Models for uplift are commonly used to isolate the marketing effect of a campaign. For customer churn reduction, uplift models identify customers who are likely to answer positively to a retention activity only if explicitly targeted. They are also used to avoid wasting resources on customers that are very likely to switch companies. In practice, the models' performance is measured with the Qini coefficient. We introduce a Qini-based uplift regression model to analyze a large insurance company's retention marketing campaign. Our approach is based on logistic regression. We show that a Qini-optimized uplift model acts as a regularization in uplift models, yielding interpretable models with few relevant explanatory variables. Our results also show that the parameter estimation based on our Qini-optimized regression significantly improves the Qini prediction performance of uplift models.

This is joint work with Mouloud Belbahri (TD Assurance), Olivier Gandouet (TD Assurance) and Vahid Partovi Nia (École Polytechnique de Montréal).