Free marketing mix modeling software: the open-source options, what they cost to run, and what else is free
The main free marketing mix modeling software is open-source code: Meta's Robyn, Google's Meridian and PyMC-Marketing from PyMC Labs. All three are free to download and use. None includes a user interface, data connectors or support, so the cost moves from a licence to the people who build and run the model. Vendor trials are rare.
Free therefore means one of two things in this market. It can mean code that costs nothing to download and a good deal to operate, or a commercial product offered at no charge for a period, which few vendors do. This page describes the three open-source libraries, what running one involves, how many vendors publish a price or a trial, what to check before starting, and when a free option is the better choice and when a paid platform is.
The three open-source libraries
Robyn, from Meta. A marketing mix modeling package using ridge regression, with an automated search over adstock and saturation settings. It returns point estimates and a budget allocator, and it can take experiment results as calibration inputs.
Meridian, from Google. A Bayesian marketing mix model built in Python, designed for causal inference and able to take experiment results as calibration. It returns a probability distribution for each channel's effect, and it replaced Google's earlier LightweightMMM.
PyMC-Marketing, from PyMC Labs. A Bayesian Python library covering marketing mix modeling and customer lifetime value, built on the PyMC probabilistic programming library, with wide control over model structure and priors.
All three are open source and free to download, and all three ship no user interface, connectors or support contract.
In short: Robyn suits R teams that want a fast, automated start and are comfortable with point estimates. Meridian suits Python teams modeling many regions, particularly in the Google data stack. PyMC-Marketing suits Python teams that want the most control over a Bayesian model.
What running an open-source library involves
The code is one of five things a working model needs, and the only one that is free.
A data pipeline. Spend from every ad platform and outcomes from sales systems, pulled on a schedule and fixed when a platform changes its export.
Modeling work. Choosing variables, delayed-effect and saturation settings and, for the Bayesian libraries, priors, then validating the result.
Refreshes. Repeating the fit as new weeks arrive, and checking that results have not drifted for the wrong reasons.
Interpretation. Someone who can explain the output to marketing and finance and turn it into a budget plan.
Calibration. Experiments such as geo tests, designed and run by the team, because the libraries accept test results but do not run tests.
The last four recur every cycle, and they are where most of the cost sits.
The first model usually takes months rather than weeks, and most of that time goes on the data pipeline and on validation. After that, each refresh needs someone to check the inputs, rerun the fit and explain what changed.
Other free tools
Beyond the three libraries, some free browser-based and warehouse-native tools offer simpler marketing mix modeling without code. They can be useful for a first look at whether a business's data shows any signal. Before relying on one, check the same things as for any model: what it assumes, what it lets you see, whether it can take experiment results, and whether it keeps a record of each run.
Free tools also vary in where data goes. Tools that run in the browser or in your own data warehouse keep data inside the business, which matters for companies with strict data policies.
How rare vendor trials are
Commercial vendors seldom offer a trial. Of the ten vendors that passed our screening, one mentions a trial or free tier on the pages we reached: Sellforte. Of the twenty-one vendors we checked, five do: Forvio, Hyros, Improvado, Rockerbox and Sellforte. Captured 14 September 2026.
Pricing is published more often, though still by a minority. Of the ten vendors that passed our screening, six publish a pricing page and two publish a price. Of the twenty-one checked, fifteen publish a pricing page and seven publish a price.
Where a trial exists, check what it includes: how much history can be loaded, which connectors are available, whether experiments are part of it, and what happens to the data and the model when the trial ends.
Why trials are rare in marketing mix modeling
There is a practical reason. A marketing mix model needs enough history behind each channel and enough movement in that spend to read against, so a meaningful trial requires ingesting and cleaning real historical data before any result appears. On a fresh account with no history loaded, a trial has nothing to fit.
That setup work falls on the vendor, the customer or both, and it happens before anyone knows whether the result will be useful. Many vendors prefer to scope that work in a sales process instead.
For a buyer, the practical consequence is that the readiness question, whether the data supports a model at all, is often worth answering before talking to vendors. A free library or a simple free tool can answer it.
What to check before starting with anything free
Does your history support a model? Two or more years of weekly data, with enough variation in each channel's spend, is the usual starting point. With less, any tool returns wide ranges.
Who will do the work? An open-source library needs someone with modeling and engineering skills and the time to keep going after the first model.
How will the model be checked? Plan at least one experiment to calibrate it, or the first model rests on correlations alone.
What happens after the first result? Budget for the refreshes and maintenance, which is where most of the effort goes.
Where will the results be used? A model is only worth its cost if its output reaches the budget decisions it is meant to inform.
When a free option is the better choice
An open-source library is the better choice when the team has modeling and engineering skills to spare, wants full control over the method, and can maintain the model over time. It is also the cheapest way to learn how marketing mix modeling works on your own data, and to test whether your history supports a model before paying for one.
When a paid platform is the better choice
A paid platform is the better choice when the team does not have spare modeling capacity, when results are needed on a planning schedule, when the model has to keep working after the person who built it leaves, and when experiments and calibration need to be part of the service.
It is also the better choice when the cost of the people a library needs is higher than the platform's price, which is the comparison to make in writing before deciding.
What we do not do
We are not free. A price is published, and on a page about what free costs that is the whole of our claim.
Reads sit at campaign level, in the planning layer, and the platform does not optimise ad sets day to day.
A channel needs enough history behind it, and enough movement in that spend to read against, whether you are paying or not. No tool changes that, and it is the main reason a first model can be disappointing whoever runs it.
And we are not an open-source library. If what you want is code you control and methodology you can read line by line, the three libraries above are the options, and they are good ones.
How to choose
You have a modeler with time, and want control: start with an open-source library. Robyn for quick automated search in R, Meridian or PyMC-Marketing if you need uncertainty in the results.
You are unsure whether your data supports a model: use a library or a free tool to find out before paying for anything.
You need results on a schedule, and nobody has time to maintain a model: use a paid platform, and compare its published price with the cost of the people a library would need.
Cassandra, the marketing mix modeling platform, publishes a price. On a page about what free means in this market, that is the whole claim, and it is checkable in one click.
The readiness question comes first whatever you choose. Whether your history supports a model is the first thing worth knowing, and the libraries above are a real way to answer it.
Moving from free to paid, or the other way
From a library to a platform, the main asset is the cleaned history of spend and outcomes. It shortens setup, and the library's results give a reference point for the new model. Run both on the same period before relying on the new one.
From a platform to a library, ask for exports of the data, the outputs and the settings used, and check what the contract allows before signing.
In both directions, keep the record of experiments. Test results remain valid evidence whichever model uses them.
Many teams start with a library, learn what their data can support, and move to a platform once measurement has to run on a schedule. That path is a sensible one, and the work done on the library is not wasted.
Questions
Is there free marketing mix modeling software?
Yes, as code. Meta's Robyn, Google's Meridian and PyMC-Marketing are open source and free to download. None ships an interface, connectors or support, so the cost moves to the people who run them.
Do marketing mix modeling vendors offer free trials?
Rarely. Of the ten vendors that passed our screening, one mentions a trial or free tier on the pages we could reach, as of 14 September 2026. Of the twenty-one vendors we checked, five do.
Why do so few MMM vendors offer a trial?
Because a meaningful result needs real historical data ingested and cleaned first, which is setup work done before anyone knows whether the result will be useful.
What does an open-source MMM library cost to run?
A data pipeline, modeling work, refreshes, interpretation and calibration experiments. The code is free, and the other four recur every cycle.
Which free MMM library should I use?
Robyn suits R teams wanting fast automated model search. Meridian and PyMC-Marketing suit teams that need uncertainty in the results, with Meridian oriented to geo-level data and PyMC-Marketing offering the most control.