An absorbing Markov chain for customer segment transition

Abstract

Abstract Firms use RFM analysis to identify valuable customers, particularly loyal customers. In practice and in prior studies, RFM-based scores are often treated as static indicators. However, when observed periodically, these scores can be analyzed dynamically, allowing firms to track changes in customer loyalty and design retention strategies. Although prior studies have modeled customer segments as states in Markov models, they have mainly focused on predicting future segment distributions and have not sufficiently examined how long customers remain in each state before reaching churn. To address this gap, this study applies an absorbing Markov chain framework to dynamic RFM-based customer segments. We define High-Active, Active, Needs-Attention, Low-Active, and Inactive as transient states and Churn as an absorbing state. Using POS transaction data from a retailer operating both physical stores and an online store, we compute the expected steps to absorption and the expected number of visits to transient states before absorption. We also compare online-only, offline-only, and multichannel customers. The results show that the expected steps to absorption and state-visit patterns before absorption into churn vary across channel-based customer groups. These indicators provide decision-support information for CRM monitoring, retention intensity, and channel-specific marketing interventions.