A Hierarchical Bayesian Market Mix Model with Causal Inference for Personalized Marketing Optimization

Abstract

This study presents a novel hierarchical Bayesian market mix model that integrates causal inference techniques for personalised marketing optimisation. As traditional market mix models struggle to capture individual-level heterogeneity and causal relationships, we develop a flexible framework that addresses these challenges. Our model incorporates multi-level data structures, combining individual, product, and market-level variables to estimate personalised and aggregate marketing effects. We enhance the model's ability to infer causal relationships between marketing actions and consumer responses by employing a potential outcomes approach and propensity score matching. The empirical application utilizes a comprehensive dataset from a multinational consumer goods company, spanning three years of marketing activities across multiple product categories and countries. Results demonstrate the model's superior performance in predicting consumer behaviour and optimising marketing resource allocation. Integrating latent variable modelling for personalisation captures unobserved heterogeneity in consumer preferences, enabling more targeted marketing strategies. This research contributes to marketing analytics by providing a robust methodology for estimating individualised marketing effects, improving attribution accuracy, and generating actionable insights for personalised marketing optimisation in complex, data-rich environments.