Evaluating Intervention Targeting with Causal Feature Selection in Retail Loyalty Programs

Abstract

Retail loyalty programs generate vast amounts of consumer data, presenting significant opportunities for targeted marketing interventions. However, traditional predictive models often conflate correlation with causation, leading to suboptimal allocation of promotional resources. This paper investigates the efficacy of causal feature selection in improving intervention targeting within a large-scale retail environment. Through a randomized field experiment involving hundreds of thousands of active loyalty program members, we evaluate the impact of isolating variables that genuinely drive behavioral change in response to promotional stimuli, as opposed to variables that merely predict baseline purchasing behavior. The study utilizes advanced uplift modeling techniques adapted for high-dimensional data, focusing on the identification and validation of causal features that differentiate treatment responders from non-responders. Our findings demonstrate that interventions targeted via causal feature selection significantly outperform those relying on standard predictive targeting or blanket promotions, yielding substantial improvements in incremental revenue and campaign return on investment. Furthermore, the results highlight the critical importance of distinguishing between inherent customer loyalty and treatment-induced loyalty. This research contributes to the growing body of literature on causal inference in marketing analytics, providing actionable frameworks for practitioners seeking to optimize customer relationship management strategies in resource-constrained environments.