|
Article Excerpt
The customer relationship management (CRM) industry is set to be worth $76.3 billion by 2005 but over 50% of projects will fail to meet benefit objectives. While CRM nirvana is the attainment of profitable one-to-one relationships, current activity is concentrated on segmentation. As technology has moved segmentation from simple classification towards more complex predictive modelling, the use of CRM analytic suites comprising statistical techniques such as decision trees, neural networks and cluster analysis is increasing. It is suggested that the subjective nature of cluster analysis may be overlooked when the technique is integrated with other 'tools' into a data-mining package and, consequently, that inadequately tested cluster analysis solutions may be contributing to CRM dissatisfaction. This paper reports the findings of a study which subjected a data set designed for segmentation purposes to a series of rigorous validity and reliability tests and went as far as to randomise the data to ascertain wheth er current methods could detect 'false' data. The study shows, alarmingly, that under certain conditions random data can 'pass' standard tests and highlights just how meticulously and thoroughly cluster analysis solutions must be tested before they can be safely used in formulating marketing strategy. Practical, theoretical and technical advice is offered for managers working with CRM analytics suites and avenues suggested for future research into improved CRM performance through effective management of the IT/marketing interface.
Introduction
As customer relationship management (CRM) sweeps through global business there is reawakened interest by managers, consultants and academics alike in the classic concept of segmentation. According to a recent PriceWaterhouseCoopers survey (Brown 1999), CRM evolves within a firm in three stages: first, the acquisition of customer data; second, the segmentation of customers into discrete groups; and third, the application of differential relationship management strategies to each identified segment and, in some circumstances, to each individual customer. Their research showed that in 1998, 76% of firms actively engaged in CRM were at stage 2 (that is, segmenting their customer base), 16% were still at stage 1, and only 8% at stage 3. A more recent survey of UK businesses in general found that only 18% of firms claimed to organise their business around customer segments, although 50% intended to do so by 2002 (Economic Intelligence Unit survey cited. in Monasco et al. 2000). Thus segmentation will be an extremel y important strategic marketing concept in the foreseeable future as firms progress from acquisition to true relationship building. Even in the distant future, when having reached stage 3, firms are able to form differential relationships with individual customers, segmentation will remain the most cost-effective method of identifying new target customers.
While CRM vendors' revenue seems set to grow, the picture for users is somewhat less optimistic. In 1998, although 80% of large US firms owned a data warehouse, nearly 50% of these organisations felt their investment had failed to meet expectations (Brown 1999). By 2001, 41% of firms with GRM systems were either experiencing difficulties or thought they were a potential flop (Eckerson & Watson 2001). A number of suggestions have been made to explain this lack of satisfaction with CRM systems. Some industry observers suggest that hype by vendors has led to a gap between expectations and performance (Johnson & Page 2001). Others claim the failure stems from a poor understanding of the HR implications (Muhney 2001).
Aims and objectives
This study aims to contribute towards an understanding of current dissatisfaction with CRM. We consider a range of current thinking on the very nature of market segments, and examine evidence that segments may simply be artefacts of the statistical programmes that generate them. This discussion is placed in a CRM context. Finally, we present a study that investigates the reliability and validity of the technique most widely used for segmentation: cluster analysis. In particular, we determine whether inadequately tested cluster solutions may contribute towards CRM dissatisfaction.
The nature of market segments
As practitioners are enthusiastically seeking out groups of profitable customers whose loyalty they can woo, some academics are beginning to question whether segments are actually stable entities, and more fundamentally whether they really exist at all. If the durability or existence of segments is questionable, then this raises important issues for firms who have invested millions of dollars in CRM technology. As Dibb (2001, p. 208) observes, 'Deep in the heart of segmentation strategy is the notion that customers will allow themselves to be managed. There is a sense that once identified, the customers within the segments should remain available for marketers to target.'
A variety of research questioning the stability of market segments has emerged over the last decade. One strand has its roots in culture theory and social psychology. The primary thrust of this literature has been to explore the notion that market segments are not discovered, enduring entities, but instead are temporary groupings resulting from successions of peer group interactions. Examples include Tepper's (1994) research into use of segmentation cues by the over 50s, which suggests segments are mediated by both private and social labelling processes. Holt reports two studies into the processes by which individuals group and regroup within society, one on baseball spectators (1995), the other an application of Bourdieu's theory of cultural capital to the US public (1998). Other research on this theme includes Cova's (1997) work on neo-tribes, Thompson and Hayko's (1997) study of fashion discourse, and Shouten and McAlexander's (1995) fascinating ethnography of Harley Davidson motorcyclists. There are two m ajor implications of this strand of research for CRM practitioners eager to identify profitable segments. First, understanding the social drivers of segment formation may be as important as delineating different segments. Second, the predictive capability of segmentation will be limited in so far as segment composition will reform as individuals realign their peer group affinity.
Postmodernists question the fundamental existence of segments. Their influence has been evident in the marketing literature since the early 1990s (Firat & Venkatesh 1993; Brown 1995, 1998; van Raaij 1996). Firat and Schulz (1997) describe a postmodern reality characterised by openness and tolerance, hyper-reality, paradoxical juxtapositions, a perpetual present, fragmentation, loss of commitment, the decentring of the subject, reversal of consumption and production, emphasis on form or style and an acceptance of disorder and chaos. If we inhabit a postmodern world, segmentation is a futile concept because given the conditions listed above consumers can no longer be anchored to a single consistent stable way of behaving. Consequently, any grouping of consumers will be the result of chance similarities in individual responses to fragmented consumption experiences. If CRM practitioners were to espouse a postmodern view of reality then investment in identifying enduring profitable segments would seem extremely il l-founded.
Rather less radical is the notion that segments are artefacts of the statistical programmes that create them (Dibb & Stern 1995). Consequently, the reliability and validity of the statistical functions used to generate segments or to build predictive models remains a possible contributory factor to CRM user dissatisfaction. Interestingly, the problematic nature of computer-generated segmentation statistics was flagged up in the Journal of the Market Research Society as early as 1970 (Inglis & Johnson 1970) and was reprinted in JMRS in the 1996 special 50th Anniversary Edition as one of the most important papers of the past half-century (Inglis & Johnson 1996). However, it has doggedly remained an unfashionable area of interest, particularly for practitioners. Instead, a sizeable body of marketing literature argues that the implementation of segmentation is the crucial issue for practitioners (Piercy & Morgan 1993; Jenkins & McDonald 1997; Dibb 1999). For instance, Dibb and Simkin (2001, p. 613) argue that, 'w hereas academic researchers emphasize segmentation techniques, stressing the need for statistically robust segmentation schemes, practitioners are more concerned with identifying segments for which clear marketing programmes can be developed and that will get them closer to their targeted customers.' This paper contends along with Inglis and Johnson (1970, 1996) and Dibb and Stern (1995) that unless segments are statistically robust, it...
|
