A Personalized Causal Inference Framework for Media Effectiveness Using Hierarchical Bayesian Market Mix Models

Abstrak

This study presents a novel framework for personalized causal inference in media effectiveness using Hierarchical Bayesian Market Mix Models (ABM). The proposed approach integrates individual-level data with aggregate market information to estimate personalized media effects while addressing the challenges of data sparsity and high dimensionality. By combining the identity layer and the optimization process in a Bayesian hierarchical model, the model captures heterogeneity across consumers and provides robust predictions of individual causality. Affect different media. The framework is used for e-commerce business data, which includes 500,000 customers across 50 markets in 24 months. The model shows better prediction performance than the integrated business model, with a 30.4% reduction in RMSE. Empirical results reveal significant heterogeneity in media effectiveness across channels and consumer segments. Email marketing emerges as the most effective channel on average, followed by TV advertising, digital display ads, and social media engagements. Sensitivity analyses and robustness checks, including alternative prior specifications and placebo tests, support the validity of the estimated causal effects. The findings provide valuable insights for media planning and marketing strategy, highlighting the importance of tailored budget allocation and campaign design approaches. This research contributes to the growing body of literature on personalized marketing analytics and offers a powerful tool for estimating individualized media effects in complex marketing environments.