Incremental ROAS answers a harder question than platform dashboards report: would this purchase have happened without the ad? Reported ROAS credits any conversion that touched a campaign inside the attribution window; iROAS counts only revenue that would not have existed without the ads.
The formula is (treated revenue − control revenue) / treated ad spend, and the split must come from a designed experiment. A simple frame: treated markets produce $500K over the test window, matched holdout markets running no ads produce $420K, and dividing the $80K gap by the $40K spent in the treated markets gives an iROAS of 2.0 — even if the platform pixel reported a 4.0 in those same markets. The numerator is what the ads created, not what they touched.
Three experiment shapes produce the treated/control split: geo-lift (markets randomized into test and holdout), platform conversion lift studies like Meta CLS and Google CLS (users randomized inside the platform), and a randomized holdout audience. In each case the numerator is a different unit from attribution-based ROAS. iROAS commonly lands below reported ROAS, and the gap is typically largest on retargeting and branded-search campaigns, where the platform takes credit for purchases that would have happened anyway. The ratio of iROAS to reported ROAS is sometimes called the incrementality multiplier.
Treat the two as different metrics with different jobs, not a real-versus-fake pair. iROAS is the budget-allocation metric for whether a channel is creating new revenue (and new customers — read it alongside CAC); reported ROAS remains the right signal for in-platform day-to-day optimization. iROAS is also expensive — holdout opportunity cost, statistical power, testing infrastructure — and is read at the channel or campaign level.