Optimizing Marketing Campaign Targeting Using Uncertainty-Based Predictive Modelling

Abstract

Over the past decades, decision makers have increasingly relied on analytical methods in order to increase revenues and reduce costs. This is also the case for marketing, where predictive models have primarily been used to identify potential customers and measure the effects of marketing campaigns. In this paper, we introduce a methodology to optimally distribute a constrained marketing budget over a group of targets for which historical behavioral information is available, though where it is deemed infeasible to set up a trial campaign involving a control and test group as is the typical approach described by "net lift" modelling. Instead, so called "swing clients" are identified based on a notion of uncertainty following from any classification technique which can be trained over the historical data, which makes that our approach is easily applicable in most marketing environments. We report on the feasible of our approach by presenting a case study which was performed at the largest airport in Belgium.