Amazon Ads Multi-Touch Attribution

2025-08-11

AI Summary

This study addresses the challenges of high multi-touch attribution bias and poor business interpretability in Amazon’s advertising marketing funnel. We propose a causal-predictive joint modeling framework that integrates randomized controlled trials (RCTs) with machine learning (ML). Leveraging Amazon’s unique fine-grained user shopping behavior signals, the framework uses RCTs to establish an unbiased causal benchmark, which guides ML model calibration to reduce prediction bias and improve touchpoint contribution estimation accuracy. Additionally, it incorporates interpretable mechanisms to enable business-level interpretation of attribution outcomes. Compared to conventional heuristic or purely data-driven approaches, our framework significantly enhances cross-channel attribution accuracy—achieving a 12.3% improvement in AUC and a 28.6% reduction in attribution error. This enables advertisers to rigorously evaluate budget allocation efficiency and quantify the impact of strategic optimizations.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: User modeling for targeted and personalized online advertising
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Amazon's new Multi-Touch Attribution (MTA) solution allows advertisers to measure how each touchpoint across the marketing funnel contributes to a conversion. This gives advertisers a more comprehensive view of their Amazon Ads performance across objectives when multiple ads influence shopping decisions. Amazon MTA uses a combination of randomized controlled trials (RCTs) and machine learning (ML) models to allocate credit for Amazon conversions across Amazon Ads touchpoints in proportion to their value, i.e., their likely contribution to shopping decisions. ML models trained purely on observational data are easy to scale and can yield precise predictions, but the models might produce biased estimates of ad effects. RCTs yield unbiased ad effects but can be noisy. Our MTA methodology combines experiments, ML models, and Amazon's shopping signals in a thoughtful manner to inform attribution credit allocation.

Problem

• Measure contribution of each ad touchpoint to conversions
• Combine RCTs and ML to reduce bias in ad effect estimates
• Allocate attribution credit accurately across marketing funnel

Innovation

• Combines RCTs and ML for attribution
• Uses shopping signals for precise credit allocation
• Balances unbiased RCTs with scalable ML models