How I'd design a geo test

1. 1. Pre-register the decision

   Before anything launches, write down the hypothesis, the KPI (booked sales by market, not platform conversions), the break-even lift, and what happens in each outcome: keep, cut, or scale.

2. 2. Choose the markets

   Pick test and control markets whose sales moved together in the pre-period. Tools such as GeoLift help search market combinations. Avoid markets with planned promotions, store openings, or anything else that would hit one group and not the other.

3. 3. Size it before launch

   Use the pre-period noise between test and a scaled control to estimate the smallest lift the test can detect and how many weeks it needs. If the detectable lift is bigger than any plausible effect, change the design: more markets, a bigger spend change, or a longer run.

4. 4. Run it cleanly

   Change only the variable under test. Freeze other media changes in the test markets, watch for spillover from national campaigns, and don't stop early because the chart looks good.

5. 5. Read it honestly

   Report the lift with its interval and the incremental return per dollar of test spend. Compare the whole interval to break-even. If the interval straddles it, the result is inconclusive, and saying so is part of the job.

6. 6. Feed it back

   A finished test calibrates the mix model and settles disputes between platforms that claim the same sale. Log it with the design, so the next test starts from what was learned.

This is how I'd approach the problem in general. It describes a method, not results from any employer or client.

Try it: Geo test readoutTry it: Experiment power calculatorThe shift from MTA to macro-measurementHow I'd approach an MMMHow I'd reconcile platform numbers with cash