GuideMeasurement and attribution
Incrementality
Definition
The revenue that would disappear if the marketing stopped. Attribution asks which ad should get credit for a sale; incrementality asks whether the sale needed an ad at all.
How it’s measured
You measure it by switching spend off on purpose. A geo holdout turns a channel off in matched regions and compares them with the rest. An on/off test blacks out the whole channel for several weeks. A conversion lift study, run inside the platform, holds ads back from a control group and counts the difference. All three are experiments, which is why this is the one measurement question with a correct answer: the control group bought, or it did not.
The gap between tested and platform-reported ROAS is widest where channels are credited for customers already on their way in: branded search, because the person typed your brand before seeing the ad, and retargeting, because the person was on your site yesterday.
The catch: the answer costs revenue. A holdout means real regions with ads off for weeks, and the budget case for deliberately losing sales is awkward. Most companies I have sat with discuss incrementality every year and never run one.
What to watch for
Sort the account by reported ROAS. If branded search and retargeting sit at the top, that is the expected signature of channels credited for demand that already existed, and a reason to test them first. When a channel’s budget changes materially, watch total revenue: if the channel’s reported revenue moves and the total stays where it was, the channel is claiming sales it did not cause. A lift result quoted from the platform’s own study still needs a holdout you control before it settles anything. And where no test has been run, the honest reading is that incrementality is unknown.
The question you ask
“If we switched this off for a month, how much revenue would we lose?”