Short answer
Meridian and Robyn are both free and credible. Meridian is Bayesian and handles regional data well. Robyn gets you a first answer faster. Neither is a product — the real cost is the person who runs it. And both will hand you a confident model that fails on new data if nobody checks.
When a credible model costs nothing to license, the question stops being "can we afford this" and becomes "who builds it and defends it."
Meridian
Built in
Robyn
Limited
Custom
As needed
Only worth it if you actually control spend by market. Hand Google a national budget and there is no geo pattern to model.
When is open source the right answer?
Take the free framework if most of these are true.
- Someone in-house is comfortable with this kind of modeling, and will still be there in a year.
- Your channels are stable and mostly digital, with clean weekly spend data.
- You want to learn the shape of the problem before paying anyone.
- The decisions riding on it are important but reversible.
A well-built Robyn or Meridian model, properly checked, beats an unvalidated paid one. The framework is rarely what makes a model good.
What is the failure nobody warns you about?
Both frameworks will give you a model. Neither will tell you the model is wrong. The charts look sensible, the fit looks strong — and nothing ever tested whether it can predict something it has not seen.
When is a custom build worth it?
- 01A real share of spend is offline — TV, connected TV, radio, billboards, print.
- 02Your KPI is not a clean online order: store visits, prescriptions, dealer leads.
- 03You control spend by market, so a model can borrow strength across markets.
- 04You want test results built in as the model’s starting point.
- 05The output has to survive finance, so assumptions and validation must be written down.
If none of that applies, use the free framework. That is a real recommendation.
Is Meridian a replacement for Robyn?
A different set of trade-offs, not a strict upgrade. Meridian handles Bayesian priors and regional structure well. Robyn usually gets you a first answer faster.
Can I put test results inside the model?
In a Bayesian framework, yes. A match-market result becomes the starting assumption for that channel. One of the most valuable things you can do with a test.
How long does a first open-source model take?
Days to run. Months to get one you would defend to a CFO. Most of that is data cleanup and validation.
Do we still need tests once we have a good model?
Yes. The model gives you every curve. The test tells you whether they are real.
Where we’re different
Click a claim. We’ll show our work.
Marketing mix modeling is a time-series problem. So we hide the last three to six months from the model, then make it predict them. If it cannot get under 10% error on months it never saw, we do not ship it — and you get your money back. Ask anyone else on your shortlist for that number. Most of them do not have it.
We are data scientists, not a software company. We do not sell a tool that competes with the vendors on these pages. That is why we will tell you when one of them — or a free open-source framework — is the better answer for you.
Book the two-week auditFind your wasted third
Want this run on your own numbers?
The two-week audit shows you where each of your channels sits on its own curve — what to cut, what to grow, and how much budget is sitting in the wrong place.
Book the two-week auditKeep reading
- When you should not buy an MMMMarketing mix modeling gets sold to companies that cannot use it yet. Five situations where the honest answer is to wait, and what to do instead.
- The model that failed, and what it cost to find outWe split a working model in two. One came back with 832% error on unseen data. Why it happened, and why it would have been invisible without a holdout.