GRIFFIN / ammm 4.0.3
Marketing mix modelling
Build, assess and interpret Bayesian marketing mix models in Python.
AMMM4 fits Bayesian media mix models, combines compatible experimental evidence
with model assumptions, and evaluates contributions and spending scenarios.
The Python distribution is ammm4; imports use ammm.
Choose a starting point for your workflow.
| Task | Start here |
|---|---|
| Choose an entry route | Getting started |
| Work in Python | Python quickstart |
| Run a model from configuration | YAML runner |
| Install this checkout | Installation |
| Prepare data and preserve fitted state | Data and lifecycle |
| Choose assumptions and an estimator | Modelling contracts |
| Assess whether estimates support an intervention | Identification and evidence |
| Find a specific public interface | API reference |
| Resolve common modelling and operational questions | FAQ |
Follow the complete modelling workflow, including prior checks, inference diagnostics, held-out prediction and sensitivity analysis. A successful run establishes execution; it does not establish that a channel effect is identified or a budget change will improve outcomes.
Documentation scope
These pages describe implementation commit 7cb7f20b357f23ee1683670953c7dd0092fa2084,
reviewed on 1 October 2026; documentation revisions may be newer than that source
commit. Source links identify the implementation behind each
contract. Read the supported surface before moving an
AMMM3 workflow: shared terminology does not imply interchangeable APIs or saved
models. The workflow organisation draws on AMMM3, with AMMM4-specific examples
and evidence limits.
The guides are ordinary Markdown. Run shell commands from the repository root. Python snippets labelled as continuations require the objects from the linked quickstart; optional stages can add fits, dependencies or provider charges.