PyMC-Marketing MMM

Build Bayesian marketing mix models with PyTorch.

Updated Mar 19, 2026

Search Tags:#saturation #bayesian #gam #marketing-mix #adstock #hierarchical-priors

Ready to run

npx skills add https://github.com/benmaier/decision-agent-placeholder --skill pymc-marketing-mmm

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: PyMC-Marketing MMM
Source: https://github.com/benmaier/decision-agent-placeholder/tree/main/decision-packs/mmm/opencode/skills/pymc-marketing-mmm
Command: npx skills add https://github.com/benmaier/decision-agent-placeholder --skill pymc-marketing-mmm

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

PyMC-Marketing MMM enables users to design and configure Bayesian Marketing Mix Models with adstock, saturation, and hierarchical priors, helping deliver principled inferences for marketing effectiveness.

Core Features & Use Cases

Flexible MMM architectures from simple to multidimensional hierarchies with Bayesian priors and GAM components
Built-in adstock and saturation transformations for carryover and diminishing returns
Priors configuration, model building, and posterior inference workflows with diagnostics
Guidance for using patterns (single-series and multi-market) and evaluating model fit

Quick Start

Provide a minimal example by creating an MMM with GeometricAdstock and LogisticSaturation, building the model with your data, and running a first fit.

Frequently Asked Questions about PyMC-Marketing MMM

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I build a Bayesian marketing mix model with adstock and saturation?▼

You can build a Bayesian marketing mix model by configuring adstock and saturation components, setting priors, and running sampling. This approach applies carryover and diminishing returns transformations to channel spend data for principled effectiveness inference.

Can I model marketing mix data across multiple regions or products?▼

Yes, you can model marketing mix data across multiple regions or products. The workflow supports multidimensional hierarchies, allowing you to apply hierarchical priors and baseline configurations to multi-market datasets.

What is needed to start configuring Bayesian MMM priors and workflows?▼

To start configuring Bayesian MMM priors and workflows, you need date variables, channel spend variables, and a target variable. Optional dimensions like regions or products can be included, followed by setting up a pipeline for priors and sampling.

Does this Bayesian MMM approach support generalized additive model components?▼

Yes, the Bayesian MMM approach supports optional generalized additive model (GAM) components. You can integrate these into the model structure alongside adstock and saturation transformations to capture non-linear relationships in the data.

How do I evaluate model fit and posterior diagnostics for a marketing mix model?▼

You evaluate model fit and posterior diagnostics by analyzing the inference workflows provided. This involves examining posterior distributions and diagnostic outputs to assess the reliability of your marketing mix model results.

What are the limitations of using hierarchical priors in multi-dimensional marketing mix models?▼

Hierarchical priors in multi-dimensional marketing mix models require careful configuration to avoid over-shrinkage across regions or products. Proper structuring of baseline priors and model components is essential to maintain principled inferences.