By the Mayab Media Team
Every ad platform is great at grading its own homework. That's the problem.
Google, Meta and everyone else will happily tell you their ads drove your sales. Some of those sales would've happened anyway. The only way to know how many is to turn ads off somewhere and see what happens. Google just made a free tool for exactly that.
What launched
On September 9, 2026, Google announced during a community livestream that Meridian GeoX is generally available, according to PPC Land. Google repeated it the next day in its broader measurement update, calling it available globally.
What it is, in plain words:
- GeoX runs geographic experiments. Some regions see your ads, matched regions don't. Then you compare.
- It's open source and free. The code lives on GitHub, and Google says anyone can audit it.
- It feeds Meridian. Meridian is Google's open-source marketing mix model. GeoX results go in as Bayesian priors, which basically means "real test results that keep the model honest." Meridian then points out which channels most need testing next.
- It designs smarter tests. GeoX groups regions into balanced sets instead of picking them at random, and checks the design against past data before you spend anything.
Google claims "more than 31%" budget savings for large advertisers compared with other open-source tools. That's Google's internal benchmark, with no sample size or comparison set given. Treat it as marketing until someone replicates it.
There are also practical notes for whoever runs the code. Meridian 2.0.0 moved to a JAX backend by default on September 2, and it includes breaking changes, so older pipelines need updating.
Why this matters for your budget
Here's how the money leaks. Your branded search campaign shows a 12x ROAS in Google Ads. Looks amazing. But many of those people searched your brand name because they already wanted to buy. Turn the campaign off in a few test regions and you might find sales barely move. That's money you could move somewhere that actually creates demand.
The reverse happens too. Upper-funnel channels like YouTube or Demand Gen often look weak on last-click reports. A geo test might show they're doing real work that never gets credited.
PPC Land cites an estimate from a Zalando researcher that marketing mix models can overstate paid search ROAS by as much as 2.5x. That's one researcher's figure, not a universal rule. But it's exactly the kind of error experiments catch and dashboards don't.
Is this for you?
Let's be straight. GeoX is a developer tool. It's Python, it's statistics, and it needs clean regional sales data. It's not a button inside Google Ads.
It makes sense if:
- You spend enough that a 10% to 20% reallocation is real money.
- You sell across many regions (US states, metro areas, DMAs) and can tie sales to location.
- You have someone who can run Python, or a partner who can.
If you're a single-location business or a small service company in one metro, this isn't your tool yet. You can still borrow the idea, though. Pause a campaign for two weeks in a test area, keep everything else the same, and compare leads against a similar area. Crude, but more honest than trusting the platform.
Your first geo test, step by step
-
Pick one question. Not "is marketing working?" Something like "Does our branded search campaign drive sales we wouldn't get anyway?" One channel, one question.
-
Get your sales data by region. You need daily or weekly sales (or qualified leads) by geography, ideally a year or more of history. Pull it from your CRM, Shopify or order system, not from the ad platform.
-
Let GeoX design the split. Feed in the history and let it build balanced test and control groups. Check that the groups look similar on past sales before you start.
-
Set the test length and stick to it. Don't peek and stop early when results look good. That's how people fool themselves.
-
Change only one thing. In test regions, pause or raise spend on the one channel. Keep everything else (other campaigns, promos, pricing) the same across both groups.
-
Read the result as a range. You'll get an estimate of lift with uncertainty around it. If the range includes zero, the honest answer is "we can't tell yet," not "it works."
-
Feed it back into the plan. If you run Meridian, use the result to calibrate it. If you don't, use it to move budget and test the next channel.
Our opinion
Attribution reports tell you who touched the sale. Experiments tell you who caused it. Those are different questions, and most businesses only ever answer the first one. GeoX being free and open source lowers the bar a lot. The hard part now is discipline, not budget.
Takeaway: Platform ROAS tells you who got credit; a geo test tells you what actually worked, and now it costs nothing but time.
Want us to check how this affects your tracking? Book a free audit.
