A geo experiment measures the effect of advertising by changing it in some geographic areas and not in others, then comparing what happens. The areas that get the change form the test group, the rest form the control, and the difference between them estimates what the advertising genuinely added. It is one of the most practical ways to measure incrementality.
How a geo experiment works
It starts with a question that leads to a decision, such as “would we lose sales if we stopped bidding on our own brand name?” or “is Performance Max adding sales or just claiming them?”. You then need two things: the ability to switch advertising on, off or up by area using location targeting, and a way to measure the outcome by area, such as sales by delivery postcode, enquiries by branch or store takings.
Areas are split into test and control. With few areas, they are matched so the two groups have tracked each other closely in the past. With many smaller areas, they can be assigned at random. A pre-period confirms the groups move together before anything changes. During the test you change spend only in the test areas, and run long enough to cover your conversion lag. Analysis then compares what the test areas did with what the control areas predict they would have done.
Free tools exist for the analysis, such as Meta’s open-source GeoLift package. Google and Meta also run lift studies for some advertisers, and eligibility varies by account and spend.
Why it matters
Platform-reported conversions often credit ads for sales that would have happened anyway, particularly on brand search and remarketing. A geo experiment does not depend on tracking individual users, so it still works when many UK visitors decline cookies under PECR. It also captures sales that happen offline, in shops or over the phone.
The UK brings its own design questions. Tests can be split by region, TV area or postcode area. London usually needs treating as a region of its own, because it behaves differently from the rest of the country and its commuters live across the Home Counties, so an ad seen at work in the City may lead to a purchase at home in Surrey. Leave buffer areas between test and control where that spillover is likely.
Common mistakes
- Running the test on too few areas or for too short a time to separate a real effect from normal noise.
- Ignoring spillover between neighbouring areas.
- Picking a test region where something else is happening, such as a new branch opening or a local promotion.
- Changing other marketing, prices or stock in the middle of the test.
- Measuring the result with the ad platform’s own conversion figures instead of real sales.
- Not agreeing beforehand what result would change the budget.
How to act on it
Write down the question, the decision it informs and the result that would trigger it. Check that you can report sales or enquiries by area reliably, ideally going back at least a year. If your volumes are small, accept that a geo test may only detect large effects, and consider a simpler campaign experiment or a longer on-and-off test instead.
Keep everything else stable while the test runs and read the result against real sales, not the platform’s figures. I design and analyse these tests within ongoing PPC management for clients who spend enough to make them worthwhile, and say so plainly when the numbers will not support one.