The Value of First Impressions: Leveraging Acquisition Data for Customer Management

Abstract

Managing customers effectively is crucial for firm’s long-term profitability. By understanding differences across customers, firms can tailor their activities towards those customers for whom the intervention will pay off, therefore increasing the value of customers while maximizing the return on the marketing efforts. Targeting effectively ultimately depends in the firm’s ability to precisely estimate differences across customers—a very difficult task when firms attempt to manage recently-acquired customers for whom only the first purchase has been observed. We propose a model that allows marketers to form “first impressions” of customers right after having been acquired. We define a first impression as an inference (based on the observed behaviors at the moment of acquisition) that the firm makes about customer’s traits that are relevant for the firm (e.g., whether the customer will purchase again, how s/he will respond to specific marketing actions).

The main aspect of the model is that it captures latent dimensions that impact both the variety of behaviors collected at acquisition as well as future propensities to buy and to respond to marketing actions. Using probabilistic machine learning, we combine deep exponential families with the demand model, relating behaviors observed in the first purchase with consequent customer behavior. We first demonstrate that such a model is flexible enough to capture a wide range of heterogeneity structures (both linear and non-linear), thus being applicable to a variety of behaviors and contexts. We also demonstrate the model’s ability to handle large amounts of data while overcoming commonly faced challenges such as data redundancy, missing data, and the presence of irrelevant information. We then apply the model to data from a retail context and illustrate how the focal firm could form customers' first impressions by merely using its transactional database. We show that the focal firm would significantly improve the return on their marketing actions if it targeted just-acquired customers based on their first impressions.

Keywords: Customer Management, Targeting, Deep Exponential Families, Probabilistic Machine Learning, Acquisition