Profit Driven Business Analytics

CHAPTER 4Uplift Modeling

INTRODUCTION

When discussing customer churn prediction in Chapter 3, we explained that by developing and adopting a customer churn prediction model we can target the customers which are most likely to churn in a retention campaign. As such, the use of a predictive model significantly increases the efficiency and return of a retention campaign by allowing to select true would-be churners and to exclude nonchurners. The reader may have realized that a further improvement may be achieved by selecting customers that are not only likely to churn but as well likely to be retained when targeted in a retention campaign. If we exclude would-be churners from the campaign that have made up their minds and therefore cannot be retained, a further increase in profitability will be achieved.

To this end, we introduce uplift modeling approaches in this chapter, which aim at estimating the net effect of a treatment, such as a marketing campaign, on customer behavior. Uplift models allow users to optimize the selection of customers to include in marketing campaigns as well as a further customization at the individual customer level of the campaign design, for example, in terms of the contacting channel and the characteristics of the incentive that is offered. Such customization may even further increase the effect and return of the campaign.

In the first section of this chapter, we will broadly introduce and motivate the use of uplift modeling as an alternative approach ...