Modern Marketing Measurement and Personalization: Innovations in AI, Experiments, and Causality
(A2025-126088)

AUTHORS

Nicolas Padilla, London Business School; Ryan Dew, The Wharton School, University of Pennsylvania; Eva Ascarza, Harvard Business School; Oded Netzer, Columbia University

ABSTRACT

This session explores how marketers can navigate the evolving landscape shaped by privacy regulations, AI advancements, and the demand for personalized customer engagement, focusing on innovative methodologies that integrate AI, experimental designs, and advanced models to enhance marketing measurement, personalization, and advertising effectiveness. The four papers in this session collectively contribute to advancing the understanding of these pressing challenges: (1) “Your MMM is Broken: Identification of Nonlinear and Time-Varying Effects in Marketing Mix Models,” Ryan Dew, Nicolas Padilla, and Anya Shchetkina Modern marketing mix models often assume nonlinear or time-varying effects. Using theory, simulations, and real-world applications, this paper demonstrates that these effects are frequently not separately identifiable and may suggest fundamentally different optimal marketing allocations, especially with autocorrelated marketing variables. (2) “AI in Disguise: AI-Generated Ads Outperform Human-Made Ads if They Don’t Look Like AI,” Yannick Exner, Jochen Hartmann, Oded Netzer, and Shunyuan Zhang This study investigates the performance of AI-generated ads, showing they outperform human-made ads in click-through rates when not perceived as AI-created. Key features like large faces and high aesthetics enhance their human-like perception, while intense color saturation signals AI origins. The findings provide actionable insights for optimizing AI-powered content creation and enhancing advertising effectiveness. (3) “Incrementality Representation Learning: Synergizing Past Experiments for Intervention Personalization,” Ta-Wei Huang, Eva Ascarza, and Ayelet Israeli This paper introduces Incrementality Representation Learning (IRL), a framework that leverages past experiments to predict heterogeneous causal effects of marketing interventions. By creating low-dimensional representations of intervention features and customer covariates, IRL generalizes to untested scenarios, improving targeting accuracy and addressing cold-start challenges. Validated across 274 experiments, IRL offers practical tools for tailoring profitable marketing interventions. (4) “Unified Marketing Measurement: How to Fuse Experimental Data with Marketing Mix Data?,” Nicolas Padilla and Ryan Dew We propose a Unified Marketing Measurement framework that fuses experiments with aggregate marketing mix data. Using Gaussian Processes, our model handles time-varying marketing effectiveness and leverages exogenous variation in test to de-bias estimates from marketing mix models.