How long should a geo lift test run?

A geo lift test should run long enough to cover at least one full purchase cycle, get past the weeks when the ads are still building, and shrink the noise below the lift you expect. For most multi-location brands that means 6 to 12 weeks with the spend change live, then a few weeks of post-period. A coffee chain can sit at the short end. A dental group, or a gym brand testing in January, sits at the long end.

The real question is “how long until the answer is trustworthy?”, and you can work that out before you spend a dollar. This guide shows the arithmetic, gives typical lengths by type of business, and covers the ways teams cut tests short or let them drag on. If you haven’t planned a geo test before, start with our guide to geo holdout tests and come back here for the timing.

The short answer: 6 to 12 weeks, plus a post-period

Meta’s open-source GeoLift guidance sets a floor of 15 days for daily data or 4–6 weeks for weekly data. It says a test should cover at least one purchase cycle, and it asks for stable history at least four to five times the test length (Source: Meta GeoLift documentation, accessed 2026).

Treat those as floors, not targets. Counts at physical locations are noisier than online conversions, and local media usually takes a week or two to reach full strength. So most location tests need more than the minimum.

What decides how long a geo test should run

Purchase cycle

A burger chain’s regulars come back every week or two. A dental patient may book weeks after the first search. A home-services customer only calls when something breaks. Cover at least one full cycle, ideally two, so someone who saw the ad in week one has had time to act on it.

Ramp-up

Ads don’t hit full strength on day one. Radio and CTV need frequency to build. Paid social campaigns spend a week or two finding their audience. Direct mail lands over several days. The first one to three weeks usually show a weaker effect, and they still count in the average. A short test is mostly ramp.

Noise

Every week your test locations drift against their twin by some amount, even with no change in spend: weather, a local event, a staffing gap, a road closure. The average of that drift over the test shrinks with the square root of the number of weeks. Four weeks halves the noise of one week. Sixteen weeks quarters it. That’s why the first extra weeks help a lot and later ones help less.

The calendar

Avoid windows that straddle Thanksgiving, spring break or a big promotion unless test and comparison markets get the same hit on the same weeks. A test that ends early and clean beats one that runs through Christmas.

Working out the length: a worked example

This example is illustrative. A 64-location bakery-café chain wants to test adding $13,000 a week of paid social and search in 22 locations. Four locations next to the test markets are left out as buffers, so 38 form the comparison. The 22 test locations serve 15,400 customers a week between them (700 each). Over the last 26 weeks, the weekly gap between the test locations and their twin had a standard deviation of 3.2%.

1. Expected lift. Past market reads put an extra customer at around $28. So $13,000 a week should buy about 464 extra customers, or 3.0% of 15,400. Assume the first two weeks run at half strength: 1.5%.
2. Smallest detectable lift. At 95% confidence and 80% power it’s roughly 2.8 × 3.2% ÷ √weeks. The 2.8 is 1.96 plus 0.84; our guide to minimum detectable effect explains where they come from.
3. Compare the two at each length. The test reads once the expected average lift, ramp weeks included, is above the smallest detectable lift.

Eight weeks, the length most teams pick by default, comes back inconclusive here. Twelve weeks reads. The extra four weeks cost $52,000 of media. That stings, but it’s cheaper than $104,000 spent on an eight-week test that was never going to answer the question.

One caveat. The arithmetic treats weeks as independent, and they aren’t quite: a slow week is often followed by another. Real detectable lifts run a bit higher than the formula says, so leave some room. Countlift’s test designer runs this check on your own sales history rather than the textbook formula. If 12 weeks is too long for the business, there are other levers:

• Concentrate the spend. Put the same budget into fewer test markets. A bigger lift per location is easier to see than a small one spread thin.
• Add test locations. More locations average out more noise, as long as the comparison pool stays large enough.
• Choose steadier markets. Leave out markets with remodels, openings or a history of wild swings.

Typical geo test lengths by type of business

Gyms and hotels deserve a special mention, because their peaks are sharp and their customers decide slowly. Our guides to gym marketing ROI by club and hotel feeder-market advertising cover their timing in more detail.

Why the post-period matters, and how long it should be

Effects outlast the spend. Someone who heard your radio spot in week 7 might come in during week 9. In a go-dark test it runs the other way: switching a channel off takes weeks to show, because customers it brought in earlier keep coming back for a while. Stop reading the day the spend changes back and you’ll undercount a heavy-up and miss most of a go-dark.

Plan two to four weeks of post-period for most channels. Go longer for channels that build slowly (CTV, radio, billboards) and for businesses with long purchase cycles. Keep the freeze on prices and promotions running through it, and report the in-flight and post-period lifts separately before adding them up.

Can you stop a geo test early?

Not because it looks good. If you check every week and stop the first time the lift clears the bar, you’ll find “significant” results far more often than the 5% you signed up for, even when the ads did nothing. Each extra look is another chance for noise to cross the line.

There are two good reasons to stop. Something broke the test: a store fire, a price change in only the test markets, a media buy that didn’t run as planned. Or the result is clearly costing money, say a go-dark where sales in the dark markets fall well beyond anything seen in the pre-period. Write both stopping rules down before launch.

Don’t extend a test because it’s “almost significant” either. That’s the same peeking problem in reverse. If the planned length came back inconclusive, report it that way, say how big an effect the test could have missed, and decide whether a bigger test is worth paying for.

Common mistakes with geo test length

• Running a two-week test. Don’t. It will lie to you. Two weeks is all ramp and noise, and whatever it shows is mostly luck.
• Taking the length from the media plan. “The flight is six weeks” is a buying constraint, not a test design. If six weeks can’t read, change the budget, the markets or the length.
• Skimping on the pre-period. A twin fitted on eight weeks of history will look perfect before launch and wander off afterwards.
• Stopping the read on the last day of spend. You lose the carry-over and make every channel look weaker than it is.
• Running through holidays that land unevenly. Easter, spring break and state tax holidays fall on different weeks in different markets. Shift the window.
• Changing spend mid-test. Topping up the test markets in week 5 “because it’s working” means weeks 1–4 and 5–12 measured different things.

Questions people ask about geo test length

Pick the length on paper, from your own noise and your customers’ buying cycle. Then leave it alone until the date you wrote down.

Written by The Countlift team at First Byte in The Woodlands, Texas. Questions or corrections? Email hello@countlift.com.