The Return of Marketing Mix Modelling: Meridian, Robyn and Beyond

Marketing mix modelling (MMM) was once the preserve of large consumer goods companies with big budgets and patient analysts. It is now one of the most discussed topics in measurement. In February 2026, the IAB reported that 60-75% of advanced measurement users say their methods fall short on rigour, timeliness, trust and efficiency, and launched a multi-year industry initiative that includes modernising MMM ¹. In a 2025 EMARKETER and TransUnion survey, 46.9% of US marketers said they planned to invest more in MMM over the next year, and MMM was the top answer when marketers were asked which methodology was most reliable ².

This article explains why MMM has returned, how the modern open-source versions work, what data they need, and how they fit alongside attribution and experiments.

Why MMM is back

Three pressures have pushed marketers back towards modelling.

• Signal loss. Privacy regulation, declined consent and browser restrictions make it harder to track individuals from ad exposure to purchase. MMM works on aggregated data, such as weekly spend and sales by region, so it does not depend on user-level tracking.
• Channel coverage. User-level attribution struggles with TV, out-of-home, podcasts and other channels without clicks. MMM can include them alongside digital media.
• Free, open-source tools. Google and Meta have both released open-source MMM packages, lowering the cost of entry for mid-sized advertisers.

The IAB's 2026 work also highlights a gap: MMM is only as good as the channels it includes. MarTech reported that in the IAB's State of Data 2026 research, 77% of marketers said gaming was underrepresented in their mix models and 50% said commerce media was significantly overlooked ³.

Google Meridian and Meta Robyn compared

Google Meridian became available to all marketers and data scientists globally in January 2025. It uses Bayesian causal inference, combining prior knowledge with new data, and can use the results of incrementality experiments as priors. It also models reach and frequency for video and can use Google Query Volume to separate the effect of paid search from underlying demand ⁴. By June 2026, Google said thousands of advertisers had used Meridian results, and it had added a no-code Scenario Planner for testing budget options and Meridian Studio, a Google Cloud product for running the model at scale ⁵. In September 2026, Google made Meridian GeoX, an open-source library for geo experiments whose results can feed into an MMM, generally available, and added support for brand signals such as branded search volume ⁶.

Meta Robyn is described by Meta Marketing Science as an "experimental, AI/ML-powered and open sourced" MMM package. It uses ridge regression and a multi-objective evolutionary algorithm to explore adstock rates and saturation curves, with the Nevergrad library handling hyperparameter optimisation. The core package is in R, with a Python API in beta ⁷. Its documentation includes calibration against lift experiments and a budget allocator that recommends how to shift spend across all channels, not only Meta's ⁸.

	Google Meridian	Meta Robyn
Approach	Bayesian causal inference 4	Ridge regression with evolutionary hyperparameter search 7
Built by	Google 4	Meta Marketing Science; R package with Python API in beta 7
Experiment calibration	Experiments as priors; GeoX for geo tests 46	Calibration with lift studies 8
Distinctive features	Reach and frequency, Google Query Volume, Scenario Planner 45	Automated hyperparameter tuning, budget allocator 8

Both are free and open source, but neither is a turnkey product. Each needs analysts who understand the modelling choices.

How modern MMM works

At its core, an MMM is a regression model that explains a business outcome, such as sales or leads, using media activity and other factors over time. Modern versions add a few important ideas:

1. Adstock (carryover). Advertising effects decay over time rather than stopping when the campaign ends.
2. Saturation (diminishing returns). Each extra pound or dirham spent in a channel eventually delivers less than the last.
3. Control variables. Price, promotions, seasonality, competitor activity and macro factors are included so the model does not credit media with sales it did not cause.
4. Calibration. Results from experiments are used to anchor the model's estimates of channel effects.
5. Scenario planning. The fitted model is used to test how outcomes would change under different budgets.

What data you need

Google's Meridian documentation gives clear guidance that applies to most MMMs ⁹:

• Weekly data is best practice. Daily data slows the model; monthly data can produce unstable results.
• History: at least two years of weekly data for geo-level models, and three years for national-level models.
• Geography: geo-level data, such as regions or states, is preferred; national data is an acceptable alternative.
• Required inputs: a KPI (such as revenue or conversions), media exposure for each paid channel (impressions, clicks or spend), media spend, and control variables.
• Optional inputs: revenue per KPI, population by region, organic media and Google Query Volume, though leaving out query volume may bias estimates.

For brands operating across EMEA, regional data often already exists in sales and finance systems. The hard part is usually consistency: aligning spend data from many platforms and agencies to the same weeks, regions and definitions.

MMM, attribution and experiments together

No single method answers every question. As one participant in the IAB initiative put it, "There's no single methodology that can answer every measurement question" ¹. A practical approach uses each method for what it does best:

• MMM for strategic budget allocation across all channels, quarterly or monthly.
• Platform attribution for day-to-day campaign optimisation within a channel.
• Experiments (geo tests, conversion lift, holdouts) to establish causal effects and calibrate the model.

Google's own guidance describes integrating Meridian with incrementality testing to validate and refine the model's signals ⁵. The aim is a loop: experiments improve the model, the model guides budgets, and attribution handles daily optimisation.

There is also a governance question. When a media platform builds the model, some practitioners ask whether it can be neutral about the channels it sells. PPC Land's 2026 analysis of Robyn noted that this debate remains open, even where there is no evidence of bias in the calculations ⁸. Running the model independently of any media seller, or at least reviewing its assumptions, addresses the concern.

Key takeaways

• MMM has returned because of signal loss, the need to measure offline channels, and free open-source tools ¹⁴.
• Google Meridian is Bayesian and focuses on reach, frequency and search demand; Meta Robyn uses ridge regression with automated tuning ⁴⁷.
• Expect to need two to three years of weekly data, ideally by region ⁹.
• Calibrating MMM with experiments, such as Google's Meridian GeoX, improves trust in its results ⁶.
• Keep the model independent, or at least transparent, when platforms supply both the tool and the media.

What this means for your brand

Start with a data audit: can you assemble two to three years of weekly spend and sales data by region, with consistent channel definitions? If not, fix that first, because the model cannot be better than its inputs. Choose a tool that matches your team's data science skills, or work with a partner who can run either. Plan at least one experiment per year in your largest channels to calibrate the model. Finally, use MMM for budget decisions, not as a replacement for the attribution data that runs your daily campaigns.

Sources

1. IAB Announces Project Eidos: A Comprehensive, Interdisciplinary Initiative to Fundamentally Modernize Ad Measurement - IAB, February 2026
2. MMM, incrementality, and other measurement trends that will define 2026 - EMARKETER, November 2025
3. 75% of marketers say their measurement systems are falling short - MarTech, February 2026
4. Google opens Meridian marketing mix model - PPC Land, January 2025
5. Marketing mix modelling with Meridian - Think with Google, June 2026
6. Drive profitable growth with new data and measurement tools - Google, September 2026
7. Robyn - Meta Marketing Science on GitHub, 2026
8. Meta's Robyn: who really benefits when a platform builds your MMM? - PPC Land, April 2026
9. Collect and organize your data - Google for Developers, 2026

Figures and product details are as reported by the sources listed, at the time of writing. Platforms change quickly, so check the latest documentation before acting on any single number.