Ask three marketing mix modelling vendors for a price and you get three things that do not compare: a software subscription, a project fee and a framework that is free to download. All three can be right. None of them is the cost of marketing mix modelling. The cost is what you pay over the first two years to have a model that is fed, retrained and used, and the largest parts of that rarely appear in the quote.
As a planning figure, budget around USD 50,000 a year for the model itself, for one market in a typical setup. Published prices run from under half that for self-serve software to several times it for enterprise programmes. Then add three lines that most quotes leave with you: the data work (the integrations, and the monthly structuring of TV, radio and other offline media), the upkeep (retraining and checking the model) and the people who turn the output into budget decisions. Compare the total over two years, and ask every vendor which of the four lines their price covers. The USD 50,000 often looks the same whether the data work is in it or not.
Every route to MMM carries the same four costs. What changes is who carries each one, and whether it is in the price you are shown.
The licence, the project fee or the framework. This is the line vendors quote and the line buyers compare. On its own it says little about the total.
Getting every channel into one dataset the model can read, and keeping it flowing. Digital platforms have APIs. TV, radio, out-of-home, print and most sponsorship do not, so their numbers arrive as media plans and spot reports from an agency, every month. And a downloaded number is not a usable one until someone has decided what it contains. Section 3 covers both.
A model built once is a report. A model that is retrained as new months arrive, checked against actual sales and adjusted when a channel changes is a system. The difference is work, every month, for as long as you want to use it.
Somebody has to read the output, explain what changed, run the scenarios and answer the CFO. In a consultancy that person is in the project until it ends. In self-serve software it is your analyst. In a managed platform it is the vendor's team.
A handful of vendors publish prices, and they show the spread. In the vendor pricing collected by whatismmm.com (updated September 2026), self-serve software priced on media spend starts at a few thousand a month: one vendor lists its full MMM at USD 4,500 a month against a USD 300,000 monthly media spend, another starts at EUR 2,700 a month and scales with last year's spend. The one managed vendor with public tiers charges USD 100,000 to 220,000 and up a year. A UK cost guide puts established econometrics vendors at £100,000 to £400,000 a year for licence and services. Most vendors, Odins included, give a price after a conversation, because the number depends on markets and channels.
Against that spread, USD 50,000 a year for one model in one market is a fair planning figure for a company with a marketing budget from EUR 1 million a year and up. What the figure does not tell you is what it includes. Two quotes at that level can differ by the whole data line. And a price that scales with media spend grows with your budget whether or not the model got any better.
Collecting and cleaning the data is most of the effort in any MMM project. One practitioner who has built them for years puts it at "80% of the job". That was true in 2021 and it is true now. What has changed is that more vendors sell the model as software and leave the data where it is. The cost does not go away. It moves to your analyst, your data engineer, a connector tool or a second invoice.
Google, Meta, TikTok, Bing and the rest have APIs. Connecting them once is a day's work for someone who has done it before. Keeping them connected is the recurring part: tokens expire, an account gets added, a platform changes its reporting. Affiliate networks, price comparison sites and smaller sources often need a connector tool such as Funnel or Supermetrics in between, which is its own licence. Run this yourself and you should count the tool licence and a share of a data engineer's time, every year.
TV, radio, out-of-home, print, sponsorship and much of influencer marketing have no API. The data exists as a media plan and a spot or airing report, held by you or your media agency. Someone has to request it, receive it every month, map each spot to the right channel and week, and chase the gaps. A UK cost guide lists this under the costs buyers forget, with two to six weeks of data engineering to assemble the history alone. The history is the smaller part. The monthly structuring continues for as long as the model runs.
A spend number is only a spend number once someone has decided what it contains. Gross or net. With or without agency fee and production. Booked or invoiced. A programmatic export such as DV360 reports the platform's total cost with fees in it, while the agency's report shows media cost; the two differ by a steady share, and the model needs one basis, applied the same way every month. Each source then has to land in one hierarchy, so that search, Google and Performance Max line up under each other the way TV, the channel and the spot do. Skip this and you have a model of your invoices, not of your marketing.
How the data should be structured is in our guide to data for marketing mix modelling, and the offline pipeline is described on our page on TV, radio and offline media spend.
What the data line costs if you run it yourself
A connector tool licence for the sources without a direct integration. A share of a data engineer for the integrations: weeks to build, hours a month to keep. And a person who structures the offline files every month and keeps the weighting consistent. Together they cost more than the tool licence looks like on its own, and they are the part that stops when that person is on holiday.
Once the data is clean, rerunning a model costs minutes of compute. That is why open-source frameworks look free. The work is everything around the rerun: checking last month's forecast against what actually happened, finding out why a channel's effect moved, deciding whether a new platform is its own channel or part of another, and rebuilding the pipeline when the agency changes. The whatismmm.com registry assumes an in-house operator spends most of a year on the first model and about half their time after that. The half holds only when the data arrives clean, which brings you back to section 3.
The upkeep has a second cost that no spreadsheet shows: the model lives with whoever built it. When that person leaves, the model usually stops. We meet companies with a Robyn or Meridian model nobody has retrained for a year. The code still runs. Nobody trusts the output.
The output of an MMM is a set of curves and a table of contributions. Turning that into "move 8% of the social budget to search, and here is the expected revenue" is a job, and so is defending the number when the CFO asks why it moved since last quarter. A consultancy includes this until the project ends. Self-serve software leaves it with your analyst, which works if you have one with time. A managed platform puts it in the price: a person reviews the recommendation before you see it and sits in the meeting where it is used. When you compare quotes, write down who this person is in each option. If the answer is a name on your own team, add their time to the total.
| Cost line | Build in-house | Consultancy or agency project | Managed platform |
|---|---|---|---|
| The model | Free framework (Meridian, PyMC-Marketing, Robyn), plus two to three data scientists with Bayesian experience for the build | In the project fee | In the subscription, usually per model or per market |
| The data work | Yours: data engineering, connector tools and the monthly offline files | Usually assumed delivered clean, or a separate line. The monthly offline structuring is yours after handover | In the price at Odins. Ask at other vendors |
| The upkeep | Yours for as long as the model runs, and tied to the person who built it | A refresh is a new project, typically quarterly or yearly | Monthly retrain in the price |
| The decisions | Your team | Included until the project ends | The vendor's team reviews and presents |
| What moves the price | Headcount | Scope and refresh frequency | Markets, sources and offline media. At some vendors, your media spend |
Two things come out of the table. The in-house route is free only at the licence line. Count two to three data scientists for the build and a large share of one to keep it running, plus data engineering, and the people cost alone lands well above a managed subscription: a single fully loaded data scientist is about USD 176,000 a year in the US figures the whatismmm.com registry uses, and the UK guide puts a one- to two-person in-house team at £150,000 to £600,000 a year. And a project is cheapest in year one and often the most expensive by year two, because every refresh is bought again and the data work starts over when the agency changes.
When the agency that buys your media also runs the model, there is a third thing to price: the independence of the answer. Our page on agency MMM versus independent measurement covers it.
Odins is a fixed annual subscription in two parts. The data platform collects and structures all your marketing data, digital and offline, into one dataset in your own hierarchy and keeps it flowing; it is priced by the scope of the data, how many markets and sources, and whether TV, radio and other offline media need a pipeline. The models we build and run on it are priced by how many you run, one per market. There are no per-user fees, no per-connector fees, no setup fee and no charge for the monthly retrain, the analyst review, support or the quarterly meeting. The price does not move with your media spend. We do not publish list prices, because two companies with the same budget can need very different setups; after a 30-minute call you have a number in writing. The details are on our pricing page.
Some companies start with the data platform alone and add a model later. The structured dataset is useful on its own for reporting, and it can flow into your own data warehouse.
Send the same brief to every vendor, and ask each of them these questions. The answers fill in the lines the quote left out.
| Ask | Why it matters |
|---|---|
| Which of the four lines does your price cover: the model, the data work, the upkeep, the decisions? | The whole guide in one question. Any line not covered is yours; add it to the total. |
| Who builds and maintains the data integrations, and what do you need from us each month? | If the answer is "your team exports a file", the data line is yours. |
| How are TV, radio, out-of-home and sponsorship collected, and who structures them every month? | The line most often missing. A pipeline for the agency's files, or your analyst's spreadsheet. |
| Who decides gross versus net, fees and currency, and applies it the same way every month? | Weighting is where models quietly go wrong. |
| How often is the model retrained, and is that in the price? | A yearly refresh bought separately doubles the year-two cost. |
| Does the price change with our media spend or our number of users? | A share of spend grows with your budget. Per-user fees grow with adoption. |
| Who reviews the recommendation before we see it, and who presents it to the CFO? | If that is your analyst, add their time. |
| Who owns the data and the model if we leave? | The exit cost belongs in the total. |
The year-two test
Write down what each option costs in year two, not year one. In-house, that is the same salaries again. A project, that is a new fee plus the data work redone. A managed platform, the same subscription. The cheapest MMM is the one that is still running, fed and trusted in year two.
Odins is a managed marketing investment platform for companies with a marketing budget from EUR 1 million to several hundred million a year. We connect the data, digital and offline, build and maintain Bayesian models, and deliver monthly recommendations our team reviews before they reach you, in one subscription that covers all four lines. Companies like CDON, Nettbil, Hyre and Megaflis run their marketing budgets on it. Set against a budget of that size, the total cost is small next to what we typically see across customers: 5 to 15% more effect from the budget you already have. That holds only while the model is running.
It is not the cheapest option for everyone. If you have a data science team with time to own both the code and the data, an open-source framework costs you salaries you already pay; our build versus buy guide for Meridian covers that choice. If your budget is well below EUR 1 million a year, the structured data is the better first investment. For the wider market, see our overview of MMM providers.
Plan around USD 50,000 a year for one model in one market as a starting point, then add the data work, the upkeep and the people who use the output where they are not in the price. Published vendor prices run from a few thousand a month for self-serve software to USD 100,000 to 220,000 and more a year for managed enterprise programmes. The total over two years is the number to compare.
The licence is. Google Meridian, PyMC-Marketing and Meta Robyn cost nothing to download. The cost is two to three data scientists with Bayesian experience for the build, a large share of one to keep it running, data engineering for the integrations and the offline files, and the risk that the model stops when its builder leaves. For a company without a data science team, that is usually the most expensive route.
If you run it yourself: a connector tool licence for the sources without a direct integration, a share of a data engineer for the integrations, and a person who structures the TV, radio and out-of-home files every month and keeps the weighting consistent. Budget two to six weeks to assemble the history, then hours to days every month. At Odins this is the data platform part of the subscription, and it is included.
It is a simple proxy for the size of the account. The drawback is that the invoice grows with your budget whether or not the model does more work. Odins prices on the scope of the work instead: markets, sources and offline media, fixed for the year.
On the quoted line, often yes. On the total, often no, because the data work, the retraining and the interpretation are in the price rather than on your team. Compare the two over two years with every line filled in, and the gap narrows or reverses.
How Odins prices, and what is included, is on our pricing page. If you want a number for your setup, book a demo: a 30-minute conversation about your markets and channels is usually enough.