A geographic boundary does not contain every exposure
People travel, media delivery crosses borders and online campaigns can reach customers outside their intended market. A geo experiment can therefore expose some control customers or fail to reach some treatment customers. The relevant question is whether the implemented contrast still matches the decision being tested.
- Map exposure boundaries
- Check customer movement
- Monitor treatment contrast
- Qualify the estimate
Start with the geo-lift design, then map where exposure and outcomes are measured. A store-location outcome and a customer-home-location outcome can respond differently to the same travel pattern.
The effect you can estimate depends on the intervention, comparison and outcome you define. The incrementality and experiments learning path connects this idea to the assumptions, practical checks and next steps.
To work through the connected decisions, read Geo-lift market selection: review more than pre-test fit and MMM calibration: keep an experiment evidence register.
Work through a contaminated contrast
Imagine treatment markets receive a planned £100,000 increase while controls unexpectedly receive £20,000 of similar media. The contrast is no longer treatment versus unchanged control. Dividing the estimated outcome difference by £100,000 without explaining the control change may misdescribe the intervention.
This does not imply a universal correction factor. Spillover can affect customers and outcomes in complicated ways, and it need not always bias the estimate toward zero. Preserve the actual delivery evidence and define the effect the design can support.
Check the likely pathways before launch
Review geo-targeting settings, media-market boundaries, commuting, delivery areas, national campaigns and retailer catchments. Consider buffer regions or a different assignment unit where justified. Decide how sales are attributed to a geography and keep that rule fixed.
Shared promotions can also contaminate the contrast even when ad targeting is accurate. Coordinate with the teams running email, pricing, stores and other media. A campaign calendar is an experimental control document, not merely a reporting convenience.
Monitor exposure without chasing the outcome
During the test, check spend and delivery by assigned market, data completeness and major operational deviations. Monitoring whether the treatment was implemented is different from repeatedly checking significance and stopping when the estimate becomes attractive.
Predefine which deviations trigger a pause, an extension or a qualified analysis. Report assigned treatment, actual delivery and the primary effect estimate together. The incrementality guide explains why the treatment definition must remain attached to the result.
When calibrating an MMM later, include the spillover assessment in the evidence record. Do not transfer a diluted or altered contrast as though it were a clean channel shutdown. If the contamination is too severe to support the original question, say so and design a more suitable intervention.
Sources and further practice
This article’s worked example is illustrative, not a measured client result. Use the linked methodology to check the assumptions before applying it.