Maximizing the Lift in Database Marketing

Database marketers use predictive models to identify the top 5% - 30% of their most-likely-to-respond customers for a marketing campaign, as contacting all customers is either too expensive, rr not practical. The standard statistical models used for finding the “top X%” customers are: ordinary regression for a continuous response variable, and logistic regression for a binary response variable. However, these models do not explicitly maximize the lift – the measure indicating the at-hand model’s percentage gain of identifying the top X% customers over the chance model (i.e., the random selection of X% of the customers from the marketer’s database). In effect, the statistical models are not optimal in maximizing lift. The statistical models work well in practice, but a model that explicitly maximizes lift should outperform them. The purpose of this article is to discuss and illustrate the predictive power the GenIQ Model©, as it explicitly maximizes lift, for any top X% depth-of-file.