A Hidden Markov Model of Customer Relationship Dynamics in Retailing Industry

This paper addresses the issue of modeling and understanding the dynamics of customer relationships. The proposed model facilitates using typical transaction data to evaluate the effectiveness of relationship marketing actions as well as other customer-brand encounters on the dynamics of customer relationships and the subsequent buying behavior. Our approach to modeling relationship dynamics is structurally different from the models in the existing literature. Theories in behavioral relationship marketing suggest that relationships evolve over time as a consequence of encounters between the customer and the company (or organization). Accordingly, in the proposed model, customer-brand encounters, such as exposure to relationship marketing activities and past buying behavior, may have an enduring impact by shifting the customer to a different (unobservable) relationship state. We construct and estimate a hidden Markov model (HMM) to relate the latent relationship states to the observed buying behavior. The HMM enables the marketer to assess the evolution of customer relationships over time. Moreover, since the relationship states are determined, in part, by exposure to marketing actions, it is possible to examine methods by which the firm can alter the customer’s relationship level and consequently affect the long-term buying behavior. To account for unobserved heterogeneity across customers, we specify a random-effect model estimated using a hierarchical Bayes procedure. We calibrate the proposed model using simulated data, as well as using longitudinal alumni gift giving data. This empirical application demonstrates the value of the proposed model in understanding the dynamics of alumni-university relationships and predicting donation behavior. Using the proposed model, we are able to identify three relationship states, probabilistically classify the alumni base into these different states, and estimate the marginal impact of different interactions between the alumni and the university on moving the alumni between these states. The application of the model for marketing program decisions is illustrated using a “what-if” analysis of a reunion attendance marketing campaign. Additionally, using a validation sample, we show that the proposed model improves the ability to predict future donations relative to several benchmark models.