Market selection is part of the experiment
A geo-lift test needs a credible estimate of what would have happened in the treated markets without the intervention. A close pre-test fit is helpful, but it does not guarantee that the comparison remains valid during the test. Market eligibility and operational constraints belong in the design before results are observed.
- Define eligible markets
- Check pre-period behaviour
- Simulate the design
- Approve operational feasibility
The geo-lift guide covers the full workflow. This review focuses on choosing a defensible market set and documenting why alternatives were rejected.
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 experiments: audit spillover before interpreting lift and Incrementality tests: choose a commercially useful MDE.
Screen markets before optimising a score
List major store openings, regional promotions, supply disruptions, unusual holidays and changes in outcome coverage. Identify markets where the planned treatment can actually be delivered and where control exposure can remain distinct. Exclude markets for documented design reasons, not because their eventual results are inconvenient.
Suppose a fictional test market has an excellent historical match but a major new store opens during the planned treatment period. The match alone cannot distinguish the store effect from advertising. Either change the timing, choose a different market or incorporate a defensible design that addresses the intervention.
Use simulations to understand the detectable effect
Run pre-period placebo or power exercises appropriate to the chosen method. Examine sensitivity to test duration, market set and effect size. Use an effect that matters commercially rather than asking only whether some large lift would be detectable.
A design that reliably detects a 20% lift may be unhelpful if the decision turns on a 3% lift. Report that limitation before committing budget. The power guide explains the distinction between statistical detectability and a worthwhile business effect.
Make the selection auditable
Record candidate markets, eligibility rules, pre-period dates, outcome definition, diagnostics, power assumptions and operational sign-off. Preserve the chosen comparison before launching the intervention. Do not repeatedly select controls after seeing the treatment-period outcome.
Inspect the resulting synthetic control or comparison weights where relevant. A design dominated by one unusual donor may be fragile. Check whether the donor markets remain comparable under plausible disruptions, without turning those sensitivity checks into a search for the preferred answer.
The synthetic control guide explains the counterfactual construction. The final decision should state what the test can learn, what effect it may miss, and which operational event would invalidate or qualify the interpretation.
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.