Approaches to Customer Segmentation

Abstract

Customer segmentation has virtually unlimited potential as a tool that can guide firms toward more effective ways to market products and develop new ones. As a conceptual introduction to this topic, we study how an innovative multi-national firm (Migros Turk) has developed an effective set of segmentation strategies. This illustrates how firms can construct novel and inventive approaches that provide great value. A-priori, and custom designed post-hoc methods are among the most important approaches that a firm should consider.

We then review general approaches to customer segmentation, with an emphasis on the most powerful and flexible analytical approaches and statistical models. This begins with a discussion of logistic regression for supervised classification, and general types of cluster analysis, both descriptive and predictive. Predictive clustering methods include cluster regression and CHAID (Chi-squared automatic interaction detection, which is also viewed as a tree classifier). Finally, we consider general latent class models that can handle multiple dependent measures of mixed type. These models can also accommodate samples that are drawn from a pre-defined group structure (e.g., multiple observations per household). To illustrate an application of these models, we study a large data set provided by an international specialty-goods retail chain.

Keywords: Latent class model, clustering, cluster regression, logistic regression, classification, conjoint analysis, random effect, multilevel model, inactive covariate, satisfaction

JEL Classification: C30, M21, M30, M31, M37