Marketing Mix Model

Bayesian Marketing Mix Model for measuring channel effectiveness and optimizing budget allocation. Built with PyMC for MCMC inference.

Overview

Estimates the incremental contribution of each marketing channel (TV, digital, search, social, etc.) to revenue, accounting for:

• Adstock: Carryover/decay effects of advertising
• Saturation: Diminishing returns at high spend levels
• Seasonality: Weekly and monthly patterns
• Control variables: Pricing, promotions, holidays

The model outputs posterior distributions of channel ROI, enabling data-driven budget reallocation.

Install

pip install -e .

Usage

from mmm import BayesianMMM
from mmm.data import load_sample_data

df = load_sample_data()
model = BayesianMMM(target='revenue', channels=['tv', 'digital', 'search', 'social'])
model.fit(df, samples=2000)
model.plot_contributions()
model.optimize_budget(total_budget=1000000)

Project Structure

mmm/
├── model.py          # Bayesian MMM with PyMC
├── adstock.py        # Geometric and Weibull adstock transforms
├── saturation.py     # Hill and logistic saturation curves
├── decomposition.py  # Channel contribution decomposition
├── optimizer.py      # Budget allocation optimization
├── plots.py          # Response curves, waterfall, diagnostics
└── data.py           # Data loading and validation

See examples/ for notebooks.