Short answer
Robyn is Meta's open-source Marketing Mix Modelling framework, first released in 2021. It is a semi-automated MMM tool that combines ridge regression, evolutionary optimisation and Bayesian elements to make MMM faster and more accessible.
Robyn is an R package from Meta that automates a lot of the MMM workflow: adstock and saturation transformations, ridge-regularised regression, Nevergrad-based hyperparameter search and model selection against experimental calibration. It is open source and widely adopted.
Robyn made MMM accessible to teams without a full econometrics function. It ships with sensible defaults, produces defensible outputs quickly, and hooks directly into Meta's own conversion-lift data. For a mid-market advertiser it is often the fastest path from zero to a working MMM.
Robyn is not fully Bayesian - it uses ridge regression with hyperparameter optimisation rather than full posterior sampling. Uncertainty quantification is limited. It can be over-fit to short data. And the vendor is Meta - a channel Robyn also measures - which is a conflict worth naming.
Robyn is faster and easier; Meridian is more statistically rigorous and better for geo-hierarchical data. Robyn suits mid-market and self-serve teams; Meridian suits large advertisers with a modelling team. Both are legitimate 2026 choices; twenty10 runs against either.
Bayesian MMM, calibrated with experiments, refreshed monthly, delivered as a decision system.
Explore our solutions