AI-Enhanced Marketing Mix Modeling: Integrating ML, XAI, and LLMs for Greater Accuracy, Interpretability, and Actionability

This paper proposes an AI-enhanced Marketing Mix Modeling (MMM) framework that integrates machine learning (ML), explainable AI (XAI), and large language models (LLMs) to evaluate marketing effectiveness with improved predictive accuracy, interpretability, and practical applicability. Moving beyond traditional MMM approaches, the framework employs the XGBoost algorithm to capture nonlinear relationships between multichannel marketing investments and business outcomes. SHAP analysis further enhances model interpretability through feature-importance rankings, beeswarm visualizations, and dependence plots that quantify each channel’s marginal contribution. In addition, a Claude-based LLM module translates complex model outputs into natural-language performance insights and budget allocation recommendations, lowering technical barriers and improving decision-making actionability. Experimental results demonstrate that, compared with an OLS baseline, the XGBoost model significantly improves predictive performance, increasing R² from 0.8572 to 0.9123 while reducing RMSE and MAE by 21.7% and 22.8%, respectively. The integrated SHAP and LLM components further elucidate channel impacts and suggest budget optimization strategies, yielding an estimated 18–25% improvement in marketing ROI. Overall, the proposed AI-enhanced MMM framework democratizes advanced marketing analytics by providing businesses with an accessible, automated solution that guides strategic marketing decision-making, enables efficient resource allocation, and improves marketing returns.