Geo test for Google Ads: a guide for DACH retailers

In the DACH region, a geo test is above all a question of the country

Within the DACH region (Germany, Austria and Switzerland), a geo test for Google Ads works reliably only in Germany, and even there only below the level of the federal states. The reason is a property that Google itself describes: the smallest measurable effect depends solely on the historical volatility and the fit of the regions, not on the budget. A market that is too small does not become testable through more ad spend. Before you think about test types, you therefore check whether your country offers enough comparable regions.

Whether you need an incrementality test at all, and what hurdle it has to separate, is covered in Measuring incrementality in Google Ads. This article is about the execution.

How many regions does your market really have?

For the simple design, Google requires at least ten regions and recommends 50 to over 100, because with few regions the statistical power stays low. Google explicitly names whole countries as an example of units that are too coarse. At the same time, the unit must not be too small: people who commute between regions see ads from the wrong group. Google's foundational paper generally considers postcodes too small.

Applied to the DACH region, this range gives the following picture. The figures come from the NUTS classification 2024 and from Google's table of geo targets (active entries, as of October 2026).

My assessment: Germany at district or city level is the only DACH level at which a geo test can regularly support a decision. Austria and Switzerland have enough entries in Google's table, but few orders per unit, and commuting between centres and surrounding areas weighs more heavily with small units. Both reduce the fit, and the smallest measurable effect rises with it.

How large does the effect have to be for the test to see it?

The test only sees effects that are larger than the natural fluctuation of your regions. This threshold is called the minimum detectable effect (MDE), and you can calculate it before the start from the historical data, as Google's GeoX and Meta's GeoLift by power analysis do. The decision rule is simple: the expected effect of the tested campaign must be clearly above the MDE. Otherwise the test almost certainly ends with "no significant difference", and that does not mean "no effect" but "not measurable".

A worked example with invented, round numbers: suppose a shop wants to switch off its generic search campaigns in Germany (go-dark). These campaigns are credited with 15% of shop revenue, and half of it would have come without the ad. The real drop in the test regions would therefore be 7.5%. If the power analysis at federal-state level returns an MDE of 15%, the test is pointless. At district level with a 5% MDE, it can answer the question.

This yields a commercial rule that the usual guides leave out: more budget does not lower the MDE. According to Google, it only makes less efficient campaigns detectable as well, whereas with little budget you only see very efficient ones. Measurability in a small market cannot be bought.

The process in Google Ads

1. Choose the test type. Google distinguishes three types: go-dark switches off existing spend in the test regions, holdback starts a new campaign only in the test regions, and heavy-up gives them additional budget to test scaling. For retailers with a running account, go-dark is the most direct question: what is missing when the campaign is missing?

2. Define the regions. Outside the US, Google Ads works with city IDs or postcodes, so you build a district as a group of its cities. The location option is set to presence, so that interest in a region does not trigger delivery outside the group. Geo-based Conversion Lift uses its own Google Marketing Areas, which are meant to minimise movement between regions. It requires, however, that the campaigns target exactly one country, and it shows a feasibility status in advance. At "low", Google itself advises against it, and I would follow that. More on this route in Conversion Lift for Search and PMax.

3. Plan the pre-period and run time. The pre-period must be at least three times as long as the test. Google calculates 7 to 14 weeks in total and 3 to 10 weeks of active phase, and the first days of that can go to the learning phase of the bid strategy. The window should cover at least one purchase cycle. A six-week test therefore needs 18 weeks of clean daily data before it. I would not plan a test whose pre-period contains Black Week, a shop relaunch or a tracking change.

4. Protect the budget against leakage. With go-dark on a limited budget, Google shifts the freed-up money into the control regions and so inflates the comparison base. Google advises manually reducing the daily budget to the share of the control regions and using individual budgets instead of shared budgets. If you overlook this, you measure an effect that is too small.

5. Move only one lever. During the run, bids, feed prices and other channels stay the same in all regions. A regional newsletter or a poster campaign in a test region makes the result unusable.

When the data per region is too thin

GeoX needs daily data, and small shops often have zero or one order per district and day. Google names three levers: test for longer, say six to eight weeks instead of four, use a shallower conversion with more volume such as add-to-basket actions, or use a larger test group. As a measure of forecast quality, Google recommends an R² of 0.8 or higher, and below 0.5 the control regions do not predict the test regions reliably.

You check the R² before the start, on the pre-period, not afterwards. If it is already below 0.5 there, no test duration will change that.

What to do about Austria and Switzerland?

For Austria and Switzerland I advise against a geo test of their own in most cases. Nine federal states are below the minimum, 26 cantons are far below Google's recommendation, and smaller units rarely have enough orders per day. The test then costs weeks of management and delivers an MDE that no realistic campaign exceeds.

My approach instead: test in Germany, carry the result over to Austria and Switzerland as an assumption, and work there with small, reversible budget steps whose effect you track in the country's total revenue, for example through MER and blended ROAS.

The transfer does not hold if the market differs markedly: a different range, prices in CHF with their own margin, a different main competitor, or a much stronger or weaker brand in the country. Then two routes remain. Either a geo-based Conversion Lift, if Google shows a high feasibility status for the account, or a decision run openly as a bet with a fixed review date. What else sets the three markets apart is covered in Google Ads in the DACH region.

When you should not run a geo test

A geo test is not worth it if any of these conditions applies:

• The power analysis returns an MDE that is above the expected effect.
• The pre-period is shorter than three times the test length, or disturbed by promotions and tracking changes.
• You cannot do without regional special promotions and price changes during the run.
• The result would not change a budget decision, because it is already settled.

For brand campaigns, the question is often easier to settle by other means, see Brand revenue is not automatically Google Ads performance. I consider a test that cannot measure more expensive than no test: it costs revenue in the test regions and creates a false certainty that everyone then relies on.