Something Approaching Science?

Abstract

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 whether 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.