Marketing Mix Modelling (MMM)

A step-by-step web application for Marketing Mix Modelling that attributes sales/revenue to marketing channels using configurable transforms and multiple model types. All saturation transforms follow the law of diminishing marginal returns.

---

🌐 Live Application

🚀 Try the Live App

---

✨ Features

Step-by-Step Wizard

• Step 1: Brand name, model type, data source, column mapping (date, target, channels, segments, controls)
• Step 2: Per-channel transformation type, curvature, adstock parameters, model fit

Data Sources

• Sample dataset – Built-in synthetic MMM data
• Upload CSV – Your own dataset with flexible column mapping
• Generate dataset – Create synthetic data with custom frequency (daily/weekly/monthly/yearly) and channel names

Transformation Types (All Diminishing Returns)

Transform	Formula	Default
Negative exponential	( 1 - e^{-x/k} )	✓
Hill	( x^\alpha / (k^\alpha + x^\alpha) )
Log	( \log(1 + x/k) )
Linear	( x / (k + x) )
Power	( x^\alpha / (k^\alpha + x^\alpha) ), α ∈ (0,1)

Model Types

• Linear – OLS with optional constraints
• Ridge – L2 regularization
• Lasso – L1 regularization
• Bayesian – PyMC with priors
• Hierarchical – PyMC with partial pooling

Constraints

• Positive coefficients – Channel effects ≥ 0
• Lag sum range – Sum of channel coefficients within [lower, upper]

---

📁 Repository Structure

MMM-Marketing-Mix-Modelling/
├── mmm/                    # Core MMM package
│   ├── __init__.py
│   ├── config.py           # MMMConfig, column inference
│   ├── transforms.py      # Adstock, saturation (5 types)
│   ├── pipeline.py        # Orchestrator
│   └── models/
│       ├── base.py
│       ├── linear.py
│       ├── ridge_lasso.py
│       ├── bayesian.py     # PyMC
│       └── hierarchical.py # PyMC
├── mmm_app.py              # Streamlit step-by-step UI
├── create_mmm_dataset.py   # Synthetic data generator (CLI)
├── data/
│   └── marketing_mix_weekly.csv
├── requirements.txt
├── .streamlit/config.toml
└── README.md

---

🚀 Quick Start

Run Locally

# Clone
git clone https://github.com/ananttripathi/MMM-Marketing-Mix-Modelling.git
cd MMM-Marketing-Mix-Modelling

# Install
pip install -r requirements.txt

# Run app
streamlit run mmm_app.py

Generate Sample Data (CLI)

# Weekly data, default channels
python create_mmm_dataset.py

# Monthly with custom channels
python create_mmm_dataset.py --freq monthly --channels "TV,Digital,Radio,Brand"

# Daily, negative exponential transform
python create_mmm_dataset.py --freq daily --transform negative_exponential -o my_data.csv

---

📊 Column Mapping

The app supports dynamic column mapping – works with any dataset. Map:

Column	Description
Date	Time period (week, month, etc.)
Target	Sales, revenue, or conversions
Channels	Marketing spend columns (TV, digital, etc.)
Segments	Optional – for segment-level modelling
Controls	Covariates (seasonality, holidays, promotions)

---

⚙️ Deploy on Streamlit Cloud

1. Fork or use this repo
2. Go to share.streamlit.io
3. New app → Select repo, branch main
4. Main file path: mmm_app.py
5. Deploy

---

📦 Dependencies

• streamlit – Web UI
• pandas, numpy – Data
• scipy – Optimization
• plotly – Charts
• pymc, arviz – Bayesian/Hierarchical models (optional – remove for faster deploy if only using Linear/Ridge/Lasso)

---

📄 License

MIT