AMMM Documentation

AMMM V2

A Stage-Gated Bayesian MMM Workflow for Production

AMMM (Advanced Marketing Mix Modelling) combines principled Bayesian inference, diagnostics-first validation, and optimisation workflows in one reproducible system.

Run QuickstartView Workflow StagesInspect Output Contract

Start Paths

Quickstart

Run your first end-to-end model with the V2 pipeline and staged outputs.

Configuration

Tune model behaviour through YAML config, priors, diagnostics gating, and CLI overrides.

Methodology

Understand the statistical rationale behind priors, diagnostics, calibration, and decisions.

Optimisation

Move from posterior inference to budget allocation under explicit uncertainty and constraints.

Core Workflow Stages

00run_metadata

Run provenance and config record.

10pre_diagnostics

Data checks and prior predictive plausibility.

20model_fit

Posterior sampling, traces, and fit artefacts.

30model_assessment

Posterior predictive fit and metrics.

40decomposition

Channel contributions and efficiency outputs.

50diagnostics

Convergence, calibration, Pareto-k, structural checks.

60response_curves

Saturation and contribution response curves.

70optimisation

Single and multi-period budget allocation outputs.

80interpretation

Generated narrative and reporting artefacts.

Inspect 50_diagnostics/ before acting on outputs from 40_decomposition/, 60_response_curves/, and 70_optimisation/.

Operational Checks

Troubleshooting

Resolve convergence, runtime, and environment issues quickly.

Testing Guide

Run canonical pytest commands and verify workflow contracts.

Config Reference

Field-by-field reference for model configuration and behaviour.

Output Schema

Canonical artefact mapping for downstream integrations.