Proactive Retention Management in Retail: Field Experiment Evidence for Lasting Effects

Abstract

The vast amount of data that is now available allows retailers to make marketing decisions at the customer level and thus to manage customer defection proactively. Although there is an extensive literature that focuses on developing models to predict customer "churn'' in contractual settings, few studies have addressed managing customer defection in the noncontractual context. Using data supplied by a large German grocery retailer, we address this gap in the literature for noncontractual settings as follows. First, we develop a definition of churn that distinguishes between random short-term volatility and true defection. Second, we compare various prediction algorithms in terms of their ability to predict customer churn from data available to the retailer; this analysis enables us to devise an approach for identifying the customers most likely to defect and for targeting them proactively. Third, our approach is tested in a large-scale field experiment involving 400,000 customers for whom we estimate defection probabilities. In particular, potential churners are targeted with a series of marketing campaigns over six months. We find that weekly revenue from this treatment group increases by 3% and that, in comparison with the control group, the number of churners declines by 6%. Lastly, we show that the effects of proactive churn management last for at least two months after the end of the field experiment.

Keywords: proactive churn management, field experiments, machine learning

JEL Classification: M31, C51