Robyn is an open-source, AI/ML-powered Marketing Mix Modeling (MMM) toolkit developed by Meta Marketing Science under the “facebookexperimental” GitHub umbrella. Its goal is to democratize rigorous MMM: what traditionally required expert statisticians and expensive consulting becomes accessible to any company with data. Robyn takes in historical data (spends on different marketing channels, conversions, or revenue, and optional context or organic-media variables) and uses a combination of techniques, regularized regression (Ridge), time-series decomposition (trend, seasonality, holiday effects), and hyperparameter optimization (via evolutionary algorithms), to estimate the incremental impact of each marketing channel. It explicitly models “carry-over” (adstock) and diminishing-returns (saturation) effects per channel, enabling realistic modeling of how advertising persists over time and saturates.
Features
- Semi-automated MMM pipeline combining ridge regression, adstock & saturation modeling, and time-series decomposition (trend, seasonality, holidays)
- Hyperparameter optimization using evolutionary algorithms (multi-objective optimization) to search over adstock decays, saturation curves, and other model parameters
- Support for both paid-media variables and organic/contextual variables (e.g. newsletter sends, organic reach) to account for non-ad spend influences
- Calibration option against ground-truth data (e.g. lift tests, geo-experiments) when available, to improve model reliability and reduce bias
- Budget allocation optimizer that uses the model to simulate different spend distributions and recommend an efficient spend plan across channels
- Open-source availability in R (stable) and a beta Python port, enabling flexible integration in data-science workflows and accessibility across environments
Project Samples
