PyMC-Marketing

PyMC-Marketing suits analysts and data science teams building Bayesian marketing models in Python. Its appeal is the combination of flexible marketing mix modeling, customer analytics and budget optimization; the trade-off is a technical setup with specific, current dependency requirements.

Overview

This open-source library covers marketing mix modeling (MMM), customer lifetime value, customer choice and Bass diffusion. It also includes an alpha-stage model for predicted incrementality by experimentation. Bayesian posterior distributions and credible intervals make uncertainty part of its predictions, which can help teams weigh a range of plausible outcomes rather than rely on a single estimate.

PyMC Labs and PyMC core developers maintain the library. PyMC Labs describes itself as a Bayesian AI consultancy founded by members of the PyMC community.

Key features

• Configurable MMM: Custom priors and likelihoods, adstock and saturation transformations, and time-varying effects let modelers tailor analyses to their assumptions and data. Diagnostics and out-of-sample predictions support evaluation, but the flexibility places more responsibility on users to make and assess modeling choices.
• Budget optimization: The built-in tool allocates spend across channels to maximize ROI, connecting MMM outputs to budget decisions.
• Experiment calibration: Users can calibrate MMM models with empirical experiments such as lift tests, bringing experimental evidence into the modeling workflow.
• Customer analytics: Predictive CLV, cohort analysis and customer segmentation suit teams studying customer value alongside marketing performance.
• Inference and uncertainty: Multiple NUTS samplers, including BlackJax, NumPyro and Nutpie, provide inference options; PyMC backends allow GPU acceleration. Posterior distributions and credible intervals express prediction uncertainty, though using these tools assumes statistical and Python expertise.
• Learning and setup: A guided learning path, example notebooks and API reference support adoption. Installation uses pip install pymc-marketing, with no conda environment required. The current documentation requires Python 3.12 or later, PyMC 6.0 or later and ArviZ 1.2 or later, so teams should check their environment before committing.

Pricing

PyMC-Marketing is free: the PyMC-Marketing plan costs nothing and provides an Apache 2.0 open-source Python library. There is no free trial because the software itself is free, and no paid tier or seat quota is stated. A free 30-minute strategy session with PyMC Labs experts and a contact option provide a route to discuss use of the library.

Platforms

PyMC-Marketing runs on Linux and is self-hosted. It is a Python library rather than a hosted web application, so teams need a suitable Python environment and the ability to manage their own modeling workflow.

Who it's for

It is a strong fit for analysts, data scientists and marketing measurement teams who can work with Python, PyMC and Bayesian modeling. Teams that want to customize MMM, include experimental calibration, quantify uncertainty or connect channel models to budget allocation can use its broad modeling toolkit in one library.

It is a poor fit for marketers seeking a ready-to-use dashboard or teams that cannot support the required Python and PyMC versions. The alpha-stage incrementality model is best approached as an early-stage option rather than a finished core capability.

Pros and cons

• Pros: Broad modeling scope. MMM, CLV, customer choice, Bass diffusion and experimentation models cover several marketing analytics needs within one library.
• Pros: Modeling flexibility with decision tools. Custom MMM components, experiment calibration and ROI-focused budget allocation let experienced teams carry analysis toward spend decisions.
• Pros: Open source and free. The Apache 2.0 license and free plan make it accessible without a software subscription.
• Cons: Technical operating requirements. Users need a self-hosted Linux/Python workflow and current versions of Python, PyMC and ArviZ, which may complicate adoption in existing environments.
• Cons: Statistical judgment is required. Custom model choices and uncertainty estimates are valuable to skilled modelers, but they demand expertise to configure and interpret responsibly.
• Cons: One capability is alpha-stage. Predicted incrementality by experimentation is not as mature as a finished model offering.

Alternatives

Choose Google Meridian if you want a free, open-source marketing measurement option with source code and methodology papers intended to support transparency and user auditability.

Robyn is another free, open-source option for teams comparing marketing mix modeling libraries.

Choose Revup Marketing if a web-based product and a free plan limited to one plan, one user and one team better match your needs; paid tiers cover more plans, users and teams at custom pricing.

For enterprise marketing measurement platforms with custom pricing, consider LiftLab or Measured. Brandsystems offers a paid web-based marketing resource management product with pricing provided by demo request. Fintastic has a paid enterprise subscription with unlimited users and data, while Planful tailors subscription pricing to company size, user count and selected modules.

For broader comparisons, see Marketing Performance Management Software and Customer Lifetime Value Software.

Verdict

PyMC-Marketing is a compelling choice for technically equipped teams that want open-source Bayesian MMM, customer analytics and budget allocation in a customizable Python library. Choose it for modeling control and explicit uncertainty; look elsewhere if your team needs a hosted, nontechnical workflow or cannot meet its Python and PyMC requirements.