Meridian GeoX Goes Global, and Google Just Rewired Your Measurement Stack
Google made Meridian GeoX globally available and rebuilt its measurement stack. What changes for your funnel, which numbers to trust, and the six moves to make this week.
Google shipped a measurement overhaul on September 10, and most coverage will tell you what launched. This is the version you actually need: what it changes about how your funnel gets credited, which numbers to trust, and the six moves to make this week.
The short version: Google made incrementality proof free and made its data plumbing universal. Every headline number in the announcement, the 14%, the 26%, the 11%, describes conversions your funnel already produced that Google’s systems could not see, verify, or use. That is not an analytics upgrade. That is a conversion-rate upgrade hiding in your tracking layer.
What actually shipped
Three things, and they connect.
1. Meridian GeoX went global
GeoX is Google’s open-source library for causal geo-experiments. Some regions see your ads, matched regions do not, and you measure the real difference. No attribution model, no modeled conversions. On September 10 it left beta and became generally available worldwide inside Meridian, Google’s open-source marketing mix model (MMM).
Google claims its JAX-based backend generates experiment designs over 94% faster than other open-source and legacy solutions, and that multicell design (several treatments against one shared control group) cuts experiment budgets by more than 31% for large advertisers. Both are vendor numbers from internal benchmarks. Hold them loosely.
The bigger change is methodological. GeoX replaced matched-market heuristics with stratified sampling using out-of-sample validation. Google’s own argument: matched-market approaches reuse the same historical data to fit the model and estimate its variance, which can underestimate real-world noise. That produces tests with attractively low minimum detectable effects that are quietly underpowered. If you have ever run a geo test that “worked” on paper and found nothing in market, this is one likely reason.
2. The data foundation got unified
Data Manager, Google’s first-party data hub, now integrates directly into Google Analytics and DV360, and enhanced conversions are launching in both. Google reports advertisers using enhanced conversions see an average 11% increase in Search conversions versus standard conversion imports.
The Data Manager API went universal, rebuilt on the IAB Tech Lab’s Event and Conversions API (ECAPI) standard. One endpoint for audiences and conversion events across major ad platforms, with built-in diagnostics that flag data issues before they touch campaign performance.
3. Google started scoring your data quality
The new Data Strength Uplift Metric in Google Ads quantifies the conversions recovered by your first-party data setup. Google reports advertisers using Google tag gateway observe an average 14% conversion uplift, and over 20% for Demand Gen campaigns.
Meridian itself got upgrades: agentic tooling that audits data quality and guides model building in real time, a faster backend, and brand signals like Branded Google Query Volume now modeled directly, so upper-funnel TV and video spend can finally show its downstream effect.
Read the numbers like an operator, not a press release
Four claims, four footnotes. Check the windows:
- 26% incremental ROAS lift: Google Data, April 2025 to April 2026. A year of data, reasonably credible.
- 11% Search conversion lift: January 1 to 14, 2026. Two weeks.
- 14% tag gateway uplift: July 2024 to June 2025, split into two six-month periods.
- 20% Demand Gen uplift: June 3 to 17, 2026. Two weeks.
Two of these rest on fortnights. That does not make them false. It makes them hypotheses, not guarantees. Google is also the media seller here, grading its own inventory. PPC Land documented the same conflict with Meta’s Robyn, and a Zalando researcher found standard MMM practice overstates paid search returns by roughly 2.5 times. Vendor-measured, vendor-benchmarked. Your own geo test is the only grade that counts.
Why this matters more for funnels than for dashboards
Your funnel produces the conversion events that every one of these systems consumes. When the signal path breaks, a redirect strips the GBRAID parameter, a form fires the wrong event, enhanced conversion fields never populate, the damage is silent. Bids optimize on partial data. MMM reads phantom channel weights. You cut a channel that was working or scale one that was not.
We wrote about this exact failure mode in the GA4 parameter stripping diagnostic: Google can now detect the landing-page defect, but only after the attribution already broke. The September 10 release is Google telling you the same thing from the other direction. The systems reading your data are getting sharper. The data your funnel sends them is still your responsibility.
There is also a compounding effect. Meta’s Andromeda and TikTok’s Smart+ both train delivery on post-click conversion signals. Platforms that receive cleaner, richer, better-matched signals bid smarter. Your tracking layer is now an input to three bidding algorithms at once.
The operator playbook: six moves this week
- Audit the conversion signal path. Trace one real conversion end to end: click, landing, form fill or checkout, CRM, offline close. Confirm the event fires once, with the right parameters, in both the browser and the backend.
- Turn on enhanced conversions in GA4 and DV360. It is new in both. If you already use it in Google Ads, this extends the match surface with no new infrastructure.
- Consolidate pipelines through Data Manager. One ECAPI-standard endpoint replaces parallel audience and conversion feeds. Fewer pipelines means fewer places to break, and the built-in diagnostics catch breakage before your campaigns do.
- Check your Data Strength Uplift Metric. It is now in Google Ads. A low number is not a shame score, it is a to-do list with a dollar value attached.
- Run one go-dark or holdback geo test. Pick your highest-spend channel. Pause spend in a set of regions, or hold out a holdback group, and measure with GeoX. Design it for conclusive results, not cheap ones. This is the step most teams skip, and it is the only one that produces causal proof.
- Feed the result into Meridian as a prior. The new calibration module converts experiment results into model inputs, and Meridian then recommends which channel to test next. That loop, test, calibrate, re-test, is what Google means by “proactive performance engine.” One warning before you start: Meridian version 2.0.0 shipped on September 2 with the new backend and calibration module as breaking changes, so existing model pipelines need updates before you build on them.
What to stay skeptical about
The 94% and 31% figures come from internal benchmarks against unnamed comparisons. The two-week uplift windows are thin. And Google has a stated plan to consolidate experimentation methodologies so GeoX becomes the primary path, which is convenient for the company that operates the auction you are testing in.
None of this makes the stack wrong. It makes it free, powerful, and biased toward its maker. Use it, and grade it with your own experiments.
The takeaway
Measurement stopped being a report card the day your competitors started feeding cleaner signals into smarter bidders. Google just handed you the tools to close that gap: free incrementality testing, universal data plumbing, and a score that tells you exactly what your broken signal path is costing.
Most funnels will adopt none of it. The plumbing is unglamorous work, and “we fixed our conversion feed” does not demo well. That is precisely why it pays. If you want a funnel built with its measurement layer treated as part of the product, not an afterthought, that is how we build. Talk to us.