How can algorithms help in segmenting users and customers? A systematic review and research agenda for algorithmic customer segmentation

Abstract

What algorithm to choose for customer segmentation? Should you use one algorithm or many? How many customer segments should you create? How to evaluate the results? In this research, we carry out a systematic literature review to address such central questions in customer segmentation research and practice. The results from extracting information from 172 relevant articles show that algorithmic customer segmentation is the predominant approach for customer segmentation. We found researchers employing 46 different algorithms and 14 different evaluation metrics. For the algorithms, K-means clustering is the most employed. For the metrics, separation-focused metrics are slightly more prevalent than statistics-focused metrics. However, extant studies rarely use domain experts in evaluating the outcomes. Out of the 169 studies that provided details about hyperparameters, more than four out of five used segment size as their only hyperparameter. Typically, studies generate four segments, although the maximum number rarely exceeds twenty, and in most cases, is less than ten. Based on these findings, we propose seven key goals and three practical implications to enhance customer segmentation research and application.