Google's geo experiments go global, and its new uplift metric counts conversions, not customers

Google has moved Meridian GeoX out of beta and made it generally available worldwide, and in the same announcement introduced a Google Ads metric that estimates how many conversions an advertiser's first-party data setup pulled back into view. The two shipped together as one measurement story. For most software marketers, only one half is usable this quarter.

The release splits cleanly. One half is plumbing. Data Manager, the hub where advertisers connect CRM records, offline events and app data, is being wired directly into Google Analytics and Display & Video 360. Its API has been rebuilt on the IAB Tech Lab's Event and Conversions API standard, so the same pipeline can feed other ad platforms instead of one bespoke integration per destination. Enhanced conversions are expanding into both products, and new diagnostics are meant to catch broken data before a campaign optimises on it.

The other half is econometrics. Meridian, Google's open-source marketing mix model, gets agentic help that audits data quality and guides model building in real time, a faster back end, and the option to feed branded query volume in as a brand signal. GeoX is the piece that matters most: an open-source library that measures lift by geography rather than by user, and which Google says can be pointed at any ad platform, not only its own.

The numbers are Google's own

Every figure above comes from Google internal data, footnoted on Google's own post and drawn from windows the company chose. The 26% ROAS figure covers April 2025 to April 2026 and is limited to Search campaigns bidding to conversion value. The 11% comparison ran over the first two weeks of January 2026. Treat them as directional evidence that better signal helps a bidding system, not as a forecast for your account.

The new metric is called Data Strength Uplift, and it lives in Google Ads. It estimates how many of your conversions were always happening but went unmeasured until you fixed tagging or connected another source.

The experiment most software companies cannot run

GeoX is the more interesting release and the less accessible one. Geo experiments split markets into treatment and control, change spend in one group, and read the difference. That sidesteps user-level tracking entirely, which is why it keeps surfacing as the answer to attribution decay.

The requirements are the problem. A market-level test only reads cleanly when every market produces conversions each day, and when the markets differ enough from one another to assemble a credible control group. Google recommends GPU resources for the modelling. The test also costs media, because producing a readable difference means deliberately raising, cutting or withholding spend in chosen markets for its duration.

There are still costs associated with the people, data, infrastructure, and media required to build models and run experiments

That assessment is from Brooke Osmundson, writing in Search Engine Journal, which covered both halves of the launch. Meridian carries no licence fee at all, and that is precisely what makes its real cost easy to underestimate.

Now apply that to a B2B software business. A geo test needs enough conversions per market per day to move a confidence interval. Enterprise software demand is usually thin, concentrated in a handful of metros, and the revenue it becomes lands quarters after the click. Holding budget out of your strongest market for the length of a test is a hard sell when pipeline coverage is already tight. GeoX is built for advertisers with volume and geographic spread. That is not most software marketers, and pretending otherwise wastes a quarter.

Recovered conversions are not qualified ones

The Data Strength Uplift metric has a narrower blind spot, and it is the one that should worry B2B teams. Recovering conversions tells you your measurement got more complete. It says nothing about whether those conversions are worth having. Search Engine Journal made the same point in its write-up of the Data Manager changes: a fuller count is not a better one.

For software marketers this is not a technicality. If the conversions you send back are form fills, demo requests and ebook downloads, then improving data strength means feeding the bidding system more of a signal that was already teaching it to find the wrong buyer, and you now get a number that rewards you for it. The metric is real and useful. It is also an easy thing to put in a board deck in place of pipeline.

The one genuinely broad win here is the ECAPI-based API. A single standard for pushing conversion events to multiple platforms cuts the integration work that keeps small teams from sending CRM-stage data anywhere at all.

What a software marketer should do

• Send stage, not submission. Wire Data Manager to fire on qualified opportunity or closed-won from your CRM, not on the form.
• Read Data Strength Uplift as a data-quality gauge, never as growth, and keep it out of revenue reporting.
• Rebuild one conversion pipeline on the ECAPI standard before adding a second destination.
• Scope GeoX honestly against your daily conversion volume per market before anyone promises an incrementality read.