Google and Meta are quietly rewriting the rules of ad measurement, and they’re doing it with open-source marketing mix modeling (MMM) tools that many marketers don’t even realize they’re using.
MMM is a method for measuring the relative performance of different marketing channels. It fell out of favor because it takes a long time and isn’t deterministic. But since Apple, Google and regulators around the world began cracking down on online tracking, MMM has come roaring back into fashion.
So much so that Google and Meta jumped in with their own open-source MMM tools, called Meridian and Robyn, respectively.
Now, Meridian and Robyn are reshaping the independent measurement ecosystem.
According to numerous agency execs and attribution tech vendors AdExchanger spoke with, Google and Meta have aggressively pushed their MMM tools as a new attribution standard.
Meta, however, appears to be pulling back while Google continues to push.
AdExchanger’s sources say Meta is winding Robyn down behind the scenes while Google continues to forge ahead, even tying sales KPIs to Meridian adoption.
What’s going on?
Meta released Robyn in 2020, while Google was actually a little late to the game with Meridian, which didn’t come out until 2024. And their motivations were pure, at least on paper.
“We’re seeing a significant industry shift toward transparent open-source measurement,” Harikesh Nair, Google’s senior director of data science and engineering, recently told AdExchanger in an email.
Open-sourcing Meridian is “crucial for building trust,” Nair wrote, because marketers and data scientists want to “look under the hood to verify the model’s methodology for themselves.”
As one data analyst at a travel brand told AdExchanger, the open-source aspect means Google and Meta can faithfully present campaign results as impartial and observable.
Campaign reports from Google’s Performance Max and AI Max are frustrating and ineffectual, the analyst said, because the ad placements can’t be examined and the algorithmic decision-making behind any given placement is completely obscure. But by changing the measurement lens “from a microscope to a telescope,” as they put it, MMM provides a high-level view that credits media by channel – TV, radio, out-of-home, Google, etc. – which allows marketers to feel like they have more control and transparency.
But that transparency and control can be an illusion.
Yes, marketers can customize Meridian and observe how the model works in detail, unlike within a walled garden. But its real value often still comes from exclusive data contributed by the platform.
Meridian is deeply wired into Google’s universe.
After Google open sourced its MMM in 2024, Mike Ryan, head of ecommerce insights at European consultancy Smarter Ecommerce, described Meridian as having mapped out the entire Googleverse of media and data in “loving, intricate detail.”
Case in point, last year, Google services consultancy Adswerve transitioned Alaska Airlines from a kludged-together hybrid of multiple measurement tools – “a Frankenstein’s monster sort of thing” was how Luka Cempre, Adswerve’s head of data modernization and cloud strategy, put it at the time – to Meridian only because it can tap into YouTube and Google Search data.
Meanwhile, at Google Marketing Live in May, Google announced a built-in integration of Meridian for Google Analytics customers.
“Reading the tea leaves here,” said Nick Stoltz, chief strategy officer at indie media measurement provider Measured, one reason Google has been pushing Meridian so hard is because it upgrades Google Analytics, which is still trying to shake its reliance on click-based attribution.
And although Meta’s Robyn has been outpaced by Meridian of late, Meta is making similar changes to improve how its platform performs in MMM campaign reports.
In March, Meta announced a major change to its campaign analytics, creating a new category called “engage-through attribution” and rebranding the metrics it ships to outside attribution platforms. As one member of Meta’s product marketing team said on background during a press briefing at the time, Meta previously struggled to “tease out” the value of social media when advertisers used Google Analytics.
Amazon is another example. Its MMM product, released in 2023, isn’t open source. But Amazon is upfront about why it invests in MMM. “Our goal is to make Amazon signal and also media available for advertisers who are running MMMs to incorporate in their model,” Jamie Fellows, director of measurement and data science at Amazon Ads, told AdExchanger.
Amazon’s MMM therefore isn’t a Trojan horse-esque freebie that’s smuggling platform self-attribution into MMM reports. The purpose of its MMM services – and Amazon Ads isn’t shy about saying so – is to make Amazon look good both as a media seller and as a retailer.
Where platform MMMs go now
Amazon is explicit about its self-interest, and so its MMM products and data feeds will continue as closed-source advertiser tools.
Google and Meta’s self-interest is far more entangled and convoluted, although it’s worth noting that, beginning late last year, the two have apparently begun moving in opposite directions.
Although, beginning late last year, Google and Meta have apparently begun moving in opposite directions.
According to three measurement vendors and execs from three agencies, Meta has stopped pushing Robyn.
Two sources even separately used the word “dismantled” in reference to the engineering team at Meta working on Robyn.
But Measured’s Stoltz noted that Google has a much more compelling incentive for MMM adoption, because Google Analytics is the default omnichannel analytics hub for many millions of advertisers. That’s why Google Analytics also has data integrations with Pinterest, Snap, Meta and other walled gardens that wouldn’t cut deals with Google Ads.
But Google Analytics also has “media mix modeling blind spots,” Stoltz said. Before May, Google didn’t allow cross-channel traffic analytics to flow into GA, and the same was true for linear TV ads. A sale from someone who views a cable commercial and promptly searches for and buys that product would be credited to search.
So it’s no surprise that Google is leaning on its sales force to drive Meridian adoption, according to AdExchanger’s agency and measurement exec sources. Many Google salespeople have hard KPIs tied to how many advertisers they get on Meridian.
Which isn’t necessarily a bad thing.
Google isn’t hiding that Meridian is created and preferred by Google, said Mutinex CEO Henry Innis. He added that any move away from click-based measurement in favor of MMM and more sophisticated statistical models is an improvement on the last-click standard.
The problem, Innis said, is that so many advertisers use Meridian’s off-the-shelf specs. Meridian provides a good baseline but should be customized for every business, which doesn’t always happen.
“There’s a huge amount of power in setting where everybody starts in a solution,” he said.
Marketers themselves often see Meridian as a free add-on within Google’s system and therefore don’t put much effort into it. But “it’s free in the sense that a puppy might be free, not like a free beer,” Innis said.
Those who fall in the latter camp could easily get drunk on a very pro-Google version of ad measurement.
The open-source secret
Thanks to Google’s efforts, Meridian is earning a greater share of the MMM measurement market. Yet many marketers aren’t even aware.
Numerous agencies and measurement vendors have built their proprietary offerings using Meridian source code.
Measured’s MMM model is proprietary, Stoltz said, and predates Meridian. But if the timing had lined up differently, Measured would have seriously considered using Google’s open-source code, he said.
“I suspect or know that many new measurement players are built on either Robyn or Meridian, so it’s not uncommon,” he said.
And there’s clearly precedent. Many web browser operators that are antagonistic to Chrome still develop their browser using the open-source Chromium code.
(As a side note, one agency exec observed that although Google’s salespeople are incentivized to distribute Meridian, Google doesn’t actually mind – and may even prefer – when brands don’t realize their MMM is running on Google code.)
Regardless, Meridian and Robyn are “really good entry points” for educating digital marketers about MMM, said Mike True, a co-founder and CEO of MMM startup Prescient AI, which also isn’t built on Meridian or Robyn. But more sophisticated marketers, he said, especially those that advertise and sell in offline channels, require more advanced MMM solutions.
Prescient, for instance, doesn’t treat Google as one holistic channel the way Meridian does. Instead, True said, it breaks Google down into component parts, such as YouTube, Search, PMax and Demand Gen.
Take a DTC brand with a Shopify site and a strong social presence that doesn’t sell in stores or advertise on TV. It might be fine with Meridian and/or a lightweight MMM product from a third-party provider like Northbeam or Triple Whale, True said. But when omnichannel sales and advertising come into play, configuring a proprietary MMM offering is worth the effort.
Innis, who said Mutinex’s own MMM is also built in-house, added that Google, Meta and Amazon have greatly improved their data feeds for MMM solutions, which is why media owners should only be data contributors to MMM systems. There’s no need for them to operate MMM code.
In a dream hypothetical scenario, he said, there would be disclosure regulations when an MMM provider is aligned with a particular player in the ad supply chain, akin to how corporate director duties require disclosure or recusal in situations where an entity stands to benefit.
In Innis’ view, it’s not a coincidence that an early wave of MMM providers emerged in the early 2000s just after the Enron scandal. “Why was that?” he said. “Because most capital expenditures required controls and reporting that were at arm’s length to people who may be executing those supply chains.”
The lesson, in other words, is that measurement is most trustworthy when it isn’t provided by the same players who stand to benefit from the results.
And the same goes for MMM. It’s an improvement over click-based attribution, but don’t expect it to be the savior of transparent, independent measurement – at least not without first checking to see what code is running under the hood.
