Your quarterly MMM deck still lands like a report card: channels graded, ROIs printed, next quarter already locked. Google just told you that loop is the product failure mode.
On September 10, 2026, Nipoon Malhotra (VP, Ads Analytics, Insights, and Measurement) posted Google’s measurement suite update. The language is blunt. Earlier this year Google said it wanted measurement to move from a “reactive report card” to a “proactive performance engine.” The Meridian piece of that post is not a footnote. Agentic help for data quality and model building, brand signals such as Branded Google Query Volume inside the model, and Meridian GeoX moving from beta to global general availability all land in the same stack. Marketing Dive (Peter Adams, September 14) and Search Engine Land (Anu Adegbola, September 10) both read it the same way: MMM is being wired for audit, upper funnel, and causal calibration, not for another static board pack.
If your team still opens Meridian only when finance asks for a slide, you are optimizing the wrong operating cadence.
What actually changed
Three Meridian moves matter for operators this week. Treat them as one system, not three feature bullets.
First, agentic capabilities. Malhotra’s post says Meridian now helps audit data quality, resolve errors, and guide model building in real time, with backend updates aimed at faster runs. Marketing Dive frames the practical intent: less manual troubleshooting before an analysis can move. Search Engine Journal’s Brooke Osmundson (September 10) is careful: the tools do not invent your inputs. Teams still choose which business and media data belong in the model and how much confidence to put in the output. What changed is that data hygiene and model guidance sit closer to the run, not in a separate analyst queue three weeks later.
Second, brand signals inside the model. Meridian can now include relevant brand signals, with Branded Google Query Volume called out explicitly, so upper-funnel video and TV effects can be modeled toward future sales rather than ignored because last-click never saw them. Search Engine Land notes TV and out-of-home as the obvious beneficiaries. Osmundson adds the caveat operators need: a branded search lift alone does not prove a campaign caused future sales. Seasonality, promotions, competitors, and news move branded demand too. The upgrade is a fuller model surface, not a causal free pass.
Third, Meridian GeoX is generally available globally. Google’s developer docs describe GeoX as a global, open-source, publisher-agnostic geo-experiment library for transparent causal tests that can calibrate MMM. Malhotra says you can run independent experiments or fold incrementality results into Meridian to boost accuracy. Developer docs go further on the join: GeoX results convert into priors that calibrate Meridian, and the stack can recommend which Meridian channels would benefit from a GeoX test. Osmundson stresses the cross-platform angle: you are not limited to testing Google media. Holdback, go-dark, and heavy-up designs are in scope for teams with enough geographic variation and daily time series to run them.
Around Meridian, the same post expands Data Manager into Google Analytics and Display & Video 360, makes the Data Manager API universal on IAB Tech Lab’s ECAPI standard, adds diagnostics, and ships a Data Strength Uplift Metric in Google Ads. Google’s own footnotes claim average lifts for advertisers connecting offline and app data (26% incremental ROAS), using enhanced conversions (11% Search conversions vs standard imports), and building data strength with Google tag gateway (about 14% conversion uplift, over 20% on Demand Gen in the cited windows). Those are Google internal figures. Use them as vendor evidence for why first-party plumbing sits next to MMM, not as your forecast.
The mechanism: audit, brand, and geo priors in one loop
Strip the launch copy and the operating loop looks like this.
You feed first-party and media data into a cleaner pipe (Data Manager, enhanced conversions, diagnostics). You build or refresh an MMM with agentic help catching quality breaks while the model is still forming. You optionally add brand signals so upper-funnel spend is not invisible to the mix. You run GeoX where the model is soft or the budget decision is large, then push experiment results back as priors. The output is not a PDF. It is a calibrated prior set and a channel recommendation that should change next week’s allocation conversation.
That is why “report card” is the wrong metaphor. A report card grades the past after the semester ends. A live measurement loop audits inputs while you model, measures brand demand while you spend, and injects causal evidence before you move money. Search Engine Land’s “why we care” line lands the operator shift: stop asking which channel received credit for a conversion and start asking which investments created incremental growth.
GeoX is the hinge. MMM estimates relationships from historical co-movement. When channels move together, estimates get mushy. A geo experiment creates treatment and control areas without user-level tracking, which matters when cookies and identity graphs are unreliable. Developer docs sell cost and design flexibility (multi-cell against a common control, time-based regression, stratified sampling). Osmundson sells the board-meeting use: if a geo test supports what the model says for a channel, you have evidence executives can follow without reading every Bayesian assumption. When model and experiment disagree, that conflict is useful too. It is a reason to reopen an assumption, not a reason to bury the slide.
Agentic auditing sits earlier in the loop than most teams staff for. Bad feeds used to kill a quarterly rebuild. Now the product story is real-time guidance while building. That only helps if someone owns the data contract. An agent that flags missing geo grain or broken spend series does not replace a measurement lead who can say which series is allowed into priors.
What operators get wrong
Mistake one: treating Meridian as a quarterly deliverable. If the only MMM artifact your CFO sees is a QBR appendix, you will miss that GeoX GA and agentic audit tools are built for a faster refresh cycle. Schedule a living model review tied to budget gates, not to calendar theater.
Mistake two: stuffing Branded Query Volume into the model and declaring brand “solved.” Osmundson is right. Branded search is a signal, not a verdict. Keep confounders in the brief. Pair brand signals with experiments on the upper-funnel bets you actually want to defend.
Mistake three: running GeoX only on Google inventory because Google shipped the library. The docs and SEJ coverage both say publisher-agnostic. If your hardest incrementality question is a rival social or CTV line, that is where a geo design earns its keep.
Mistake four: confusing open source with free. Meridian has no license fee. Osmundson lists the real costs: people, data, GPU compute Google recommends for Meridian, and media you must raise, cut, or hold to create treatment and control. Agencies that sell “we’ll stand up Meridian this sprint” without geo capacity and experiment budget are selling a toy.
Mistake five: letting Google grade Google without a second ruler. Marketing Dive notes the perennial concern that Google is grading its own homework. GeoX being usable across platforms is the partial answer. Your process answer is independent experiment design, holdout discipline, and a written rule for when platform-reported ROAS can override a calibrated prior (almost never without a documented exception).
Mistake six: celebrating Data Strength Uplift Metric as proof your MMM is done. Uplift from first-party setup is a Google Ads metric about recovered conversions from data plumbing. Useful for the data foundation half of Malhotra’s triad. It is not Meridian calibration. Do not merge those dashboards in the same sentence to leadership.
Mistake seven: skipping diagnostics because “the agent will catch it.” Built-in Data Manager diagnostics and Meridian’s agentic audit reduce time to find breaks. They do not invent business definitions. If your offline revenue feed double counts returns, you will get a faster wrong model.
Second-order effects
Measurement vendors and in-house science teams will compete on loop speed. The scarce skill is not fitting another regression. It is owning the join between clean inputs, brand-aware MMM, and geo priors that finance will fund. Agencies that only deliver attribution screenshots will lose share to teams that can run multi-cell geos and translate priors into budget moves.
Upper-funnel creative and media leads get a new seat at the measurement table. If branded query volume and TV or OOH can enter the model, brand teams can no longer be told “not measurable” as a default. They will also face harder questions when brand signals move and sales do not. That is healthy. It kills both brand mysticism and performance tunnel vision.
Procurement will see “agentic MMM” pitches everywhere. Demand a demo of data-audit behavior on your dirty feeds, not a slide of Meridian’s GitHub stars. Ask which GeoX designs they have run at your geo grain. Ask how priors are versioned when an experiment contradicts last quarter’s model.
Platform politics intensify. A Google-built open stack that can test non-Google media is still a Google narrative. Sophisticated buyers will keep a second MMM or an independent incrementality partner in the room when the budget move is large. That is not paranoia. It is how you keep causal claims honest when the library author also sells inventory.
What to do this week
- Rewrite the measurement operating cadence. Replace “MMM deck due Friday” with a living loop: weekly data-quality review, monthly model refresh triggers, and geo tests queued against the three channels with the softest priors or the largest proposed spend shift.
- Inventory brand signals you can legally and cleanly feed. Branded Google Query Volume is the named example. Decide which upper-funnel lines enter Meridian this cycle and which stay out until you have a GeoX plan.
- Pick one GeoX candidate, not five. Prefer a channel where MMM and platform attribution disagree, or where leadership is about to move seven figures. Confirm daily geo-level outcomes and enough markets for control. Price the media distortion of the test before you sell the science.
- Staff the data contract. Name the owner who accepts or rejects feeds when Meridian’s agentic audit or Data Manager diagnostics flag a break. Without that owner, agentic tooling becomes alert spam.
- Separate Google’s internal lift footnotes from your forecast. Cite Malhotra’s Data Manager and enhanced conversions averages as vendor context. Rebuild incremental impact on your geo tests and calibrated priors.
- Brief finance and brand together. Show how brand signals and geo priors change the budget conversation. If brand still lives in a different scorecard from performance, this launch will create two decks that still do not talk.
- Ask your agency or science partner three questions in writing: Which Meridian channels would you GeoX first and why? How do you convert GeoX results into priors? What happens when experiment and model disagree? If they cannot answer without marketing copy, you do not have a partner for this stack yet.
Sharp close
Google did not ship a prettier report card. It shipped audit agents, brand signals, and globally available geo experiments that are meant to sit inside the same Meridian loop that should move money while the quarter is still alive. Operators who keep MMM as a quarterly ritual will keep optimizing last semester’s grades. Operators who treat data quality, upper-funnel signals, and causal priors as one live system will change budget on evidence the board can follow. The mix model is not a report card. Run it like a control loop, or keep losing arguments to whoever still trusts last-click.
