Gain: Retail-Media Incrementality

AI agents measure true incremental lift, expose last-click overstatement, and recommend scale, hold, or cut

Problem statement

Retail-media advertisers cannot trust retailer-reported ROAS: last-click attribution overstates ad impact by 30 to 50%, so budgets get misallocated. Gain reads a geo-experiment, measures the true incremental lift with a confidence interval, exposes the last-click overstatement, and recommends scale, hold, or cut, with both a Growth / Media and a Finance reading. The demo runs on a synthetic geo-experiment.

Run Demo is an analysis of the default sample dataset.

Run Demo (Alternate Dataset) is an analysis of a randomly selected sample dataset from previously seeded synthetic data.

Sample dataset & assumptions

Data: synthetic geo-experiments (region, period, outcome, treatment), 30 regions over 18 periods, ~4 treated vs 26 control, with shared parallel pre-trends and a planted post-period lift

Method: a deterministic rule picks CausalImpact vs synthetic control from the panel shape

Refutation: two DoWhy tests (placebo treatment, random subset) plus a placebo-in-time check

Last-click overstatement: the disclosed 1.3 to 1.5x inflation band (ANA 2025) when no retailer number is supplied

Confidence intervals are reported at the 95% level.