Geo Lift Testing for Influencer Campaigns

Link and code attribution answers a narrow question: which tracked path preceded a purchase. It cannot answer the question a finance team actually asks, which is whether those purchases would have happened anyway. A discount code redeemed by an existing customer who was already going to buy looks identical in the report to a new customer the creator introduced.

Geo lift testing answers the causal question by comparison. Run the campaign in one set of regions, deliberately withhold it from a matched set, and read the difference in outcomes between them. The withheld regions act as an estimate of what would have happened without the campaign.

It is a blunt instrument and it costs real reach, because a holdout region is a market you chose not to sell to for the duration. That is the price of an answer that survives scrutiny.

Choose the outcome first: orders, revenue, new customers, or store visits, defined exactly as your commerce or CRM system records it. Then pick regions that are similar on the baseline of that outcome, on seasonality, and on existing marketing pressure. Matching on population alone is the classic mistake, because two cities of the same size can have very different existing demand.

Use several regions per group rather than one against one. A single test region and a single control region can differ for reasons that have nothing to do with your campaign, and there is no way to tell after the fact.

Decide the flight length and the analysis window before you start, and write down what result would count as a success. A test whose success criterion is chosen after seeing the data is not a test.

The main threat is spillover. Creator content does not respect regional boundaries, so a national creator with a large audience in the holdout region contaminates the control and shrinks the measured effect toward zero. Prefer creators whose audience is concentrated in the test regions and check the audience geography before booking.

Hold everything else steady. If paid media, email, or a promotion runs in one group and not the other during the flight, the test measures the combination and cannot separate the creator contribution.

Keep a written log of everything that changed during the window, including site releases, stock outages, and competitor promotions. Most tests that produce a confusing result do so because something entered the window that nobody recorded.

Compare the change in the test group against the change in the control group over the same window, rather than comparing the test group against its own past. A before and after comparison inside one group absorbs every seasonal and market effect that the control group exists to remove.

Report the result as a range, not a point. Small tests produce noisy estimates, and a single confident number invites decisions that the evidence does not support. Where the range includes no effect, say so; a null result on a costly channel is a valuable finding.

Then use the result to calibrate your everyday attribution rather than replacing it. If lift consistently exceeds what tracked links credited, your dashboard is understating creator contribution and the correction factor is now measurable.