Robyn
What it is
Robyn is an experimental, open-source Marketing Mix Modeling package from Meta Marketing Science. It uses machine-learning techniques to estimate media channel efficiency and effectiveness, adstock, and saturation. Its time-series modeling separates trend, seasonality, and holiday patterns, while ridge regression regularizes multicollinearity and helps prevent overfitting. Evolutionary algorithms automate hyperparameter optimization, and models can be calibrated against ground-truth methods such as geo-based tests, Facebook Lift, and MTA. The budget allocator uses a constrained nonlinear solver to explore reallocations intended to maximize outcomes; model one-pagers support comparisons. Robyn is designed for granular datasets with many independent variables, particularly for digital and direct-response advertising with rich data sources. It does not require personally identifiable or individual-level data and does not depend on cookies or pixel data. A stable R version is on CRAN, with a development version on GitHub. A Python version is marked beta, may have translation issues, and requires the R package installed. Robyn is free and MIT licensed.
For and against
For
- Free and MIT licensed.
- Supports calibration against causal tests and other methods.
- Budget allocator explores outcome-focused reallocations.
- Does not require individual-level data or cookies.
- Stable R version available on CRAN.
Against
- Python version is beta and may have translation issues.
- Python API requires the R package installed.
- Paid media variables and spend vectors must match in length and order.
The verdict
Robyn combines media modeling, calibration, and budget allocation for data-rich marketing teams. The stable R version is the more established listed option; its Python offering is beta.
Who it suits
Robyn suits digital and direct-response advertisers with granular data and many independent variables. It may also fit teams that want budget planning, forecasting, scenario planning, or ROI reporting.
On the record
- Adstock options
- Robyn offers geometric, Weibull CDF, and Weibull PDF adstock transformations.facebookexperimental.github.io · read 30 Sep 2026
- Availability
- Robyn has a stable R version on CRAN and a development version on GitHub; the maker also documents a Python version marked beta.facebookexperimental.github.io · read 30 Sep 2026
- Budget allocation
- Its budget allocator uses a gradient-based constrained nonlinear solver to maximize outcomes by reallocating budgets.facebookexperimental.github.io · read 30 Sep 2026
- Calibration
- Robyn can calibrate models against ground-truth methodologies including geo-based tests, Facebook Lift, and MTA.facebookexperimental.github.io · read 30 Sep 2026
- Input requirement
- Paid media variables and paid media spend vectors must have the same length and media order.facebookexperimental.github.io · read 30 Sep 2026
- Integrations
- Robyn uses Nevergrad for optimization, Prophet for trend and seasonality decomposition, and glmnet for ridge regression fitting.facebookexperimental.github.io · read 30 Sep 2026
- Intended users
- The package is built for granular datasets with many independent variables and is described as especially suitable for digital and direct-response advertisers with rich data sources.github.com · read 30 Sep 2026
- License
- The repository states that Robyn is MIT licensed.github.com · read 30 Sep 2026
- Model calibration
- Robyn calibrates marketing mix models using causal experiments such as randomized controlled trials and geo experiments.facebookexperimental.github.io · read 30 Sep 2026
- Model comparisons
- Robyn generates model one-pagers to support intuitive model comparisons.facebookexperimental.github.io · read 30 Sep 2026
- Modeling
- Robyn uses machine-learning techniques to estimate media channel efficiency and effectiveness, adstock rates, and saturation curves.github.com · read 30 Sep 2026
- Optimization
- Robyn uses a multi-objective evolutionary algorithm for hyperparameter optimization.facebookexperimental.github.io · read 30 Sep 2026
- Privacy
- The maker describes Robyn as privacy friendly, requiring no PII or individual-level log data and not depending on cookies or pixel data.facebookexperimental.github.io · read 30 Sep 2026
- Privacy design
- Robyn does not require personally identifiable information or individual-level data and does not depend on cookies or pixel data.facebookexperimental.github.io · read 30 Sep 2026
- Product
- Robyn is an experimental, AI/ML-powered, open-source Marketing Mix Modeling package from Meta Marketing Science.facebookexperimental.github.io · read 30 Sep 2026
- Product type
- Robyn is an experimental, AI/ML-powered, open-source Marketing Mix Modeling package from Meta Marketing Science.facebookexperimental.github.io · read 30 Sep 2026
- Python API limitation
- The beta Python API requires the Robyn R package to be installed first.facebookexperimental.github.io · read 30 Sep 2026
- Python availability
- The Python version is a beta rewrite of Robyn's R package and may have translation issues.facebookexperimental.github.io · read 30 Sep 2026
