[1H4-OS-12b] OS-12 (2)

[1H4-OS-12b-02] A new evaluation metric and issue of uplift modeling for coupon marketing

〇Akihiro Shimizu^1,2, Riku Togashi¹ (1. Mercari, inc., 2. JAIST)

Keywords:Machine learning, Causal Inference , Uplift Modeling

What is the correct evaluation of Uplift Modeling in coupon marketing? Recently, Uplift Modeling receives attention for improving the cost-effectiveness of coupon marketing. However, the measures, the Qini Curve and Coefficient, utilized in Uplift Modeling, do not align with the structure of cost-effectiveness in coupon marketing. In this paper, we demonstrate examples that the Qini does not measure the cost-effectiveness correctly. Moreover, we also propose a measures "Sure Things Curve" and "Sure Things Coefficient" that considers cost-effectiveness in coupon marketing.