What is Incrementality Testing?
TL;DR
An experimental method that subtracts outcomes that 'would have happened without the ad' to measure only the pure incremental effect. It verifies causation via holdout tests and geo-lift, preventing overestimation.
Incrementality Testing: Definition & Explanation
Incrementality testing is a method that subtracts the outcome an ad effort 'would have achieved even without the ad' to measure only the truly added (incremental) effect. The biggest weakness of attribution is 'confusing correlation with causation.' For example, if a customer who already had high purchase intent happens to click a retargeting ad and buys, last-click measurement gives all credit to that ad—but in reality they likely would have bought anyway, so the effect is overestimated.\n\nIncrementality testing verifies this through experiments (causal inference). Representative methods include (1) holdout tests (creating a control group shown no ads—by user or region—and comparing outcomes against the exposed group), (2) geo-lift tests (toggling ads on/off by region and measuring the sales difference), and (3) running PSAs (public service ads) to the control group. This answers the true question directly tied to budget decisions: 'how much would sales drop if I turned off this ad?'\n\nRecent AI attribution tools (Triple Whale, Northbeam, Rockerbox, etc.) and MMM platforms are increasingly integrating incrementality testing as a feature. It plays the role of 'checking the answers' for daily optimization (MTA) and budget allocation (MMM) figures, serving as the last bastion of reliability in measurement during the cookie-restriction era.