basis of anextensive review of relevant articles in marketing journals,the business press, and exploratory interviews with man-agers who are responsible for CRM systems, we identified42 items that were evaluated by participants of pretest inter-views as capturing all major subprocesses in the implemen-tation of CRM at the customer-facing level. These indica-tors are listed in the Appendix. All items were measured ona seven-point Likert scale.

1Two items were stated as strongly disagreestrongly agree seven-point Likert scale formats (CRM is a central aspect of our business strat-egy and CRM has become a top management issue in our SBU),whereas the other two items were measured on seven-point semantic scaleformats (With regard to your SBU, to what extent do each of the follow-ing activities represent a strength or weakness for you? The institutional-ization of a CRM philosophy and Getting top management commitmentto CRM), where 1 = major weakness, 4 = neither strength nor weak-ness, and 7 = major strength.

Indicator collinearity. Because formative measurementmodels are based on linear equation systems, substantialcollinearity among indicators would affect the stability ofindicator coefficients. In our example, none of the 42 indi-cators revealed serious multicollinearity problems.

External validity. The very nature of formative measure-ment renders traditional assessments of convergent validityand individual item reliability irrelevant (Hulland 1999, p.201). However, this does not enable researchers to link setsof items to constructs arbitrarily. Aside from strong theoreti-cal foundations, researchers must ensure that all indicatorsthat form a construct are included. To test for external valid-ity, we follow the suggestions of Diamantopoulos and Win-klhofer (2001) to estimate a multiple indicators and multiplecauses model with our aggregate indexes INITIATE, MAIN-TAIN, and TERMINATE, as well as their respective subdi-mensions and formative indicators. We used four variablesthat capture the commitment of top management to imple-ment CRM as reflective indicators of the implementation ofCRM processes.1 The loadings of all four items were highlysignificant, with loadings of .861, .839, .729, and .802.

As we described previously, the CRM process is concep-tualized as a second-order factor measurement model thatcan be approximated with various procedures. One of theeasiest procedures to implement is the hierarchical compo-nent model suggested by Wold (1980). In essence, a second-order factor is directly measured by observed variables forall the first-order factors. Partial least squares (PLS) isappropriate for estimating our measurement model becauseit provides a means for directly estimating component scores(i.e., the three dimensions of relationship initiation, mainte-nance, and termination). Because the latent variable scoresare determinate, PLS can be used to model formative indica-tors, as is the case here. The determinate nature of the PLSapproach avoids parameter identification problems that canoccur under covariance-based analysis (Bollen 1989).

Nomological validity. Given that the formation of imple-mentation of CRM processes as a new formative constructis a key objective of our study, we included 11 additionalitems in our survey. The items were measured on seven-point semantic scale formats (With regard to your SBU, towhat extent do each of the following activities represent astrength or weakness for you?), where 1 = major weak-ness, 4 = neither strength nor weakness, and 7 = majorstrength. To check the nomological validity of our ninesubdimensions and the three higher-level indexes, we esti-mated the bivariate correlations between the subdimensionsor indexes and the respective independent weakness/strength indicators.

Our formative index for acquisition management activi-ties shows correlations of .36 and .34 with the independentstrength/weakness items acquiring high value customersand implementing systematic customer acquisition, and

Customer Relationship Management Process 299

2Details of the measurement model, particularly a description of the datastructure, correlations, and PLS coefficients, are available from the authorson request.

our regain management index reveals a correlation of .35with the statement regaining high value customers as astrength of the SBU. The measurement at the maintainingstage index is significantly correlated with the item under-standing and determining the value of a customer ( =.55). Similar strong associations are observable between ourretention management index, retaining high value cus-tomers (.38) and building long-term relationships with ourvalued customers (.32). Our management of up-selling/cross-selling index reveals even stronger correlations withthe items implementing procedures for up-selling (.50)and implementing procedures for cross-selling (.51). Cor-relations of .36 (management of word-of-mouth) and .47(managing customer referrals) emphasize that the cus-tomer referral management index measures the degree ofactivities related to customer referrals. Because theweakness/strength statement discontinuing relationshipswith low-value customers is also significantly correlated toactivities to demarket customers index (.44), we concludethat all our indexes represent valid measures of the respec-tive constructs.

For the regression analysis, we constructed the relation-ship initiation, maintenance, and termination indexes byweighted multiplication of the individual indicators with thestandardized PLS weights, similar to the American cus-tomer satisfaction index (Fornell and Johnson 1996).2

Model Specification and EstimationThe complete model specification is given in Equation 1.

Variables are grouped into main effects (s), interactioneffects (s), and control variables (s). The control variablesin our system of equations are dummy variables for industryeffects.(1) Economic performance =

+ 1 relationship initiation + 2 relationship maintenance + 3 relationship termination + 4 CRM-compatible organiza-

tional alignment + 5 CRM technology + 1 (CRM-compatible organiza-

tional alignment relationship initiation)

+ 2 (CRM-compatible organiza-tional alignment relationshipmaintenance)

+ 3 (CRM-compatible organiza-tional alignment relationship termination)

+ 4 (CRM technology relation-ship initiation)

+ 5 (CRM technology relation-ship maintenance)

+ 6 (CRM technology relation-ship termination)

+ 1 Industry 2 + 2 Industry 3 + 3 Industry 4 + 1,

where

Economic performance (perceptual) = formative, multi-itemmeasure (adapted from the work of Desphand, Farley, andWebster [1993] and Kohli and Jaworski [1990]) with fourindicators,

Economic performance (objective) = net income in year x/totalassets in year x (ROA),

Relationship initiation, relationship maintenance, relationshiptermination, CRM-compatible organizational alignment, andCRM technology = formative multi-item measures,

Industry 2 = financial services,Industry 3 = power utilities, andIndustry 4 = hospitality.

All multi-item measures are given in the Appendix. Table 1lists the summary statistics for the measurement scales.

On the basis of our previous discussion, we estimatethree different models:

Model 1: Economic performance (perceptual)Sample 1= f(covariates)Sample 1,Model 2: Economic performance (objective) = f(covariates)Sample 1, andModel 3: Economic performance (perceptual)Sample 2= f(covariates)Sample 1.

Given our data structure, this configuration maximizes thedegrees of freedom for each estimation and simultaneouslyaddresses the issue of common-method bias. We mean-centered the variables for the analysis.

RESULTSThe results of the estimation are summarized in Table 2.

The effective sample size for the estimation with perceptualperformance (Model 1) is 211 observations; for objectiveperformance (Model 2), the effective sample size is 98observations. Both estimations fit the data well; the R2 is.24 for perceptual performance and .49 for objective per-formance. Thus, our model helps highlight some factorsthat are associated with more successful CRM processimplementations.

We report one-tailed significance levels. This is appropri-ate because we exclusively test directional hypotheses.Because the hypothesized effects are equal for both per-formance measures (perceptual and objective), we discussthe results together.

Relationship Stages and Economic PerformanceWe hypothesized that the degree of CRM process imple-

mentation is positively associated with economic perform-ance (H1) at the three stages of initiation, maintenance, andtermination. For our perceptual performance measure, sup-port is strongest for maintenance (2 = .71, p < .01). For ini-tiation, support is marginal (1 = .47, p < .05), and it is notsignificant for termination. In the case of objective perform-ance, all three coefficients are marginally significant (1 =9.04, p < .1; 2 = 8.16, p < .05; 3 = 6.97, p < .05). Thus, itseems that the more firms engage in implementing CRMprocesses, especially at the initiation and maintenancestage, the better they perform.

300JO

URNALO

F MARKETING

RESEARCH, AUGUST

2004

Table 1SUMMARY STATISTICS FOR THE MEASUREMENT SCALES

Sample 1 Sample 2

Number Standard Standard Variable of Items Frequency Mean Deviation Minimum Maximum Mean Deviation Minimum MaximumPerformance (perceptual) 4 18.4 4.4 6.0 27.0 19.0 4.0 8.0 26.0Performance (objective) 1 .009 .06 .19 .269 Relationship initiation 15 5.1 1.8 1.5 9.9 4.9 1.6 1.6 8.4Relationship maintenance 20 7.1 1.8 2.4 11.2 6.7 1.6 3.2 10.5Relationship termination 4 4.1 2.0 1.3 9.3 3.6 1.6 1.0 6.9CRM-compatible organizational alignment 4 13.7 4.1 4.0 21.0 12.6 3.9 4.0 20.0CRM technology 4 16.4 5.3 4.0 28.0 15.6 6.2 4.0 28.0Industry 1 (online retailers) 1 64 Industry 2 (financial services) 1 78 Industry 3 (power utilities) 1 28 Industry 4 (hospitality) 1 41

Table 2RESULTS OF MODELS 13

Performance (Perceptual): Performance (Objective): Performance (Perceptual):Model 1 Model 2 Model 3

Dependent Variable Description Coefficient Estimate Standard Error Estimatea Standard Error Estimate Standard ErrorIntercept 18.4***0 .55 N.S. 17.3*** .74

Main Effects Relationship initiation 1 .47*** .26 9.04** 5.59 .93*** .39Relationship maintenance 2 .71*** .23 8.16** 4.80 .60** .33Relationship termination 3 N.S. 6.97** 3.44 N.S.

CRM-compatible organizational alignment 4 N.S. N.S. N.S.CRM technology 5 .16*** .07 N.S. .17* .11

Interactions CRM organizational alignment relationship initiation 1 .17*** .08 2.45** 1.77 N.S.CRM organizational alignment relationship maintenance 2 N.S. N.S. N.S.CRM organizational alignment relationship termination 3 .18*** .05 N.S. N.S.

CRM technology relationship initiation 4 .11*** .06 N.S. N.S.CRM technology relationship maintenance 5 N.S. N.S. N.S.CRM technology relationship termination 6 .09*** .04 N.S. N.S.

Control Variables Industry 2 (financial services) 1 N.S. N.S. 1.79 .92Industry 3 (power utilities) 2 N.S. N.S. N.S.

Industry 4 (hospitality) 3 1.53*** .80 33.92** 20.72 3.23*** 1.11N 211.00*** 98** 101.R2 00.24*** .49** .29

*p .1.**p .05.***p .01.aDependent variable has been rescaled (103).Notes: We report one-tailed significance levels. N.S. = not significant.

Customer Relationship Management Process 301

3We also tested for differences in the effects of interest (initiation, main-tenance, and termination) for the various industries for Model 1. None ofthe interactions was significant. We also explored the possibility of meandifferences between the different countries. Taking Switzerland as a basecase, we found a significant, positive effect for the cross-validation (Model3) for Austria and Germany. For the other two equations (Models 1 and 2),the country effects were not significant. When we included the countryeffects, none of the other parameter signs or significances changed.

Moderating Effects of CRM-Compatible OrganizationalAlignment

We hypothesized that a CRM-compatible organizationalalignment has a positive, moderating effect on the CRMprocesseseconomic performance link at each stage ofCRM (H2). For the perceptual performance measure, H2was marginally supported for the initiation stage (1 = .17,p < .05) and fully supported for the termination stage (3 =.18, p < .01). H2 was not supported at the maintenancestage, but the association at least was in the hypothesizeddirection (positive). For objective performance, the moder-ating effect was marginally significant for initiation (1 =2.45, p < .05), but not for the other two stages.Moderating Effects of CRM Technology

We hypothesized that CRM technology has a positive,moderating effect on the CRM processeseconomic per-formance link at each stage of the relationship (H3). Forperceptual performance, H3 was supported only in the caseof termination (6 = .09, p < .05). Notably, for the initiationstage, the moderating effect was negative (4 = .11, p