Incrementality

Attribution divides credit among the touchpoints a converter happened to encounter. Incrementality asks a harder question: would this person have bought anyway?

The distinction is not academic, because the two routinely disagree in a specific and expensive direction. Branded search is the classic case. It attributes beautifully — people search your name, click the ad, buy — and a large share of those people were going to reach you regardless. The report says the campaign works. Turning it off is how you find out whether it did.

Retargeting has the same shape. You are advertising to people who already visited, already added to cart, already decided. They convert at flattering rates and some meaningful portion needed no advertising at all.

How you actually measure it. You need a group that did not see the ads. Geographic holdouts, where some regions get the campaign and comparable ones do not. Audience holdouts, where a random slice is excluded. Or a clean on-off test with enough time either side to separate the effect from normal variation. Some platforms offer built-in lift studies, which are convenient and are marked by the party selling the advertising — worth knowing when you read the result.

Why nobody does it. It requires deliberately not advertising to people, which feels like burning money, and it produces a number smaller than the one in the platform dashboard. The awkwardness is the point: the gap between attributed and incremental is the actual finding.