Predicted Incrementality by Experimentation (PIE) for Ad Measurement

Measuring the causal impact of digital advertising is challenging: randomized controlled trials (RCTs) provide credible estimates but are costly and difficult to scale. We propose Predicted Incrementality by Experimentation (PIE), which reframes ad measurement as a campaign-level prediction problem. PIE uses a set of RCTs to learn a mapping from campaign features to the causal quantity of interest, then applies that mapping to campaigns not run as RCTs. Our key innovation is the use of post-determined features---campaign-level aggregates such as average outcomes, click-through rates, and attribution counts, computed solely from test-group users after campaign completion. We formalize why these features carry information about treatment effects: they encode variation in organic conversion rates, selection into exposure, and ad responsiveness across campaigns. Critically, they remain available for non-RCT campaigns where control groups do not exist. Using 2,226 Meta ad experiments, PIE achieves an out-of-sample R^2 = 0.88, far surpassing industry-standard 7-day last-click attribution (R^2 = 0.19). In a decision-making framework, PIE disagrees with RCT-based decisions in only 8--12% of campaigns, compared to 12--20% for last-click attribution. PIE enables advertising platforms to scale causal measurement by extrapolating from limited RCTs to large sets of non-experimental campaigns.