The growing use of rewards aggregators in digital platforms has highlighted the need for precise estimation of causal impacts on user engagement, retention, and spending behavior. Traditional A/B testing and uplift modeling approaches often fail to capture heterogeneous treatment effects across diverse user segments, leading to suboptimal incentive allocation. This review paper examines the role of causal uplift modeling with a focus on doubly-robust estimators as a reliable framework for reducing bias and variance in treatment effect estimation. We explore the integration of SQL and Python pipelines for scalable data processing, model training, and real-time inference in production environments. Emphasis is placed on heterogeneous treatment-effect modeling techniques that enable personalized reward optimization by identifying subgroups with differential responsiveness to interventions. Furthermore, the review synthesizes methodological advancements in doubly-robust causal inference, system design considerations for deploying uplift models in large-scale rewards ecosystems, and practical challenges such as data sparsity, confounding, and latency constraints. By bridging causal inference theory with applied pipeline engineering, this study provides a comprehensive perspective on building robust, interpretable, and production-ready solutions for real-time decision-making in rewards aggregation platforms.
Amebleh et al. (Mon,) studied this question.