Beyond Last-Click: How We Prove Incremental ROAS in Retail Media

Summary

The retail media landscape is shifting from first-mover advantage to proving measurable business outcomes, with incrementality becoming the key performance metric. According to a 2024 ANA survey, 71% of advertisers now consider incrementality the most important metric for RMN investments. Yet most still base budget decisions on legacy attribution models like ROAS, which may credit purchases that would have happened anyway.

The article highlights the limitations of traditional incrementality testing: it is manual, requires resource-intensive setup, demands large holdout groups that sacrifice revenue, and interferes with machine learning optimization. To address these challenges, Moloco introduces an AI-native ghost bidding methodology, which uses a randomized controlled trial framework during live campaigns. Ghost bidding identifies when an ad would win an auction, then withholds it from a randomized subset of users, creating a clean control group without disrupting normal auction dynamics.

This allows advertisers to isolate the causal effect of ads from seasonality, organic demand, and brand loyalty. The methodology carefully defines conversion populations to ensure apples-to-apples comparisons: only users eligible for ad exposure are included in both groups, and all purchases during the test period are counted—not just attributed conversions. Early results from multiple advertisers show iROAS ranging from 253% to 1,609%, with incremental conversion rates from 4% to 29%.

The article outlines best practices: ensure sufficient sample sizes and test duration for statistical significance, carefully define exposed and control groups, count all relevant conversions, account for external factors like promotions, and establish a regular testing cadence. As more platforms integrate testing capabilities, incrementality measurement will become cheaper and more accessible. Future-proof solutions will rely on real-time session behavior and contextual intelligence rather than cookies or PII, making them privacy-compliant and more reliable.