The Working Concepts for
Building a Database Retention Model
Bruce Ratner, Ph.D.
Database marketers are often tasked with holding customers in place as mature markets fizzle and new markets overtake existing ones. They use models as a key component in their marketing programs to make progress towards retaining a customer database. For example, in the financial services and telecommunications industries, database marketers use retention models to identify individuals who are likely to renew their credit cards and cellular services, respectively, and then develop campaigns targeted to those individuals intended to excite rather than cancel activity. Logistic regression analysis is the standard method for building a retention model to explain and predict a binary target variable - defined by renewers and non-renewers - based on static variables (e.g., age and gender of customer) and time-series variables (e.g., January through December balances due). Specifically, the model provides an individual's likelihood of renewal in a prescribed time period in the future, e.g., one month prior to renewal of the product or service. The time-series data must be in correct relative position with respect to the prescribed time period before the data analyst begins model building.
This article discusses the working concepts for building a retention model by 1) reviewing the basics of logistic regression analysis, 2) presenting an explicit definition of the retention model, 3) offering the retention-cycle component for potential increase in accuracy and stability of the retention model, 4) comparing and contrasting retention and attrition models, and 5) providing the SAS-code program for aligning times-series data, which should be a welcomed entry in the tool kit of data analysts who frequently work on the retention problem.