Google's Meridian MMM Is Now Open to All Marketers — And It's Getting Serious

Google's Meridian, the open-source marketing mix model the company has been developing as its answer to Meta's Robyn, is now generally available to all marketers and data scientists. And the latest updates signal that Google is treating this as critical infrastructure, not a side project.

What's New in Meridian

The core of Meridian remains its Bayesian causal inference engine, which blends prior business knowledge with observed data to estimate the incremental impact of marketing spend. But several 2025 updates have meaningfully expanded its capabilities:

Non-media variables. Meridian now lets you include pricing, promotions, distribution changes, and other non-media factors in your model. This is a significant upgrade — without controlling for these variables, MMMs can misattribute sales lifts caused by a price cut to a media campaign that happened to run simultaneously.

Channel-level contribution priors. You can now inject your own business knowledge about expected channel contributions as Bayesian priors. This is particularly useful for channels where you have strong intuitions or existing incrementality test results that should inform the model.

Reach and frequency modeling. Unlike many MMM tools that only look at impressions or spend, Meridian accounts for reach and frequency curves — especially important for video campaigns where diminishing returns from over-frequency are a major concern.

Search data integration. Meridian incorporates Google search query volume, including organic searches, as a proxy for category demand. This helps the model distinguish between media-driven demand and organic interest.

The Partner Ecosystem

Google has certified over 20 measurement partners on Meridian, including major consultancies and analytics firms. This is a deliberate strategy to drive adoption — most brands won't build an MMM in-house, so having trained partners is essential.

Meridian vs. Robyn

The comparison to Meta's Robyn is inevitable. Robyn, which has seen over 75,000 downloads and has 32 contributors on GitHub, takes a different approach: it uses ridge regression with evolutionary algorithms to reduce human bias in model selection. A revised academic paper on Robyn's methodology was published in January 2025.

The practical differences come down to philosophy. Meridian leans into Bayesian methods and wants you to bring your business knowledge to the model. Robyn leans into automation and wants to minimize human bias. Both are legitimate approaches, and sophisticated measurement teams will likely want to run both and triangulate.

What This Means

The open-source MMM movement — driven by Meridian, Robyn, and tools like Uber's Orbit and PyMC-Marketing — is fundamentally changing the economics of media mix modeling. What used to require a six-figure consulting engagement can now be built by an in-house data science team for the cost of their time.

But "open source" doesn't mean "easy." These tools still require statistical expertise, clean data pipelines, and thoughtful model calibration. The brands getting the most value are those pairing open-source MMM with incrementality experiments for calibration — using geo-tests to validate what the model tells them.