Google GeoX: designing a test of advertising's real contribution

Prepare a geographic experiment, choose suitable regions and data, and assess incremental sales without confusing them with reported ROAS.

Start with a change that can be implemented

Meridian GeoX supports the design and analysis of geographic experiments. A company changes advertising exposure in selected regions and compares outcomes against a reference. The question is how much additional sales that change caused. This differs from assigning existing conversions to clicks. First define the intervention, outcome and budget decision the study needs to support.

Do not divide the map arbitrarily

Regions must support implementation and a credible comparison. Examine historical sales, variation, market size and campaign targeting options. Movement between regions and shared exposure to other channels can dilute the difference. GeoX documentation describes group design and minimum detectable effect estimates. Insufficient power is a reason to redesign the study, not conceal uncertainty.

Prepare the pre-experiment data

You need consistent outcome series by region and period, plus spending data for applicable test types. Decide whether the measure is orders, net revenue or something else. Check missing values, definition changes, returns and geographic assignment. Do not mix customer residence with branch location without justification. Historical data errors can undermine group selection before the campaign begins.

Choose a feasible intervention

GeoX describes designs including stopping activity, increasing intensity and withholding a new campaign from some regions. Not every option suits every business. Reducing advertising can have a sales cost; increasing it requires budget and delivery capacity. Agree what will change and what must remain stable. A design that is never implemented in the media platform is not an experiment.

Separate design from execution

The documentation treats implementing the intervention in advertising platforms as a separate stage. Building the model does not automatically change campaigns. Record settings and check that delivery follows the plan. Regional promotions, stock shortages and price changes need documenting. If the groups did not receive the planned difference in exposure, the interpretation must acknowledge that.

Read lift together with uncertainty

The analysis compares observation with a scenario without the intervention. Consider incremental outcome, uncertainty interval and assumptions rather than one attractive number. An inconclusive effect does not automatically prove advertising has no effect; the test may have been underpowered. Even an illustrative 100 additional orders must be evaluated using margin and extra cost before deciding to scale.

When GeoX is worth considering

It suits businesses with enough scale, regional data and the ability to change media in a controlled way. A small local business may lack these conditions. Begin with feasibility rather than promising a definitive result. Experiment findings can inform MMM calibration, but they apply to the particular intervention and circumstances. Do not transfer them unquestioningly to every product, season or budget level.

FAQ

Does GeoX automatically run the campaign?

Design and analysis are separate from implementing changes in the advertising platform. Execution and monitoring must be arranged independently.

Can every business obtain a reliable result?

No. Suitable data, scale, comparable regions and a feasible intervention are needed. Assess them before the test.

Sources

Information checked: 25 September 2026. Feature access may depend on the account.