Build a Marketing Mix Model with Claude Code: From OLS to Bayesian
Gui Diaz-Berrio
Marketing Analytics Leader & Author
1-Day Workshop: You understand MMM. The code is where it falls apart.
Marketing Mix Modeling is the foundation of marketing measurement. But building one from scratch takes weeks of coding: data prep, feature engineering, model fitting, diagnostics, iteration, and visualization. Most teams either outsource the work or settle for off-the-shelf tools they don't fully understand.
Agentic AI changes the economics. With Claude Code, you describe what you want, evaluate what comes back, and iterate in minutes instead of hours. The statistics are the same: geometric adstock, Hill saturation, Bayesian posterior distributions. What changes is how you orchestrate the work.
This workshop teaches you both: the methodology and the modern workflow. You'll understand what's happening under the hood and build fast enough to iterate on model specs in real time.
Workshop agenda
- 8:00AM EDT
Why this matters
The measurement problem and the agentic workflow
- 8:20AM EDT
Setup
Configure Claude Code, create your CLAUDE.md, verify environment
- 8:30AM EDT
Agentic EDA
Load data, explore correlations, visualize time series, all via prompts
- 8:50AM EDT
Break
Coffee Break
- 9:00AM EDT
OLS MMM
Adstock, saturation, grid search, model fitting, diagnostics, contributions
- 9:30AM EDT
Why OLS isn't enough
Point estimates, disconnected estimation, no domain knowledge
- 10:00AM EDT
Bayesian MMM
Priors, PyMC, MCMC sampling, posterior interpretation
- 10:45AM EDT
Break
Coffee Break
- 11:00AM EDT
OLS vs. Bayesian
Side-by-side comparison, what changed and why
- 11:30AM EDT
Build your SKILL.md
Package the pipeline into a reusable Claude Code Skill
What you’ll learn
Go from raw marketing data to two working MMMs in one hands-on session, and leave with a reusable AI-powered pipeline for your own data.
Apply geometric adstock and Hill saturation transforms via grid search
Fit an OLS model, run diagnostics (VIF, MAPE, Durbin-Watson), and iterate
Produce a channel contribution chart showing each media variable's ROI
Define priors that encode domain knowledge about channel decay and saturation
Run MCMC sampling and read posterior distributions and trace plots
Compare credible intervals vs. OLS point estimates to see what you gain
Turn the full EDA-to-model workflow into reusable commands
Customize the Skills for your own company's data and modeling standards
See the difference between vague prompts and specific, outcome-oriented ones
Learn directly from Gui
Gui Diaz-Berrio
Led measurement for €100M+ in ad spend at Kindred Group & Just Eat Takeaway
Who this workshop is for
Marketing scientists and data scientists who want to build MMMs (or already do) and want to work faster
Marketing analysts who want to understand what's inside the models their team or agency delivers
Anyone who attended the "Build an MMM in 30 Minutes" lightning session and wants to go hands-on
Prerequisites
Python 3.10+
Python 3.10+ installed on your machine
Claude Code
installed and authenticated
PyMC 5.x
PyMC 5.x and dependencies installed
What's included
Live sessions
Learn directly from Gui Diaz-Berrio in a real-time, interactive format.
Lifetime access
Go back to course content and recordings whenever you need to.
Community of peers
Stay accountable and share insights with like-minded professionals.
Certificate of completion
Share your new skills with your employer or on LinkedIn.
EDA notebook
An EDA notebook with summary statistics, correlation analysis, and time series visualizations
Claude Code Skill
A reusable Claude Code Skill that runs the full pipeline on new data with a single command
Maven Guarantee
Your purchase is backed by the Maven Guarantee.
Free resources
Frequently asked questions
$399
USD