What a channel actually adds versus what would have happened if you had spent nothing. Every attribution method tells you what a platform claims it drove; incrementality tests the claim by turning the channel off somewhere (a geo holdout or ghost audience), leaving it on everywhere else, and measuring the difference. The gap between reported ROAS and incremental ROAS is the whole game - branded search inflation commonly runs 5-10x. The mistake is running one test and treating the answer as permanent truth: incrementality drifts as creative fatigues and competitors move, so it is a recurring cost, not a project.
Benchmark. Test from roughly $3M+ revenue - below that, post-purchase surveys plus MER plus nCAC do the same job for a fraction of the effort. Sobering anchor: the median incremental ROAS on branded search is around 0.70x, whatever the platform reports.