Marketing mix modeling statistics at a glance
| Category | Statistic | Source |
|---|---|---|
| Adoption | 60% of US advertisers were already using marketing mix models; 58% of non-users were considering it | Kantar via Google, 2023 |
| Adoption | 87% say MMM matters to their organisation, but only 28% are very effective at acting on it | HBR Analytic Services, 2026 |
| Adoption | Between 67% and 76% of buy-side decision makers use attribution, incrementality tests or marketing mix models, from 400+ US planning and analytics leaders | IAB and BWG Global, 2026 |
| Investment | 40% name MMM their top measurement investment for the next 12 months | Supermetrics, 2026 |
| Trust | 27.6% of US marketers call MMM the most reliable methodology; 19.4% say multi-touch attribution | EMARKETER and TransUnion, 2025 |
| Trust | Up to 75% of advertisers say advanced measurement falls short on rigour, timeliness, trust and efficiency | IAB and BWG Global, 2026 |
| Action | 80% of MMM leaders always or frequently act on model output; 14% of laggards do | HBR Analytic Services, 2026 |
| Barriers | 47% cite data quality problems in MMM inputs; 46% cite siloed data; 45% cite thin in-house expertise | HBR Analytic Services, 2026 |
| Coverage | Gaming is underrepresented in MMM for 77% of marketers, commerce media for 50%, creator for 48% | IAB and BWG Global, 2026 |
| Accuracy | Non-experimental methods overstated ad lift threefold against 663 Facebook randomised tests | Gordon, Moakler and Zettelmeyer, 2023 |
| Data cost | Google Meridian asks for at least 104 weeks of weekly geo data, or three years for national models | Google, 2026 |
| Open source | Meridian has 1,540 GitHub stars in 32 months; Robyn has 1,516 in 77 months | GitHub, September 2026 |
| Context | Only 32% of marketers measure traditional and digital spend holistically; 23% in Europe | Nielsen, 2025 |
How many marketers use marketing mix modeling?
Marketing mix modeling is used by roughly two thirds to three quarters of large advertisers in 2026, and the most recent survey to put a single number on it, Kantar's 2023 research for Google, found 60% of US advertisers already running models.
| Statistic | Source |
|---|---|
| 60% of US advertisers were using marketing mix models, and 58% of non-users were considering it, from 300 US brand, full funnel and performance marketers surveyed between 22 February and 24 March 2023. The most quoted MMM adoption figure, and now three years old, so treat it as a floor. | Kantar, Measurement Solutions Research, US 2023, cited by Google |
| Between 67% and 76% of buy-side decision makers currently use incrementality tests, attribution analysis or marketing mix models, from a survey of more than 400 senior planning and analytics leaders at US brands and agencies. | IAB and BWG Global, State of Data 2026, February 2026 |
| 87% say it is important to their organisation to use MMM for data-driven insight, from 547 marketing-involved HBR audience members surveyed in October 2025. | HBR Analytic Services, March 2026 |
| 40% name marketing mix modeling their top measurement investment for the next 12 months, ahead of AI creative testing (37%) and A/B testing (36%), across 435 marketers in five countries. | Supermetrics, 2026 Marketing Data Report |
| 61% of respondents say multichannel and cross-channel consumer journeys have changed the way their organisation approaches MMM. | HBR Analytic Services, 2026 |
The adoption debate is over, and it was over before most vendors noticed. When two thirds of the buy side already runs some form of advanced measurement, "should we do MMM" stops being a question worth a panel. What still splits the market is whether MMM runs next to experiments and attribution or on its own. A model that nothing else checks keeps every blind spot it started with, and nobody notices when it drifts.
Why is marketing mix modeling coming back?
Marketing mix modeling returned because user-level measurement stopped being reliable while the pressure to prove media impact rose: 73% of marketers are under increasing pressure to demonstrate the business impact of their media channels (HBR Analytic Services, 2026).
| Statistic | Source |
|---|---|
| 73% of respondents agree their organisation is under increasing pressure to demonstrate the business impact of using various media channels. | HBR Analytic Services, 2026 |
| 68% name a greater focus on return on investment as the top factor driving their organisation to use MMM. | HBR Analytic Services, 2026 |
| Google decided on 22 April 2025 to keep third-party cookie choice as it is in Chrome, ending a five-year deprecation plan that much of the MMM pitch was built on. | Google Privacy Sandbox, April 2025 |
| Only 32% of marketers measure traditional and digital media spend holistically, falling to 23% in Europe and 29% in Latin America, from 1,400 marketers surveyed in February and March 2025. | Nielsen, 2025 Annual Marketing Report |
| 40% of marketers say proving ROI across channels is their biggest challenge, and 45% say they are still struggling with measurement. | Supermetrics, 2026 |
| Martech fell to 19.4% of marketing budget in 2026 from 26.6% in 2021, while paid media rose to 31.4%, across 401 CMOs surveyed by Gartner. | Gartner 2026 CMO Spend Survey, via Chief Marketer |
Notice what did not happen. Google cancelled cookie deprecation in April 2025 and MMM adoption kept climbing anyway, which tells you the cookie story was always a convenient headline rather than the cause. The real driver is in the Gartner split: money moved out of martech and into media, and nobody wants to explain a bigger media line with a tool that only sees the clickable part of it. If you want the click-level side of that argument, our marketing attribution statistics page covers it.
Can marketers act on MMM results?
Most cannot. Only 28% of organisations say they are very effective at converting MMM insights into timely and impactful action, against 87% who say MMM matters to them, in the HBR Analytic Services survey of 547 respondents published in March 2026.
| Statistic | Source |
|---|---|
| 28% of respondents say their organisation is very effective at converting MMM insights into timely and impactful action. | HBR Analytic Services, 2026 |
| The survey splits respondents into 22% leaders, 36% followers and 42% laggards, based on how effective they are at both producing MMM insight and acting on it. | HBR Analytic Services, 2026 |
| 80% of leaders always or frequently act on MMM insights, against 50% of followers and 14% of laggards. | HBR Analytic Services, 2026 |
| 61% of leaders cite better data-driven decision making as a business outcome improved by MMM, against 44% of followers and 29% of laggards. | HBR Analytic Services, 2026 |
| 47% cite data quality issues in MMM inputs as a current technological challenge. | HBR Analytic Services, 2026 |
| 46% cite difficulty integrating siloed data from multiple sources. | HBR Analytic Services, 2026 |
| 45% cite limited in-house expertise to interpret and apply MMM outputs, while only 34% are doing anything about it. | HBR Analytic Services, 2026 |
| 47% cite slow internal processes that delay data-driven changes, while only 27% are taking steps to speed them up. | HBR Analytic Services, 2026 |
| 72% say MMM insights are only selectively accessible to some business teams rather than shared across the organisation. | HBR Analytic Services, 2026 |
| 54% update or refine their MMM design quarterly or more often, rising to 65% among leaders and falling to 41% among laggards. | HBR Analytic Services, 2026 |
This is the section I would hand to anyone about to sign an MMM contract. The gap between the 45% who lack the expertise to read a model and the 34% doing something about it is the whole problem in two numbers: companies buy the model and skip the translation layer. Leaders act on their models 5.7 times more often than laggards, and almost none of the difference is statistical sophistication. It is a quarterly refresh cadence, a shared vocabulary, and someone whose job is to turn a coefficient into a budget line.
How does MMM compare with attribution and incrementality testing?
Marketers rate marketing mix modeling as their most reliable measurement method more often than any alternative: 27.6% pick MMM against 19.4% for multi-touch attribution, a 42% relative edge, in EMARKETER and TransUnion's July 2025 survey of US brand and agency marketers.
| Statistic | Source |
|---|---|
| 27.6% of US brand and agency marketers rate MMM the most reliable measurement methodology, against 19.4% for multi-touch attribution and 18.9% for unified measurement. | EMARKETER and TransUnion, July 2025 |
| 52% of US brand and agency marketers use incrementality testing and experiments to measure campaigns. | EMARKETER and TransUnion, July 2025 |
| Up to 75% of advertisers report that advanced measurement approaches fail to deliver the rigour, timeliness, trust and efficiency they need. | IAB and BWG Global, 2026 |
| 77% of marketers say gaming is underrepresented in MMM, 50% say commerce media, 48% say creator and influencer marketing, and 46% say traditional media such as radio, print and out of home. | IAB and BWG Global, 2026 |
MMM wins the reliability vote by 8 points over attribution, which sounds decisive until you notice that barely a quarter of marketers name any method at all as the most reliable one. Nobody trusts anything much. My own answer for 2026 is that MMM sets the guardrails and geo experiments settle the arguments, with attribution left to handle week to week tuning. The channel coverage numbers are the uncomfortable part: if gaming and commerce media are missing from three quarters of models, the model is not measuring the mix.
How accurate is marketing mix modeling?
Marketing mix modeling has no published accuracy benchmark, and the best available evidence on the wider family of non-experimental methods is unflattering: across 663 large-scale Facebook experiments, non-experimental estimates overstated advertising lift by roughly threefold (Gordon, Moakler and Zettelmeyer, Marketing Science, 2023).
| Statistic | Source |
|---|---|
| Across 663 large-scale randomised experiments at Facebook, with access to more than 5,000 user-level features, median true lift was 29%, 18% and 5% for upper, middle and lower funnel outcomes. The best non-experimental method returned 83%, 58% and 24% for the same campaigns. | Gordon, Moakler and Zettelmeyer, Marketing Science 2023 |
| Google Meridian asks for a minimum of two years of weekly data for geo-level models and three years for national models, which is 104 weekly observations per geo before a model can run. | Google Meridian documentation, 2026 |
| C-level leaders who rated marketing mix modeling highly were more than twice as likely to exceed revenue goals by 10% or more, across 870 senior marketing leaders and 111 C-level leaders surveyed between October 2022 and February 2023. Dated, and a correlation rather than a causal claim. | Google and Deloitte Ads Measurement Market Research, 2023 |
| The IAB puts the value of better AI-powered measurement at about $26.3 billion across total US ad spend, plus $6.2 billion in productivity value. | IAB and BWG Global, 2026 |
| Half the buy side already scales AI inside advanced measurement, but the split is lopsided: 69% of analytics teams against 30% of planning teams. AI clauses appear in about 40% of brand-agency contracts today, with 70% to 80% expected within one to two years. | IAB and BWG Global, 2026 |
The Facebook paper is about propensity matching and double machine learning rather than MMM specifically, and people who sell MMM will tell you that. They are right, and it still matters, because MMM shares the same underlying problem: you are reading effects off data that an ad platform's targeting already selected. That is why every serious modern MMM stack calibrates against geo experiments instead of arguing with them. Treat any MMM channel ROI that has never been checked against a holdout as a hypothesis.
Who is winning open source MMM, Meridian or Robyn?
Google Meridian is pulling ahead of Meta Robyn in open source MMM: Meridian's GitHub repository has gathered 1,540 stars since January 2024, about 2.5 times Robyn's rate of accumulation since April 2020, as of 20 September 2026.
| Project | Stars | Forks | Contributors | Repo created |
|---|---|---|---|---|
| google/meridian | 1,540 | 298 | 22 | 31 January 2024 |
| facebookexperimental/Robyn | 1,516 | 434 | 33 | 30 April 2020 |
Source: GitHub API, read 20 September 2026. Robyn's latest tagged release is v3.12.0, published in December 2024; Meridian's repository was last updated on 19 September 2026.
| Statistic | Source |
|---|---|
| Google announced Meridian on 7 March 2024 and made it generally available on 29 January 2025 with more than 20 trained and certified measurement partners. | Google, January 2025 |
| Meridian uses Bayesian causal inference and needs no individual identifiers or cookies, which is the reason it can be run on aggregated geo data. | Google, March 2024 |
| Google added agentic AI features to Meridian on 14 September 2026, aimed at auditing input data quality and guiding model building. | Marketing Dive, September 2026 |
| Meta describes Robyn as an experimental, semi-automated open source MMM package using ridge regression and evolutionary hyperparameter optimisation. | Meta, facebookexperimental/Robyn |
Star counts are a vanity metric and I am using them anyway, because there is no better public signal for which open source MMM engineers are picking up. Robyn has more forks and contributors, which is what six extra years buys you. Meridian has the momentum, a certified partner network and a release cadence Robyn has not matched since December 2024. Choose on whether you have two years of clean weekly geo data, not on the logo.
How we calculated the original numbers
Four figures on this page do not appear anywhere else. Here is the arithmetic, so you can check it or change the inputs.
- 32%: the share of MMM believers who can act. HBR Analytic Services found 87% of respondents say MMM is important to their organisation and 28% say they are very effective at converting MMM insight into timely action. 28 / 87 = 0.322. Same survey, same 547 respondents, so the two bases match. Read it as roughly one organisation in three, not as a precise rate.
- 5.7x: leaders against laggards on acting. 80% of leaders always or frequently act on MMM insights; 14% of laggards do. 80 / 14 = 5.71. Leaders are 22% of the sample and laggards 42%, so the larger group is the one that mostly ignores its own models.
- 42%: MMM's reliability premium over multi-touch attribution. EMARKETER and TransUnion found 27.6% of US marketers rate MMM the most reliable methodology against 19.4% for MTA. The absolute gap is 8.2 points; 27.6 / 19.4 = 1.42, so 42% more marketers pick MMM.
- 2.5x: Meridian's star rate against Robyn's. The google/meridian repository was created on 31 January 2024 and had 1,540 stars on 20 September 2026, which is 31.6 months and 48.7 stars a month. facebookexperimental/Robyn was created on 30 April 2020 and had 1,516 stars over 76.7 months, or 19.8 a month. 48.7 / 19.8 = 2.46. The repository was created in January 2024 but only opened to everyone in January 2025, so 2.5x is a floor rather than a best case. Stars measure attention, not installs.
What the 2026 numbers actually say
MMM won the argument and lost the follow-through. Two thirds to three quarters of the buy side runs it, 87% say it matters, and 28% can act on it. Every barrier in the HBR data is organisational rather than statistical: dirty inputs, siloed data, no in-house reader, slow approval. Nobody in that survey complained about the maths.
Running one method is not enough. Between 67% and 76% of the buy side uses incrementality testing, attribution or MMM, and up to 75% say those approaches still fall short on rigour, timeliness, trust and efficiency (IAB and BWG Global, 2026). The coverage gaps say the same thing from the other side: 77% of marketers report gaming underrepresented in their models, 50% commerce media and 48% creator spend.
The next constraint is data, not modelling. Meridian wants 104 weeks of clean weekly geo data before it produces anything useful, and 47% of marketers already name input data quality as their live problem. Open source made the model free. It did not make the two-year data history free, and that is where most MMM projects will stall in 2027. For the channel-level backdrop, our digital vs traditional marketing statistics page has the spend split MMM is trying to explain.
FAQ
What percentage of marketers use marketing mix modeling?
Between 67% and 76% of buy-side decision makers currently use incrementality tests, attribution analysis or marketing mix models, according to the IAB and BWG Global's State of Data 2026 survey of more than 400 senior planning and analytics leaders. The narrower MMM-specific figure comes from Kantar, which found 60% of US advertisers using marketing mix models in a survey of 300 US marketers fielded in early 2023, with 58% of non-users considering it. That Kantar number is three years old and is best read as a floor.
Is MMM more reliable than multi-touch attribution?
Marketers say yes, by 42%. EMARKETER and TransUnion found 27.6% of US brand and agency marketers rate MMM the most reliable measurement methodology in July 2025, against 19.4% for multi-touch attribution and 18.9% for unified measurement. The caveat matters: barely a quarter of marketers name any method as most reliable, so MMM is leading a weak field rather than winning a strong one.
Why do MMM projects fail to change budgets?
Because the barriers are organisational. HBR Analytic Services (2026) found 47% of organisations blame data quality in MMM inputs, 46% blame siloed data, 45% lack the in-house expertise to interpret model output, and 47% report internal processes too slow to act on it. Only 28% say they are very effective at converting MMM insight into timely action, and 72% keep those insights accessible to a few teams rather than sharing them.
How much data does an MMM need?
Google Meridian asks for a minimum of two years of weekly data for geo-level models and three years for national models, per its 2026 documentation. That is 104 weekly observations per geo. Monthly data works only with three years or more of history. This requirement, not licence cost, is the real entry barrier for most advertisers.
How accurate are marketing mix models?
No published benchmark exists for MMM specifically. The closest evidence is Gordon, Moakler and Zettelmeyer's 2023 Marketing Science study of 663 large-scale Facebook experiments, where non-experimental methods returned median lifts of 83%, 58% and 24% by funnel stage against true experimental lifts of 29%, 18% and 5%. Calibrating MMM against geo experiments is the standard answer to this problem.
Meridian or Robyn: which open source MMM should I use?
Google Meridian has the momentum. Its GitHub repository holds 1,540 stars gathered since January 2024, roughly 2.5 times Meta Robyn's rate since April 2020, and it reached general availability in January 2025 with more than 20 certified measurement partners. Robyn has more forks (434) and contributors (33), but its last tagged release was v3.12.0 in December 2024.
Sources
- Harvard Business Review Analytic Services, Bridging the Marketing Mix Modeling Actionability Gap (March 2026, survey of 547 respondents fielded October 2025, sponsored by Google)
- IAB and BWG Global, State of Data 2026: The AI-Powered Measurement Transformation (February 2026)
- PPC Land, reporting the IAB State of Data 2026 findings (February 2026)
- EMARKETER, FAQ on incrementality: How to prove your ads actually work in 2026 (2026)
- EMARKETER, FAQ on media mix modeling (2026)
- EMARKETER, Marketing mix modeling topic page (2026)
- EMARKETER, Media Mix Modeling Trends 2026 (2026)
- Supermetrics, The 2026 Marketing Data Report (2026, 435 respondents)
- Nielsen, 2025 Annual Marketing Report press release (May 2025, 1,400 marketers)
- Google, Empowering your team to build best-in-class MMMs (March 2024, citing Kantar 2023 and Google/Deloitte 2023)
- Google, Meridian is now available to everyone (January 2025)
- Google, Meridian documentation: collect data (2026)
- Google Privacy Sandbox, Next steps for Privacy Sandbox and tracking protections (April 2025)
- GitHub, google/meridian repository (read September 2026)
- GitHub, facebookexperimental/Robyn repository (read September 2026)
- Gordon, Moakler and Zettelmeyer, Close Enough? A Large-Scale Exploration of Non-Experimental Approaches to Advertising Measurement (Marketing Science, 2023)
- Gartner 2026 CMO Spend Survey, reported by Chief Marketer (2026, 401 CMOs)
- Marketing Dive, Google upgrades Meridian with agentic AI, upper-funnel capabilities (14 September 2026)
Methodology. Numbers were collected in September 2026 and checked against the original publisher's report, press release, documentation or repository, not against other statistics roundups. The IAB State of Data 2026 report sits behind a registration wall, so its figures are cited through PPC Land's reporting, which names the survey, the sample and the research partner. Vendor claims without a stated sample or method were excluded. Numbers older than 24 months are flagged as dated in the text, and every computed ratio is shown with its arithmetic so you can disagree with it. See the rest of the series on the statistics hub. Last updated 20 September 2026.