Causal measurement · Growth Planning Studio
A Bayesian MMM is only as good as its priors.
The transparency layer beneath a Bayesian MMM. Three prior strategies compared, the update made visible, and the winner scored against a known truth.
Live tool · one self-contained file · yours to keep
The problem it solves
Priors are where an MMM quietly goes right or wrong.
Priors are the beliefs the model starts from before it sees your data. Most tools set them from generic industry defaults, hidden from view. A model can be confidently wrong, and no one can say why.
This sandbox shows three ways to set priors: industry defaults, your own frequentist decomposition, or a hybrid that takes priors from incrementality experiments where you've run them. It makes the Bayesian update visible, prior times data giving a tighter, brand-grounded contribution for every channel. And it proves, against a known truth, that the hybrid lands closest.
The decisions it supports: which prior strategy to trust, where an experiment would most improve the model, and how to defend the model's numbers to a finance team that calls MMM a black box.
How it's different
Franco makes the priors the part you can see.
Explicit, and grounded in evidence: your own frequentist decomposition for every channel, and experiment results for the channels you've tested.
The sandbox shows the full update for every channel and scores all three strategies against a hidden truth. Experiment-grounded priors win, and you can see exactly why.
In your hands
What you get.
- Three prior strategies, comparedSwitch between them and watch the contributions move.
- The update, made visibleFrequentist estimate, active prior, posterior and the hidden truth for every channel, with the response curve and marginal ROI.
- The thesis, scoredPosterior error against a known truth for all three strategies, like for like.
- The evidence checklistExactly what a live build needs from you: weekly target and spend, controls, external signal, and lift-test results to turn into priors.
And the rest
One self-contained file. No login. Works in any browser, exports any chart or table to CSV or image. A walkthrough of the findings and a refreshed build when new data lands, all included.
- No login, no seat to learn
- Exports to CSV or image
- Walkthrough of the findings included
- Refreshed build when new data lands
Alternatives
What it beats.
Open-source MMM
Ships with default priors that aren't always visible to the team using them, and they materially affect the result.
SaaS MMM
Sets the priors for you, inside the platform. You take the calibration on trust.
Consultancies
Bake in their own assumptions. You get the output, not the dials.
Not sure if this is the one?
Eight quick questions. Within 24 hours we'll come back with the solutions worth considering and the approximate budget.