Marketing Mix Modeling (MMM) is a key analytical approach used to understand how different marketing levers contribute to sales. In recent years, advances in machine learning have enabled the exploration of alternative methodologies beyond traditional parametric or Bayesian regression-based models. Among the most notable recent developments is the NNN framework (Next-Generation Neural Networks for Marketing Mix Modeling), introduced by Google (Mulc et al., 2025), which leverages Transformer architectures and high-dimensional embeddings to represent both paid and organic channels, moving beyond the conventional scalar input paradigm.

This thesis aims to investigate and critically assess the NNN framework, using it as a foundation for comparison with traditional MMM approaches. The results presented by the authors are promising, but further validation is needed in diverse real-world settings. The objective of the thesis will be to evaluate the effectiveness and applicability of the NNN approach using simulated data and, where available, real marketing datasets. Key areas of focus will include attribution accuracy, out-of-sample generalization, and model interpretability.

Note: This thesis addresses cutting-edge topics at the intersection of marketing modeling and applied artificial intelligence. It is therefore intended for students with solid programming skills and good knowledge of statistical modeling or machine learning, who are motivated to engage in a rigorous, comprehensive, and ambitious research project.