Net Lift Modeling vs. Propensity Modeling for Skewed Data

• Activity Number: 203
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 11:35 AM to 12:20 PM
• Sponsor: Section on Statistics in Marketing
• Author(s): Zhen Zhang* and Lei Zhang and Kendell Churchwell and Jim Veillette
• Companies: C Spire and Mississippi State Department of Health and C Spire and C Spire
• Keywords: net lift model ; propensity model ; skewed data ; campaign segment ; lift ; incremental sales

Abstract

Propensity modeling has been extensively used in telecommunication companies to optimize marketing outcomes in cross sell and up sell campaigns, retention tactics, recruiting strategies, etc. Uplift modeling, on the other hand, is less familiar territory due to its complexity1. Net lift models are reported as superior in terms of maximizing return on investment by some practitioners2,3, while others cautioned on its trade-offs and limitations4. Little was reported, however, on the comparison of these two techniques with respect to skewed data. Our research shows that for highly skewed data, while the net lift model produced much improved incremental sales rates compared to the traditional propensity model, the propensity model outperformed the net lift model in terms of number of incremental sales, due to its much larger segments.