Customer Targeting
(A2025-125730)

AUTHORS

Patrick Bachmann, ETH Zurich; Zhuoyu Shi, Rotterdam School of Management; GILIAN PONTE, Erasmus University; Ta-Wei Huang, Harvard Business School

ABSTRACT

The evolution of data-driven marketing has created opportunities to optimize customer targeting but also raised critical challenges. This Special Session aims to address challenges such as optimizing long-term customer value, ensuring fairness in campaign design, and balancing profitability with privacy. By integrating complementary perspectives, the session presents a cohesive framework for advancing socially responsible and customer-centric targeting. 1. Targeting Campaigns for Long-Term Incremental Customer Value (Bachmann, Stromberg, Reutterer, Meierer) This study introduces a probabilistic framework to optimize customer targeting by jointly estimating customer attrition, repurchase propensity, and spending behavior. Long-term field experiments with a European retailer demonstrate up to 17% improvement in targeting performance over random selection and 8% over current practices. This approach focuses on scalability and long-term value, offering a more effective allocation of marketing resources. 2. Fair Active Learning for Targeted Marketing Campaigns (Shi, Lemmens) Fair Active Learning (FAL) improves fairness in data acquisition for causal machine learning models predicting heterogeneous treatment effects. Validated through simulations and a field experiment, FAL enhances statistical separation in protected groups without harming unprotected ones. When combined with bias-eliminating adapted trees (BEAT), it achieves fairness across two dimensions, offering a comprehensive solution for fair targeting. 3. Differentially Private Targeting Strategies (Ponte, Boot, Reutterer, Wieringa) To address privacy concerns, this research introduces two strategies—DP (Differential Privacy)-CATE and DP-policy—that balance profitability and privacy risk. Using a randomized controlled trial with 747,975 customers, the authors show these strategies maintain profitability while adhering to strict privacy standards. Privacy risk is quantified in terms of profit elasticity, guiding firms in navigating trade-offs between personalization and privacy. 4. Debiasing Treatment Effect Estimation for Privacy-Protected Data (Huang, Ascarza) This study addresses biases introduced by Local Differential Privacy (LDP) in Conditional Average Treatment Effect (CATE) estimation. A Model Auditing and Calibration approach is proposed to iteratively correct prediction errors while preserving privacy. Simulations and two real-world marketing experiments validate its superior accuracy and targeting performance, enabling firms to deliver effective interventions within privacy regulations. Together, these four studies highlight innovative solutions that advance ethical, efficient, and scalable targeting practices. By addressing interrelated challenges, the session equips researchers and practitioners with actionable tools to design equitable and effective marketing campaigns.