Targeting customers under response-dependent costs

This study provides a formal analysis of the customer targeting problem when cost for marketing action depends on response and proposes framework to estimate decision variables campaign profit optimization. Targeting is profitable if impact associated treatment are higher than its cost. Despite growing literature uplift models identify strongest treatment-responders, no research has investigated optimal costs unknown at time decision. Stochastic ubiquitous in direct retention campaigns because incentives conditioned positive response. makes two contributions literature, which evaluated an e-commerce coupon campaign. First, we formally analyze under response-dependent costs. Profit-optimal requires effect probability treatment. The empirical results demonstrate that consideration substantially increases used combination with average or customer-level effect. Second, propose jointly by combining methods causal inference hurdle mixture model. proposed model achieves competitive while streamlining building. Code available https://github.com/Humboldt-WI/response-dependent-costs.