Google Meridian Demo

This repository demonstrates how to set up and run Google Meridian, an advanced open-source Marketing Mix Modeling (MMM) framework, using sample marketing data. The example focuses on a use case with 5 geo locations and 4 marketing channels, including impressions, spend, conversions, and reach/frequency data.

---

Features

• Data Loading: Load and preprocess marketing, conversions, and reach/frequency data
• Model Configuration: Define model specifications and priors tailored to your data
• Model Diagnosis: Run Bayesian sampling and evaluate convergence diagnostics
• Visualization: Generate insightful plots for ROI, saturation, reach/frequency, and budget allocation
• Budget Optimization: Perform media spend optimization to maximize incremental KPI (conversions)

---

Repository Structure

.venv/              # Python virtual environment with google-meridian installed
data/               # Sample dataset (demo_v1.csv)
models/             # Save and load trained Meridian models
notebooks/          # Jupyter notebook demo.ipynb showcasing full workflow
reports/            # Generated model reports and visualizations
LICENSE             # MIT License
README.md           # This documentation file

---

Setup Instructions

1. Clone the repository:

   git clone https://github.com/bhauryal-eliya/Meridian-Demo.git
   cd meridian-demo

2. Create and activate the Python virtual environment:

   python3 -m venv .venv
   source .venv/bin/activate   # On Windows: .venv\Scripts\activate

3. Install required packages including Google Meridian:

   pip install --upgrade pip
   pip install google-meridian

4. Open the Jupyter notebook and run the demo:

   jupyter notebook notebooks/demo.ipynb

---

Usage

The notebook demo.ipynb walks through the entire modeling pipeline:

• Loading the example dataset (data/demo_v1.csv) containing impressions, spend, conversions, revenue per conversion, population, and reach/frequency data
• Setting up the model with geo-location hierarchy and control variables
• Running MCMC sampling for model fitting
• Diagnosing model performance and generating plots
• Performing budget optimization to suggest spend adjustments across channels

You can replace the sample data with your own by updating the CSV file and adjusting the configuration accordingly.

---

License

This project is licensed under the MIT License.

---