Marketing Mix Modelling and Google Meridian

Olamide Sule, founder of Dolphin Analytics: a digital analytics expert based in London delivering solutions for agency and in-house clients.

Most teams ask us about marketing mix modelling for the same reason: their attribution stopped reconciling, and paid channels started claiming credit the sales figures could not back up. We rebuild measurement for agency and in-house teams most weeks, so this is the explainer we would give a measurement lead weighing whether MMM and Google Meridian belong in a post-cookie stack. It covers what MMM actually is, how it differs from the attribution you already run, where Google Meridian fits, and when the method is worth the effort.

What is marketing mix modelling?

Marketing mix modelling (MMM) is a top-down statistical technique that estimates how much each marketing channel, alongside factors like price, seasonality and promotions, contributed to sales or another outcome. It works from aggregate historical data, weekly spend and results by channel, so it needs no cookies, user IDs or individual tracking. That makes it a measurement method that survives signal loss, which is why interest has climbed as user-level tracking has eroded from consent rules, browser restrictions like Apple’s Intelligent Tracking Prevention, and ad blockers.

The technique is not new. Econometricians have used regression to relate marketing spend to sales for decades, and large consumer-goods brands have run MMM in-house or through agencies since long before digital tracking existed. What changed is who needs it. When cookies and pixels measured most of the journey, digital teams could lean on user-level attribution and leave MMM to the big spenders. As that signal fades, aggregate modelling has become relevant to a much wider set of marketers.

A model looks at how spend on each channel moves with sales over time, while controlling for the things that also move sales: price changes, promotions, seasonality, even weather or competitor activity. The output is a contribution estimate per channel, a return-on-investment figure, response curves that show diminishing returns as spend rises, and budget scenarios you can plan against. It answers a planning question, not a tracking one.

How is marketing mix modelling different from attribution?

Marketing mix modelling is top-down and multi-touch attribution is bottom-up, so they answer different questions from different data. MMM works from aggregate spend and results, needs no cookies, and tells you how to split a budget across channels, including offline and brand activity that user tracking never sees. Attribution works from individual user journeys stitched together with cookies and IDs, and tells you which touchpoint preceded a specific conversion. One is a map of where money should go; the other is a receipt for where a single sale came from.

	Marketing mix modelling (top-down)	Multi-touch attribution (bottom-up)
Data source	Aggregate weekly spend and outcomes	User-level events, cookies, IDs
Needs cookies or IDs	No	Yes
Granularity	Channel and campaign level	Individual touchpoints
Sees offline and brand	Yes	No
Survives signal loss	Yes	Degrades as tracking erodes
Speed	Periodic, weeks to quarters	Near real-time
Best question	How should I split budget across channels?	Which touch drove this specific conversion?

When we audit a measurement setup, the most common mistake is treating one of these as the whole truth. Attribution overcredits the channels it can see and ignores everything it cannot, which is part of why platform dashboards routinely overstate their own contribution. We cover that specific gap in why Meta Ads overreports conversions. MMM corrects for it at the aggregate level, but it cannot tell you which keyword or creative did the work. The strongest measurement setups triangulate: MMM for budget allocation, attribution for tactical optimisation, and incrementality experiments to calibrate both.

What is Google Meridian?

Google Meridian is Google’s free, open-source marketing mix modelling library, released for anyone to use in early 2025. It is a Bayesian MMM framework written in Python that estimates channel contribution and return on investment from aggregate data, and it can fold in reach and frequency data plus calibration from incrementality experiments. Google positions it as the replacement for its older LightweightMMM project.

Being Bayesian matters in practice. A Bayesian model lets you set priors, your existing beliefs about how a channel performs, and updates them with the data, which helps when history is short or a channel’s spend has barely varied. Meridian also supports geo-level modelling, so it can learn from differences across regions rather than national totals alone, and it produces the response curves and budget-optimisation outputs a planning team needs.

The catch is that Meridian is code, not a dashboard. It needs a data scientist or a modelling-literate analyst to set up, validate and interpret, and it needs enough clean history to learn from: Google recommends two to three years of weekly data. Meta publishes a comparable open-source MMM, Robyn, if you want to compare approaches. Both are free to run; neither is free to operate.

When is marketing mix modelling worth it?

MMM is worth the effort when you spend meaningfully across several channels, run real offline or brand activity, and have genuine cross-channel budget decisions to make. If most of your marketing is one or two digital channels and your questions are tactical (which campaign, which audience, which creative), attribution and platform reporting will serve you better and faster. MMM is a budget-planning tool, so it pays off where budget-planning stakes are high.

It is a poor fit in three situations. Small or single-channel spend gives the model too little variation to learn from. Questions that need per-user or per-campaign detail sit below what an aggregate method can resolve. And thin or messy data history, less than a couple of years, or numbers that shift every time someone re-tags the site, will produce a confident model built on sand.

Budget the running cost honestly. You need either a data science resource or a commercial vendor, a consistent data pipeline feeding the model, and a cadence: most teams refresh the model quarterly and calibrate it against geo-lift or holdout experiments so the estimates stay honest. MMM is not a one-off report; it is a measurement practice you maintain.

Marketing mix modelling tools

The tools split into open-source libraries and commercial platforms. Google Meridian and Meta’s Robyn are the two best-known open-source options, both free code you run yourself; PyMC-Marketing is a third, for teams who want a general Bayesian framework to build on. Commercial platforms such as Sellforte, Nielsen and Analytic Partners wrap the modelling in managed software and consulting, which removes the build work in exchange for a licence and services fee.

The choice is less about the tool and more about who operates it. Open-source libraries cost nothing to license but need in-house modelling skill and clean data plumbing. Commercial platforms cost money but hand you the model, onboarding and support. Either way, the model reflects the data you feed it, which is the part most teams underestimate.

Where MMM fits with the measurement you already have

Marketing mix modelling is only as good as the data underneath it, so getting that foundation right comes before any modelling decision. A model built on inconsistent tracking, double-counted conversions or spend that never reconciled against real sales will produce confident, wrong budget advice. Clean event data and numbers that match what the business actually booked are what MMM depends on, and rebuilding that foundation is the measurement work we do for agency and in-house teams most weeks.

The payoff of getting measurement right is finding value your reporting was hiding. On one private aviation engagement, rebuilding attribution traced a $767,500 charter enquiry back to a single organic search visit the client’s reporting had missed entirely. That is a bottom-up example, but the principle holds for MMM too: the method only earns its keep once the inputs reconcile.

Before you invest in MMM or Meridian, it is worth checking whether your current measurement reconciles at all. If you are not sure, start with the tell us what’s broken, or book a call. For the wider picture of how we turn clean data into decisions, see our marketing analytics work.

FAQ

What is marketing mix modelling? It is a top-down statistical technique that estimates how much each channel, plus factors like price and seasonality, contributed to sales, using aggregate historical data instead of cookies.

Is Google Meridian free? Yes. It is a free, open-source Python library; the cost is the data science skill and the clean data needed to run it well.

Marketing mix modelling versus attribution: MMM is top-down and needs no cookies; attribution is bottom-up and tracks individual journeys. They answer different questions, and strong teams run both.

Do I need marketing mix modelling? Only if you have meaningful spend across several channels, real offline or brand activity, and two to three years of clean data. Otherwise, fix your attribution and platform reporting first.