Challenges and Methods of Causal Promotions Recommendation in E-commerce

This paper tackles the challenges of causal promotions recommendation in e-commerce, with a focus on scalable, real-world implementations for managing large-scale promotional campaigns. We begin by outlining the unique domain-specific complexities of promotions in dynamic settings such as online travel platforms, including triggered costs, sparse conversions, and multi-product interactions. To address these, we introduce a steering framework that enables systematic, continuous experimentation and measurement. We then present methods and best practices for causal uplift modeling and constrained optimization, highlighting a retrospective estimation approach that leverages transaction-only data. These contributions are validated through real-world deployments at Booking.com, demonstrating how causal machine learning can be operationalized to optimize promotions effectively. Our work bridges rigorous methodology with practical business needs, offering a blueprint for balancing strategic impact and operational constraints in industrial settings.