Blinded by science: The managerial consequences of inadequately validated cluster analysis solutions

Abstract

Cluster analysis has been successfully used in market segmentation for several decades. However, alongside evidence for the value of the technique, a number of studies have highlighted the importance of testing the reliability and validity of cluster solutions. Yet, in a time-poor technologically sophisticated age when alluring output falls effortlessly from user-friendly statistical packages, managers may fail to appreciate the rigorous testing required to ensure robust solutions. The authors designed an experiment to investigate whether managers could distinguish between cluster analysis outputs derived from real and random data. Given information on only cluster centroids and demographic profilers, random data devoid of meaningful structure were perceived as equally useful for purposes of market segmentation as real data. If these findings generalise, then managers could be formulating segmentation strategy based on appealing statistics that are at best untested and at worst completely misleading. As cluster analysis is incorporated into the analytics suites of popular CRM systems, marketing managers are becoming increasingly distanced from the raw data. Yet, the consequences of inappropriate use of cluster analysis, and in particular inadequate validation, can be dramatic.