In September 2026, Google upgraded Meridian, its free, open-source marketing mix modeling tool, with AI-guided model-building, upper-funnel brand signals for channels like TV and out-of-home, and globally available causal geo-testing. The upgrade lowers the technical bar for building an independent measurement model outside your agency’s reporting — which means “we don’t have the resources to verify that” is a weaker excuse than it was a year ago.

For most of the last decade, the honest answer to “how do we know this performance number is real” has been: you don’t, not independently. Marketing mix modeling required a data science team most brands didn’t have, so clients read whatever model their agency or a paid vendor handed back and asked follow-up questions from a position of limited leverage. That gap is closing, and it’s worth understanding why before your next quarterly review.

Google’s mid-September update to Meridian — its open-source MMM platform — adds agentic AI guidance that walks a marketer through data-quality checks, flags likely errors, and helps construct the model in something closer to plain language rather than requiring a specialist to hand-code it.^[1] It also folds in upper-funnel brand signals, like branded search query volume, to help attribute channels such as television and out-of-home that have always been harder to tie to sales than a last-click digital ad. And Meridian’s GeoX function — which runs causal, incrementality-based geo experiments rather than relying purely on correlation — is now available worldwide, not in limited beta.[^1][^2]

Google’s own framing of the problem is worth repeating: measurement shouldn’t just be “a record of where budget went.” A report that tells you what happened after the money is already spent isn’t performance management — it’s an expense summary with a nicer chart. Where MMM is actually useful is answering the harder, forward-looking question of whether a given channel is driving real incremental growth, or just getting credit for demand that existed anyway.^[2]

None of this obligates you to use Google’s tool specifically, and it isn’t an endorsement of one vendor over another — the point is what its existence changes. When a credible, no-cost, causally grounded measurement option is one download away, “trust our attribution model, we don’t have bandwidth to show our work” stops being a reasonable position for an agency to hold at a quarterly business review. You don’t need to become a data scientist. You need a scorecard structure, built before the relationship gets comfortable, that specifies how performance claims get shown, not just reported.

A note on role: We don’t run your measurement or replace your agency’s analytics team. What we advise on is the scorecard and accountability structure — what your agency should be required to show, and how often — so a performance claim is something you can verify, not something you take on faith.

---