Geo Testing

Geo testing changes advertising in some geographic areas and leaves others alone, then compares results to estimate the effect. The unit is a region, such as a city, a metro area, or a state, not a person. Google's Conversion Lift includes a geography-based design, built from aggregated geographic data and able to include offline data, and Google's open-source Meridian marketing mix model has a tool called Meridian GeoX for experiment results. Geo tests are also run by advertisers on their own, using spreadsheets or statistical packages, with campaigns targeted by location. Google's geo-based Conversion Lift builds groups from aggregated geographic data, and Meridian GeoX is designed to turn experiment results into priors for the model.

The design starts with matching. You pair or group regions that behaved similarly in the past on sales, traffic, and seasonality, then assign some to test and others to control. In the test regions you raise, lower, or switch off spend, while control regions continue as before. You model what the test regions would have done without the change, using the control regions and history, and compare. Because it uses aggregate results by region, it needs no user-level identifiers, which suits measurement under privacy limits. Google's geo-based Conversion Lift reports incremental ROAS, incremental conversions, incremental conversion value, and incremental cost. Choose a conversion source that you can measure by region, such as orders by shipping zip code or leads by area code.

Geo testing interacts with how your ads are targeted and where your customers are. You need the ability to target by location (Google Ads and Meta both allow it), enough regions to form matched groups, and sales or leads data by region, including offline sales if they matter. It suits national or multi-city advertisers with some geographic spread, and fits channels that no pixel can track, such as TV, radio, or out-of-home. Duration matters too: you need enough weeks to see an effect above normal week-to-week swings, and a pre-test period of stable data to build the baseline. Avoid regions that share a media market, because ads aimed at one can reach people in the neighboring one.

The classic pitfall is contamination. Media spillover can cross a market boundary, since people commute, travel, and see the same national content. A promotion or store opening in a single region can inflate the numbers. Another mistake is choosing test regions because they are your best markets, which breaks the match. A third is too few regions, so one unusual city dominates. A fourth is ending the test early. Also remember that a geo test measures the net effect of a spend change in those regions, not the effect of one ad. Its result applies to similar markets, not automatically to every audience. Keep the rest of the marketing plan steady, including prices and promotions, so the only major difference is the spend you changed.

If you want to try one, start simple. Pick six to ten similar markets, split them into test and control, and decide in advance what you will change and for how long, for example a 50 percent budget increase for four weeks. Keep pricing, promotions, and other major campaigns the same across all regions. Record daily sales by region from before the test to a couple of weeks after. When it ends, compare actual sales in test regions with what the control trend predicts, and divide the difference by the extra spend. After the test, review whether the pre-test trend of test and control regions matched, since a poor match weakens the answer.