Start Date
10-12-2017 12:00 AM
Description
The measurement of the effectiveness of a marketing campaign is a challenging task. Whereas established approaches do not consider causality, uplift models take into account which customers display some behavior because of the marketing action and model this target as differential response. The paper categorizes existing approaches toward uplift modeling collected from different fields into a conceptual taxonomy to establish the state-of-the-art and proposes a novel approach named revenue uplift modeling. Contrary to existing approaches, which model incremental response, revenue uplift models predict the incremental revenue with the goal to maximize the gain per marketing incentive for heterogeneous customers. An experiment based on a large real-world dataset of e-commerce shops across several industries provides a benchmark on the choice of machine learning methods to implement the identified uplift modeling approaches and demonstrates the effectiveness of the revenue uplift model in a real-world e-commerce environment.
Recommended Citation
Gubela, Robin Marco; Lessmann, Stefan; Haupt, Johannes; Baumann, Annika; Radmer, Tillmann; and Gebert, Fabian, "Revenue Uplift Modeling" (2017). ICIS 2017 Proceedings. 24.
https://aisel.aisnet.org/icis2017/DataScience/Presentations/24
Revenue Uplift Modeling
The measurement of the effectiveness of a marketing campaign is a challenging task. Whereas established approaches do not consider causality, uplift models take into account which customers display some behavior because of the marketing action and model this target as differential response. The paper categorizes existing approaches toward uplift modeling collected from different fields into a conceptual taxonomy to establish the state-of-the-art and proposes a novel approach named revenue uplift modeling. Contrary to existing approaches, which model incremental response, revenue uplift models predict the incremental revenue with the goal to maximize the gain per marketing incentive for heterogeneous customers. An experiment based on a large real-world dataset of e-commerce shops across several industries provides a benchmark on the choice of machine learning methods to implement the identified uplift modeling approaches and demonstrates the effectiveness of the revenue uplift model in a real-world e-commerce environment.