Through a causal impact model, not through a before-and-after comparison. Causality is inferred from a stratified geo comparison against a counterfactual prediction, and from the deviation from that prediction during the campaign window.

Without user-level attribution, a geo comparison can look like correlation wearing a statistics costume. Sales rose in the market where the billboards were, and so did several other things.

That is the most credible version of the objection, and it deserves a real answer rather than a reassurance.

It builds a counterfactual: a prediction of what your campaign markets would have done during the campaign period, trained on your own pre-period history and on comparable markets where OOH was not running.

The campaign is then treated as an intervention on a time series. Causality is inferred from the size and shape of the deviation from the prediction during the intervention window, not from the fact that the line went up.

The underlying approach is a published causal inference method from Google researchers, using synthetic control and Bayesian regression. It is not proprietary arithmetic invented for OOH, which matters when your analytics team asks.

Every result carries a statistical significance figure. Ninety-five percent is the bar for calling an impact causal, and in practice sales lift studies commonly land higher.

The control markets are selected by the model against your pre-period rather than chosen by whoever wants the result to look good.

The output includes the counterfactual itself, so the prediction can be inspected against the actual rather than taken on trust.

Each of those is detectable in advance, which is why feasibility runs before the study rather than after it.

It will not tell you which unit or which creative drove the result, and it will not de-duplicate a person across channels. It answers whether the channel produced incremental business in a market. That is a narrower claim than a full attribution story, and it is a defensible one.

Myth: A geo lift result is correlation and cannot be called causation.

Reality: These are causal impact models: counterfactual prediction, stratified geo comparison, and deviation during the intervention window.

Myth: Search lift can establish causality.

Reality: It cannot. It is a relative comparison. Causal lift is the study that addresses causality.

Myth: Footfall correlation and causal lift are the same claim.

Reality: Footfall attribution is single-touch correlation and should be caveated as such.