Precision Coupon Targeting with Dynamic Customer Triage

Coupon is a powerful tool for promotional marketing and customer targeting. Improving customer conversion effects by optimally allocating coupons to the right customers, however, is a nontrivial task. In particular, offering coupons to all customers may not be cost effective while offering coupons to randomly selected customers may not lead to optimal effects. In this paper, we address challenges in precision coupon targeting with a customer triage framework which would ration coupons precisely to customers who will be the most promising commercial conversions. Specifically, first we use random user-experiments to quantify the coupon effects of our interest. Then, we model the observation from user-experiments to predict the expected coupon effects on all customers. With the predicted coupon effects, we develop a joint decision strategy for customer triage to target customers with the most influential coupons under budget constraints. The joint decision strategy can dynamically balance exploitation and exploration in the decision process spanning multiple decision periods. Implemented on a real-world online takeout service system, our results show significant improvements in comparison with alternative approaches.