Incremental revenue (the lift over a holdout) divided by incremental spend; the causal, double-count-free version of ROAS.
Incremental ROAS, iROAS, is incremental revenue divided by incremental spend. The numerator is not platform-reported revenue; it is the measured lift over a holdout (see incrementality). This is the version of ROAS that survives scrutiny, because it counts only the sales the spend actually caused.
The gap between reported ROAS and iROAS is large and systematic. Eight X, summarising Haus case studies, finds platform-reported ROAS overstates measured iROAS by roughly 1.5x to 3x (eightx.co). An operator optimising to platform ROAS is optimising to a number inflated by 1.5 to 3 times.
Worked example (Demonstrative, range-anchored to the Haus studies): a channel reporting a 4.0x platform ROAS, where platform attribution overstates lift by about 2x, implies a measured iROAS near 2.0x. Note that a 2x overstatement sits toward the high end of the published 1.5x to 3x range, so the conclusion that follows is the more pessimistic case, not the typical one: if that channel's products carry a 50% contribution margin (2.0x break-even ROAS), the channel is barely at break-even on a causal, profit-true basis, not the comfortable 4.0x the dashboard shows. At the lower end of the range (around 1.5x overstatement) the same channel would clear its break-even with room to spare.
Blufire pairs iROAS with contribution margin so the causal revenue figure is converted to causal profit: incremental revenue at your real CM% against incremental spend tells you whether the next dollar builds margin, not just revenue.