Data-Driven Market Segmentation – A Structure-Based Conceptual Framework for Managerial Decision Support

Market segmentation increasingly uses homogeneous groups of consumers determined on the basis of empirical market data as target segments (a posteriori-, data-driven-, post hoc segmentation) rather than splitting individuals according to single, typically socio-demographic or geographic, criteria (a priori-, commen sense segmentation). A vast amount of contributions has been made to improve methodology of identifying or constructing data-based market segments. However, real world data sets often do not contain clearly separated density clusters. Therefore all techniques used in data-based market segmentation can render multiple solutions of similar quality. So far no attempt has been made to construct a framework enabling managers to systematically choose between different segmentation solutions with regard to their practical usefulness. We propose a framework of such kind. Market segmentation is a widely accepted strategic marketing tool in industry. Over time the interest has shifted from a priori market segmentation approaches to a posteriori procedures (Mazanec 2000) where an empirical data set providing disaggregate consumer information is used to split individuals into homogeneous groups rather than using single variables, typically of socio-demographic or geographic nature. The main focus of development in the area of a posteriori segmentation was on increasing methodological sophistication of techniques for identification or construction of such data-driven market segments (Wedel & Kamakura 2002). However, while methodology improved in sophistication, numerous fundamental questions remain unsolved, leaving managers without guidance with regard to the appropriateness of applying grouping techniques to their data. The core research question underlying this lack of guidance is, which data conditions require which kind of treatment to derive managerially useful market segments. As Aldenderfer & Blashfield (1984, p.16) put it a long time ago: ''The key to using cluster analysis is knowing when these groups are ``real'' and not merely imposed on the data by the method.'' From the managerial perspective this means that an increased level of transparency is needed, especially when managers have to make the decision, which one of many possible