Most companies that spend from a million euros a year on marketing already have an agency, and most agencies now offer to add marketing mix modelling to the contract. The price is often low, sometimes folded into the media fee, and the agency already holds the spend data. So the question lands on the CMO's desk in a specific form: should the people who buy our media also be the people who tell us whether it worked? This guide is for the marketing and finance leaders asking that question, and for the ones looking past Kantar, Circana or a holding-company offer to the independent platforms that exist today.
An independent marketing mix model is one run by a company that does not sell you media and does not depend on the result. That rules out the agency that buys your media. It does not rule out research houses like Kantar or Gain Theory, which buy no media; with them the question is engagement versus product, refresh cadence and price. The practical alternatives fall into four groups:
Attribution and reporting suites like Funnel and Triple Whale now add MMM features too. They are a fifth option for digital-heavy businesses, and a good reporting layer under any of the four above.
Start with what the agency offer gets right. The agency has the media data, often the creative calendar, and a team that knows your plan. A model built there starts fast. Dentsu, WPP Media (the former GroupM) and Omnicom's Annalect all sell measurement, and the big networks increasingly build it on Google Meridian or their own engines rather than on a spreadsheet.
Two things change when the measurement sits with the media buyer.
The first is incentive. The same company earns on the media it recommends and grades the media it bought. That does not make the model wrong. It means nobody outside the agency can tell whether it is right, and a CFO asked to move budget on the back of it will ask exactly that. Many companies want the measurement to sit outside the agency for that reason, so the model survives the next pitch.
The second is portability. An agency model is usually a project deliverable: a deck, a workshop, a set of curves at a point in time. When the contract ends, the model, the structured data and the history of what it said tend to end with it. An independent platform keeps the data and the model with you, and keeps updating them.
Neither point applies to the research and effectiveness houses. Kantar, with its always-on Athena platform, and WPP's Gain Theory, a Leader in the Forrester Wave for marketing measurement services in Q1 2026, do not buy media. Nielsen's marketing mix business was sold to Circana, with the acquisition completed in August 2025, so a brief that says "alternatives to Nielsen" should name which company it means. With these houses the independence argument is already settled. What you are comparing is a consulting engagement against a product, the refresh cadence, and a price tier set for the largest advertisers.
The table sorts the options by who does the work and who ends up owning the result. It leaves out the names that fit only one market.
| Research houses | Specialist platforms | Open-source frameworks | Managed MMM platform | |
|---|---|---|---|---|
| Examples | Kantar, Gain Theory, Circana | Recast, Analytic Partners | Google Meridian, PyMC-Marketing, Meta Robyn | Odins |
| Who builds and runs the model | Their consultants | Their data scientists | Your data scientists | The vendor's team |
| Who collects and structures the data | Mostly you, with their templates | Shared | You | The vendor, offline channels included |
| Refresh | Project or subscription, varies by house | Subscription, cadence varies by vendor | As often as your team retrains it | Monthly retrain, data flowing in daily |
| What you need in-house | A project owner | An analyst to work with theirs | A data team and a data warehouse | Someone who owns the budget decision |
| Fits when | Large global budgets, multi-country studies | Large budgets with an analytics function | You already have data scientists and a warehouse | Budgets from EUR 1M to several hundred million, no data science team required |
A note on the open-source column. Meridian and PyMC-Marketing are Bayesian and actively developed; Meta has stopped developing Robyn, which is also frequentist rather than Bayesian. All three are frameworks, and a framework is not a product. The statistics are rarely what takes the time. The data foundation, the monthly update and a tool the marketing team will actually open are the work, and that is what you are buying from any of the other three columns.
On "managed versus self-serve"
Several of the platforms above are managed and Bayesian, so neither word separates vendors on its own. The separating questions are who structures the offline data, whether a person reviews the recommendation before you see it, and whether the price fits the budget being optimised.
The vendor's statistics rarely decide the choice. The answers to these five questions usually do, and our guide on how to choose a marketing mix modelling vendor goes through each in more depth.
Keep it. A reporting hub or an attribution suite is where day-to-day campaign numbers belong, and most of these tools now add saturation curves and a budget optimiser of their own, which is useful inside the digital channels they see.
What they do not answer well is the investment question: how much to spend in total next year, and how to split it between television, search, social, out-of-home and the agency retainer. That needs one model that sees offline and digital in the same place, with business drivers like price and stock alongside. The sensible stack is the reporting suite for the weekly view and an independent MMM for the budget. The two feed each other: the MMM uses the platform's conversion data as a starting belief for the digital channels, and the suite's reports carry on as before. Our guide to triangulation in marketing measurement covers how the pieces fit without double counting.
Odins is a managed marketing investment platform. We collect and structure marketing data across digital and offline channels, run a Bayesian marketing mix model, retrain it monthly, and deliver recommendations on how much to spend and where after an analyst has reviewed them. We sell no media, so we have no stake in which channel wins. The platform fits companies with a marketing budget from EUR 1M to several hundred million a year and a sales signal the model can learn from. Companies like CDON, Nettbil, Hyre and Megaflis run their marketing budgets on it. Scenarios and forecasts are self-serve and available any time; the model update and the recommendations arrive monthly.
It is not the right choice for everyone. A global advertiser running multi-country studies may want a research house. A company with its own data science team may prefer to build on Meridian or PyMC-Marketing, and our page on Odins versus in-house MMM sets out when that is the better call.
Any provider that does not buy your media: a research house such as Kantar or Gain Theory, a specialist platform such as Recast or Analytic Partners, an in-house build on Google Meridian or PyMC-Marketing, or a managed platform such as Odins. The first question to settle is whether you want an engagement or a product, and whether you have a data team.
Nielsen's MMM business now belongs to Circana. Alongside it and Kantar, the independent options are the specialist platforms, the open-source frameworks and the managed platforms described above. The main differences are cadence (monthly retrain versus periodic study), delivery (product versus engagement) and a price set for the budget being optimised rather than for the largest advertisers.
It can, and the model may be good. The difficulty is that nobody outside the agency can verify it, and the model usually leaves when the agency does. Many companies want the measurement that decides the budget to sit with a company that earns nothing from the answer.
Keep Funnel for the weekly reporting and add an independent marketing mix model for the yearly budget and the split across offline and digital. The MMM can take the platform data as a starting belief, so the two agree on the digital channels and the MMM adds what the platforms cannot see.
Rarely. Meridian and PyMC-Marketing are strong frameworks, but somebody has to build the data pipeline, set the priors, retrain the model and explain the output every month. For a team without that capacity, a managed platform is usually cheaper in total and faster to a first usable model. Our guide to how much marketing mix modelling costs puts numbers on the comparison.
Kantar and Gain Theory are London-based. Odins is based in Oslo and works with companies across the Nordics, the UK and the rest of Europe. Several of the attribution-led tools are European too, including Funnel in Stockholm. Ask any provider where the data is hosted and which markets' media data they have modelled before.
Start with the complete guide to marketing mix modeling if the method itself is new to you, read Odins versus in-house MMM if you are weighing a build, or book a demo to see an independent model run on a budget like yours.