Integrated search glossary

Incrementality testing

Incrementality testing is a controlled experiment that measures how many conversions advertising actually caused, by comparing a group exposed to the ads with a similar group that was not. The difference between the two groups is the incremental lift: the part of the results the ads can genuinely claim.

Reviewed by Gareth Hoyle, Managing Director
Last reviewed 31 October 2026

Why platform reports are not enough

Attribution credits ads with conversions that follow a click or view, including some that would have happened anyway. A study of eBay's search advertising by Thomas Blake, Chris Nosko and Steven Tadelis, published by NBER in 2014 and in Econometrica in 2015, found that brand keyword ads had no measurable short-term benefit. For non-brand keywords, frequent eBay users whose buying was not influenced by ads accounted for most of the spend, so average returns were negative. That is one very well-known brand, but it shows how far attributed and incremental results can differ.

How are incrementality tests run?

Every method splits the audience into a treatment group that sees the ads and a control group that does not.

User holdouts: Google's Conversion Lift based on users splits people into exposed and control groups
Geo experiments: regions are randomly assigned to treatment or control, a method Google researchers described in 2011
Conversion Lift based on geography: Google's geo version, which supports offline conversion data
Pause tests: ads are switched off for a period or region and the change in total conversions is measured

Reading the results

Divide spend by incremental conversions, not reported conversions, to get the true cost per result. Change one thing at a time and run the test long enough to cover a full buying cycle. Google says Conversion Lift is not available for all accounts.

Example

A retailer reports a 900% ROAS on brand search. It switches brand ads off in half of its UK sales regions for four weeks and total online sales in those regions fall by 3% compared with the rest. The ads were adding some sales, but far fewer than the reported ROAS implied.