As tracking collapses, marketing mix modeling has become the fallback — and the two most widely adopted MMM tools were built by the two companies with the most to gain from what those models conclude. Google’s Meridian, launched in 2024, and Meta’s Robyn, launched in 2020, are both open-source, which sounds like neutrality and isn’t quite. Three measurement vendors and agency executives have told AdExchanger that Meta has dismantled its Robyn engineering team — leaving Google’s tool as the increasingly default answer to a question Google has a stake in.
01Open source is not the same as neutral
Google senior director of data science Harikesh Nair argues that open-sourcing Meridian is “crucial for building trust,” and there’s something to that — the code is inspectable in a way legacy MMM never was.
But inspectable code doesn’t mean neutral defaults. Mike Ryan of Smarter Ecommerce observes that Meridian maps “the entire Googleverse of media and data in loving, intricate detail,” which is the whole issue in one sentence. A model that represents one seller’s inventory with high fidelity and everything else coarsely will tend to find that the high-fidelity inventory works. Not through fraud — through resolution.
The commercial machinery around it is visible too. Measured CEO Nick Stoltz notes that Google pushes Meridian partly to drive Google Analytics upgrades, and that Google has tied sales-force KPIs to Meridian adoption. That is not how a company treats a neutral public good. Mutinex CEO Henry Innis puts the structural point best: “there’s a huge amount of power in setting where everybody starts in a solution.”
Amazon is doing the same thing without the pretense. Amazon Ads’ Jamie Fellows confirmed the company’s MMM aim is making “Amazon signal and media available.”
02Why marketers should care
| Your budget model was written by two of the parties it measures | Meridian is Google's, Robyn is Meta's, both open-source — which reads as neutrality and is not quite. Open code means you can inspect the assumptions. It does not mean anyone has. |
|---|---|
| The defaults are the product | Priors, channel taxonomies and starting assumptions shape an MMM's output more than most users realise, and they ship pre-set. If open-web and publisher-direct media arrive bucketed into an undifferentiated line while platform media gets granular treatment, the model has decided your allocation before you read it. |
| The field is narrowing | With Meta reportedly standing down Robyn's engineering, the practical open-source landscape converges on a Google-authored tool. Fewer independent models means fewer chances anyone checks the result. |
Measurement is being rebuilt at exactly the moment nobody can agree on what counts, and the parties writing the new tools are the parties whose media those tools evaluate.
03What marketers should do
04The bottom line
Measurement is being rebuilt at exactly the moment nobody can agree on what counts, and the parties writing the new tools are the parties whose media those tools evaluate. Open source makes that legible, not harmless. Publishers can’t out-engineer Google’s modeling team, but they can insist on being properly specified inside the model and bring independent evidence when the output looks wrong. Get your data model-ready, ask what tool is grading you, and never let someone else’s defaults be the last word on what your inventory is worth.