Short answer
Meridian is Google's open-source Bayesian Marketing Mix Modelling framework, released in 2024. It is a research-grade MMM library built on TensorFlow Probability, designed for advertisers who want a modern, calibrated MMM they can run in-house or with a partner.
Meridian is a Python library from Google's Marketing Analytics team implementing a hierarchical Bayesian MMM: geo-level modelling, reach-and-frequency inputs, adstock and Hill saturation, priors informed by experiments. It is free, open source and actively developed on GitHub.
Two things: first-class support for reach and frequency (not just spend), and native hooks for calibrating with geo-experiments. Meridian is opinionated about doing MMM properly - which is a feature, not a bug, if your data can support it.
Robyn (Meta) is faster to run, less opinionated, easier to bend. Meridian is more statistically rigorous, better for large advertisers with geo data. PyMC-Marketing is the most flexible - a general Bayesian MMM toolkit you can shape to your problem. In 2026 many advertisers run more than one.
Meridian suits mid-to-large advertisers with geo-varying spend, weekly-or-better data and a measurement team (in-house or external) that can operate a Bayesian model. It is not a self-serve tool for a marketing manager - it is a framework for a modelling team.
Bayesian MMM, calibrated with experiments, refreshed monthly, delivered as a decision system.
Explore our solutions