PyMC-Marketing vs Robyn

PyMC-Marketing is an open-source Python library for Bayesian marketing analytics, with tools for marketing mix modeling (MMM) and customer lifetime value. Its MMM API supports custom priors and likelihoods, adstock and saturation transformations, time-varying effects, diagnostics and out-of-sample predictions. Models can be calibrated against empirical experiments such as lift tests, and built-in budget optimization allocates spend across channels to maximize ROI. Bayesian outputs include posterior distributions and credible intervals for predictions. The library also includes customer choice, Bass diffusion and alpha-stage predicted incrementality by experimentation models, alongside predictive CLV, cohort analysis and customer segmentation. It runs on Linux and can be self-hosted. Current documentation requires Python 3.12 or later, PyMC 6.0 or later and ArviZ 1.2 or later; installation is available through pip with `pip install pymc-marketing`, without requiring a conda environment. The library is free under the Apache 2.0 license. PyMC Labs and PyMC core developers maintain it, and learning resources include a guided path, example notebooks and an API reference.

Robyn is a free, open-source Marketing Mix Modeling package from Meta Marketing Science. It estimates media channel efficiency and effectiveness, adstock rates and saturation curves using machine-learning techniques. The package is intended for granular datasets with many independent variables, particularly those used by digital and direct-response advertisers with rich data sources. Robyn automates hyperparameter optimization with evolutionary algorithms and uses ridge regression to address multicollinearity and overfitting. Its time-series modeling accounts for trend, seasonality and holidays, 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 seek outcome gains through budget reallocation; model one-pagers support comparisons. Robyn does not require personally identifiable or individual-level data and does not rely on cookies or pixel data. A stable R version is available on CRAN, alongside a development version on GitHub. A Python version is documented as beta and may have translation issues. The package is MIT licensed; support options include a public user group and GitHub issues.