How to Separate the Wheat from the Chaff: Improved Variable Selection for New Customer Acquisition

Abstract

Steady customer losses create pressure for firms to acquire new accounts, a task that is both costly and risky. Lacking knowledge about their prospects, firms often use a large array of predictors obtained from list vendors, which in turn rapidly creates massive high-dimensional data problems. Selecting the appropriate variables and their functional relationships with acquisition probabilities is therefore a substantial challenge. This study proposes a Bayesian variable selection approach to optimally select targets for new customer acquisition. Data from an insurance company reveal that this approach outperforms nonselection methods and selection methods based on expert judgment as well as benchmarks based on principal component analysis and bootstrap aggregation of classification trees. Notably, the optimal results show that the Bayesian approach selects panel-based metrics as predictors, detects several nonlinear relationships, selects very large numbers of addresses, and generates profits. In a series of post hoc analyses, the authors consider prospects’ response behaviors and cross-selling potential and systematically vary the number of predictors and the estimated profit per response. The results reveal that more predictors and higher response rates do not necessarily lead to higher profits.