Your other markets can serve as the control only if they are genuinely comparable. Control markets have to match on historical sales and addressable population, and the analytics team selects them rather than accepting a list.
What makes a control market
Equivalence on historical sales trend and on addressable population, not geographic proximity or gut similarity. Two markets that feel alike can behave nothing alike in the data.
The model picks them, iterating across candidate sets and running dozens of comparisons to find the combination that predicts your pre-period best. That is why the selection is not a conversation about which cities feel right.
A feasibility check runs first, before anyone commits to the study.
You need markets where OOH is not running
This is the constraint most often missed. Control data has to come from markets where OOH is not running. If you are running everywhere you care about, there is no control and no study.
Five clean markets is the floor. Below that there is not enough to build a counterfactual from, and the analysis degrades toward simple market matching with materially less confidence. Ten or more is the comfortable ask.
Include those markets in the data file you send. A file containing only your campaign markets cannot be modeled, and this is the most common reason a causal lift request goes back for a second export.
What will invalidate the result
A large concurrent digital campaign in the same markets that we do not know about. The model attributes the deviation to the intervention it knows about, so tell us what else is running.
Renewing the same creative year over year in the same market. There is nothing new to detect, and a flat result is the correct answer rather than a measurement failure. New creative is what creates something to measure.
A test too small for the market. A brand large enough that a modest test barely moves the aggregate will not register lift, and that is arithmetic rather than channel performance.
Common misconceptions
Myth: Any two markets can serve as test and control.
Reality: They must be equivalent on historical sales and addressable population, and the analytics team selects them.
Myth: A flat result means the measurement did not work.
Reality: If the creative and the presence were unchanged year over year, flat is the honest answer.