Open Source MMM or a Commercial Platform and How to Choose
Open-source MMM against a commercial marketing mix modeling platform or a measurement partner, compared on cost, refreshes, transparency and lock-in, with a vendor checklist.
On this page
- What do Meridian and Robyn hand you, and what stays with your team?
- What does a commercial platform add, going by its own price list?
- The pros and cons side by side
- A hypothetical break-even on analyst time
- How should you judge a partner who will build and maintain the model?
- What belongs on the checklist for any vendor or partner?
- Sources
The open-source MMM vs commercial marketing mix modeling platform trade-offs come down to four questions: what arrives on day one, who maintains the pipelines and refreshes, how much of the model you can inspect, and what you keep if you leave. Our position, from the licenses, documentation and pricing pages we read on 7 October 2026, is that the license fee is the least important difference between the routes. Meridian and Robyn cost nothing to license and leave every surrounding job with your team, while a platform or a partner sells that labor back to you, and the contract terms that deserve the hardest negotiation are the ones deciding what leaves with you. On any route the model is one way of estimating incrementality, and it earns trust the way the other methods do, through tests you can check.
What do Meridian and Robyn hand you, and what stays with your team?
Open source marketing mix modeling with Google Meridian or Meta Robyn hands you the modeling code and its documentation, and the MMM tradeoff against a commercial measurement platform is that every job around the code stays yours. The Meridian README describes "an MMM framework that enables advertisers to set up and run their own in-house models", and the Meridian documentation adds that "Because Meridian is open-source, its underlying code is fully visible and available for you to modify." The Meridian license, Apache-2.0, grants a "perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license" and supplies the work on an "AS IS" basis. The Robyn README calls the package "an experimental, semi-automated and open-sourced Marketing Mix Modeling (MMM) package from Meta Marketing Science", and the Robyn license, MIT, permits anyone "to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software".
Support on this route is a forum with a reply cycle. Meridian's README sends questions to GitHub, where "The Meridian team responds to these questions weekly in batches, so please be patient and don't reach out directly to your Google Account teams." It also recommends a minimum of one GPU, and Robyn's README warns that "the current Python version is a LLM-translated Beta version and might encounter bugs".
Data is the larger job, and Google covers only its own share of it. The MMM Data Platform guide says advertisers new to the platform "may request an account" and that "By default, data is delivered to Google Drive", with delivery to "your own Amazon S3, Google Cloud, or SFTP storage" as an option. Feeds from other ad platforms, sales, pricing and promotions are pipelines your team builds and keeps running.
What does a commercial platform add, going by its own price list?
A platform bundles the jobs open source leaves open: connectors, a refresh schedule, a planning interface and a named support contact. Sellforte publishes its prices, and when we opened the page on 7 October 2026 it showed the Full-Funnel MMM package at $4,500 a month plus customizations and Enterprise at $8,500 a month plus customizations, both at an average monthly media spend of $300,000. The page states that "Pricing depends on your package, average monthly media spend, and the sales channels included in your scope."
Every plan lists training, online support and a dedicated Customer Success Manager. Offline media measurement "is updated monthly on Full-Funnel MMM and Enterprise, subject to customer data availability", and onboarding typically takes one to three weeks for small and mid-sized ecommerce businesses and two to four months for larger enterprises. Both MMM tiers describe a model calibrated with experiment results. Three of the items listed only under Enterprise matter for this comparison: self-serve model calibration tools, automated daily exports to the customer's data warehouse, and a Service Level Agreement with the option to negotiate contract terms. We found no commitment period on the page, which refers buyers to the sales team for it.
Lifesight shows no figures on its pricing page and says: "Our experts will be able to give you an estimate after understanding your requirements." Its FAQ describes "a single annual subscription for all modules", and its plan list places "Causal MMM with geo-test calibration support" in the Precision plan and above. Read together, the two pages show where to look for platform lock-in: the tier that gives you control of calibration, the tier that sends results back to your warehouse, and a contract term that is either an annual subscription or a question for sales.
The pros and cons side by side
The pros and cons of Robyn and Meridian as open source MMM, set against a commercial vendor, fit in one table, with a measurement partner added as the third route.
| What decides it | Meridian or Robyn in-house | Commercial platform | Measurement partner |
|---|---|---|---|
| What you pay for | No license fee; staff time, compute and experiments | A subscription; Sellforte lists $4,500 a month plus customizations for its MMM tier at $300,000 in monthly media spend | Fees set by contract; we found no rates on Google's partner page |
| Pipelines and refreshes | Your team builds and runs them | Vendor connectors and schedule, dependent on the data you supply | Whoever the statement of work names |
| Transparency | Code is public and yours to modify | What the vendor documents; ask for the model specification | Open code when built on Meridian or Robyn; priors and data preparation depend on the handover |
| Support | GitHub and a user community | A customer success contact; an SLA on the top tier of the page we read | A named team for the length of the contract |
| Where lock-in sits | The people who know the model | Exports, calibration controls and contract terms | The handover: code, runbooks and data ownership |
A hypothetical break-even on analyst time
Take a hypothetical advertiser, invented for illustration, that spends $300,000 a month on media, the level at which Sellforte's page displayed its prices. The Full-Funnel MMM package at $4,500 a month comes to $54,000 a year before customizations, and Enterprise at $8,500 a month comes to $102,000. Meridian and Robyn carry no license fee, so the figure to set against the subscription is the share of an analyst's year the model consumes beyond what a platform would still require of that person. The loaded analyst costs below are our assumptions, to be replaced with yours.
| Assumed loaded analyst cost per year | Break-even share of the year against $54,000 | Break-even share against $102,000 |
|---|---|---|
| $120,000 | 45% | 85% |
| $150,000 | 36% | 68% |
| $180,000 | 30% | 57% |
At the assumed $150,000, open source costs less in cash only while the model takes under 36% of one analyst's year, measured against the MMM tier. The work recurs, too: the IAB best practices guide of December 2025 recommends "quick refreshes weekly and full retrains monthly or quarterly based on data stability". The table leaves out compute, the cost of experiments and the customization fees on the platform side, so treat it as the number to beat when you estimate your own hours.
How should you judge a partner who will build and maintain the model?
Evaluate a measurement partner hired to build and maintain your MMM (marketing mix modeling) on what they will hand over, because the code is free and the value sits in the priors, data preparation and runbooks written around it. Google's Meridian partner page invites advertisers to "Choose from our certified global partners to help you deploy Meridian's advanced MMM and GeoX incrementality measurement capabilities", and lists firms by region, among them Accenture, Tinuiti, Merkle and Ekimetrics. We found no certification criteria and no prices on that page. That certification is Google's own program for its own tools. The IAB RFI template notes that formal accreditation of MMM does not exist, listing "Claims of formal MMM accreditation (none exists)" among its red flags, so no badge stands in for the partner's own evidence of modeling quality.
The handover is where the IAB is most specific. Its guide says to "Require modular toolsets, open APIs/exportable model packages, vendor handover playbooks, and a clear in-house checklist", and gives the reason: "Flexible deployment lets you scale measurement capability without vendor lock-in or brittle transitions." Data access needs the same care. Google's data platform guide says that before it shares MMM data with anyone else, "prior permissions must be obtained from the data owner (which is typically the advertiser but can be the media agency or other third-party)", so settle who holds that role before a partner's first data request. A partner who maintains the model should also feed it experiments, and the IAB's instruction is to "Feed validated results (geo/holdout/lift results) into the model and document how they were applied." A quarterly geo-holdout test, with the result analyzed before it is used, is one way to produce them.
What belongs on the checklist for any vendor or partner?
A checklist of criteria for choosing an MMM vendor does its job when each question names the evidence you expect back. The IAB template is built that way, telling buyers to "Ask vendors to provide examples, screenshots, or documentation wherever possible." The questions below are quoted from it, and the evidence column is ours.
| What you are testing | Question from the IAB template | Evidence to request |
|---|---|---|
| Reproducibility | "How are methodologies, assumptions, and inputs documented and reproducible by independent reviewers?" | The model specification and, on an open-source build, the code and prior settings |
| Stability | "Provide results or ranges for ROI or contribution stability when retrained; how do you monitor and address drift?" | Channel ROI across several past refreshes, with out-of-sample backtest results |
| Calibration | "Do you maintain an experiment priors log or calibration record for model baselines?" | The log, showing which test moved which estimate |
| Full cost | "Are onboarding, support, and channel updates included in base fees?" | A price sheet that separates data ingestion, model refreshes and user licenses |
| Portability | "Can data be exported or connected via API for external analysis?" | A sample export of results and cleaned inputs |
| Exit | "What are your termination and renewal policies, including notice and penalties?" | The draft contract and its data ownership clause |
The template's red flags belong in the meeting as well: "Hidden analysis or refresh fees", "Non-transferable data rights or perpetual vendor lock-in", and "PDF or static PowerPoint deliverables with no data lineage". Put the first three rows to an in-house team too, since a model that nobody else can reproduce is locked in to one analyst.
Our recommendation follows from who you employ. A team with an analyst who will own the model beyond its first year, and with data already in a warehouse, should run Meridian or Robyn, bringing in a partner for the build only with the handover written into the contract. A team without that person should buy a platform and negotiate exports and calibration control into whichever tier it pays for. On both routes the model's estimates have to be reconciled with platform-reported numbers, which the multiplier method covers. That work sits with our data and analytics service, and the Attribution Doctor is a further resource for the attribution side of your measurement.
Sources
- Google, Meridian repository and README
- Google, Meridian license, Apache-2.0
- Google, An introduction to Meridian
- Google, Use MMM Data Platform
- Google, Find Meridian Partners for MMM and GeoX
- Meta Marketing Science, Robyn repository and README
- Meta Marketing Science, Robyn license, MIT
- IAB, Modernizing MMM Best Practices for Marketers, December 2025
- IAB, Modernizing MMM Request for Information Questions, December 2025
- Sellforte, Sellforte pricing
- Lifesight, Lifesight pricing