Google Meridian
Bayesian MMM framework (Python)Open source, license $0. The cost is the team that runs it.
Meridian is Google's open-source marketing mix modeling framework, released as the successor to LightweightMMM. It is a Bayesian model, meaning it treats channel effects as probability distributions rather than single point estimates, and it is built in Python.
Its distinctive positions: a preference for geo-level data (modeling many regions at once tightens the estimates), support for reach and frequency inputs on video channels rather than spend alone, and a calibration API that lets you feed incrementality experiment results into the model as priors, prior beliefs the model starts from.
Is it free? The license is. The docs assume a working Python data science environment, comfort with Bayesian tooling, and someone who can defend modeling choices to a skeptical CFO. That someone is the real line item; the build-vs-buy calculator on this site prices them.
At a glance
- Model
- Hierarchical Bayesian regression with adstock and saturation transforms, geo-level by design.
- Stack
- Python. GPU recommended for larger geo hierarchies.
- Calibration
- Experiment results enter as ROI priors, the mechanism vendors charge for elsewhere.
- Governance
- Google maintains it. Google also sells the ads being measured; read the priors documentation yourself.
Who it is right for
- Brands with geo-splittable revenue data and an in-house data scientist.
- Teams that want experiment calibration without a managed contract.
- Anyone leaving LightweightMMM, which is deprecated.
Verdict: The strongest free option if you have geo-level data and a data scientist. Without both, look at managed vendors or PyMC-Marketing.
Source ledger
Every numeric claim on this page resolves to one of these primary sources.
- [1]Google Meridian documentationhttps://developers.google.com/meridianretrieved 2026-08-02
- [2]Google Meridian repository (GitHub)https://github.com/google/meridianretrieved 2026-08-02