Geo test vs conversion lift test: which one should you run?

A conversion lift study is a test run by the ad platform inside its own system. Google describes Conversion Lift as a tool that measures the causal impact of ads by comparing a treatment group with a control group; the difference in conversions is the lift¹. Google offers user-based and geo-based versions, with groups built from aggregated user attributes or aggregated geographical data, and says the geo-based version supports offline data¹.

Meta's lift studies randomly assign user accounts (Meta calls them Accounts Center accounts) to a test or control group². Access is not open to everyone. Google says Conversion Lift is not available for all Google Ads accounts¹, and Meta says Conversion Lift Measurement is currently limited and points advertisers to a Meta representative². Details of setup and eligibility are in conversion lift study requirements.

A geo test splits regions, not people. Some regions keep the ads, some lose them (or get more), and you compare sales between the two. The open-source GeoLift package on GitHub, for example, describes itself as an end-to-end geo-experimental method based on synthetic control³. Google's Meridian GeoX documentation lists holdback, go-dark and heavy-up among the experiment designs it supports⁴.

The key difference is whose data decides. In a geo test, the result can come from your own shop orders by region, not from what one platform tracked. Step by step, see how to run a geo holdout test. For the tools, see GeoLift vs Meridian GeoX vs CausalImpact.

In our view, this is the part finance cares about most.

In a platform lift study, the platform designs the split, runs the ads, counts the conversions and reports the lift. That is convenient. It also means the seller is grading its own product. An AdExchanger article warns against letting a platform decide whether a sale was incremental or whether a customer was new, since a customer the platform counts as new may already be in the brand's CRM⁵. On an agency podcast, the host called platform-reported revenue the weak spot of running tests inside the platforms⁶.

That does not make lift studies useless. It makes them evidence from an interested party. For a small, in-channel decision, that can be fine. For a decision that moves money between channels, you want the result built on your own sales.

A typical A/B test compares two versions: two ads, two landing pages, two bids. It tells you which one does better. It does not tell you whether running ads at all adds anything, because both groups still get the activity.

An incrementality test compares "with" against "without". It tells you what the spend added. The mechanics can be the same as an A/B test, a random split, but one group gets no ads at all: that is a holdout. A geo test and a lift study are both ways to run one.

So the rule is simple. Choosing between versions: A/B test. Deciding whether the spend earns its place: an incrementality test. For when each makes sense at your scale, see incrementality for mid-size ecommerce brands.