Firms use privacy-sensitive data to make targeting decisions, which can inadvertently reveal the underlying information driving those decisions—a risk the authors term targeting privacy risk . The authors use differential privacy to quantify and control this risk. Although firms increasingly adopt differential privacy, policymakers offer limited guidance on its implementation in a targeting framework. The crucial question, therefore, becomes: where should firms implement differential privacy? The authors find that profitability critically depends on where differential privacy is implemented within a typical targeting workflow. This insight stems from two novel targeting strategies: one implements differential privacy during model training, the other at the targeting-decision stage. Using a large-scale field experiment involving 747,975 customers and extensive simulations, the authors show that both strategies remain profitable under strong privacy protection. Notably, the decision-stage strategy yields, on average, fivefold higher profits than alternative implementations. To generalize this finding, the authors derive the expected profit for a given privacy risk level and a privacy elasticity of targeting profits. Collectively, the results offer guidelines for firms and policymakers to ensure privacy protection and profitability.
Ponte et al. (Mon,) studied this question.