AI Marketing Mix Copilot - Bayesian MMM & Causal Incrementality

Bayesian media-mix modeling with causal incrementality and budget simulation.

No hosted demo for this project — it runs locally. See install instructions below.

AI & Automation

Overview

AI Marketing Mix Copilot is a Bayesian marketing-mix modeling (MMM) and causal incrementality copilot built with Python and FastAPI. Five specialized AI agents decompose multi-channel revenue, evaluate MCMC chains, analyze geo-holdout experiments and simulate counterfactual budget shifts with 90% Bayesian credible intervals.

Screenshot of the ai-marketing-mix-copilot project — Dashboard mockup from the repository docs: budget slider, blended vs marginal ROAS, revenue contribution donut and approval queue.

Problem It Solves

Blended ROAS misleads marketers into overspending on saturated channels. Teams need uncertainty-aware, causal media allocation decisions rather than last-touch reporting, and they need those decisions explained in plain language for executives.

Target Users

Performance marketers and media buyers
CMOs and CFOs reviewing budget allocation
Agencies managing multi-channel portfolios

Key Features

Bayesian MMM with adstock carryover and Hill saturation curves
Difference-in-Differences causal lift estimation from geo-holdout experiments
Five specialized agents with strict read-only tools plus a governed write path
Interactive budget simulator and allocation optimizer with credible intervals
Executive explanation agent for plain-language briefings
19 passing unit and integration tests

Technologies

Python
FastAPI
Bayesian MCMC
Pydantic
Docker
Chart.js dashboard
Model Context Protocol (MCP)

Contribution & Skills

Sole author. Implemented the econometric engine (adstock, Hill saturation, marginal ROAS, DiD), the five-agent orchestration with human-in-the-loop governance, the reactive dashboard and the 19-test evaluation suite.

Install & Usage

Quick demo

git clone https://github.com/hamid-digitalmarketer/ai-marketing-mix-copilot.git
cd ai-marketing-mix-copilot
python demo.py

Launch the dashboard

pip install -r requirements.txt
uvicorn backend.app.main:app --host 0.0.0.0 --port 8000

Containerized

docker compose up --build

Full documentation lives in the repository README →

FAQ

What problem does the Copilot solve?

It replaces guesswork in media budget allocation by modeling diminishing returns, computing marginal ROAS per channel and estimating causal lift with credible intervals, then translating results into executive briefings.

Is there a hosted demo?

No hosted demo exists. The repository includes a standalone demo script and a dashboard mockup image; the full app runs locally via uvicorn or Docker.

Related Projects