Incrementality: Proving the Ads Actually Caused the Sale

Attribution answers a narrow question: which touchpoints appeared on the path to a conversion. It does not answer the question that actually matters: would this conversion have happened without the ad?

Branded search is the clearest illustration. Someone who already decided to buy searches your name, clicks the ad above the organic result, and converts. Attribution records a win. In reality the ad intercepted a customer you already had and charged you for the privilege.

Why Platform Reporting Overstates

Every ad platform reports on its own performance using its own attribution model, and each has a structural incentive toward generosity. Summing reported conversions across platforms routinely exceeds total actual conversions — which tells you plainly that some credit is being double-counted.

Automated bidding compounds this: algorithms optimise toward the conversions they are shown, which pushes budget toward audiences that were already going to convert, because those look most efficient.

Geographic Holdout Tests

The most practical method for most advertisers. Split comparable regions into test and control groups, turn spend off in control, and measure the difference in total conversions — not platform-reported conversions.

The design requirements matter: regions must be genuinely comparable, the test must run long enough to cover the full sales cycle, and it must be long enough to separate real effect from seasonal noise. Underpowered tests produce confident wrong answers.

Budget-Split Tests

Where geographic splits are impractical, varying spend levels across matched cells and observing the response curve gives a usable approximation. Less clean than a true holdout, but it reveals the point at which additional spend stops producing additional revenue — often the single most valuable number in the account.

Expect Uncomfortable Results

Incrementality tests routinely show that branded search, retargeting and some audience segments contribute far less than attribution claimed. That is the point of running them. The organisational challenge is that someone's reported performance gets worse, and the finding needs framing as budget efficiency rather than blame.

Test on a Cycle

Incrementality is not a fixed property. It shifts with market conditions, competitor activity, brand strength and creative. A result from eighteen months ago is history, not evidence. Build testing into the calendar rather than treating it as a one-off audit.

The accounts that scale profitably are the ones that know which portion of their reported return is real.