§ 01
The builder is a statistical grid
Every variable on one grid: toggle it in, tune adstock and saturation, and the model refits live. If‑Next previews show what each candidate does to R² — 0.730 → 0.741 — before you commit.
Vol. 01Econometric attribution
Marketing mix modeling, with the rigor showing.
openMMM is the open platform for teams bringing MMM in-house. Build attribution models that print their work — coefficients, standard errors, real diagnostics — and keep full control of the data, the method, and the process.
Apache-licensed core · on your laptop or in the cloud
Figure 1
The output reads like a regression table, because it is one.
Coefficients, standard errors, t‑statistics and p‑values on every driver — significance stars included. Nothing hidden behind a single ROAS number.
Coefficients
OLS · 236 obs · ● p < 0.05
R² 0.730
| VARIABLE | COEF. | STD.ERR | T | P>|T| |
|---|---|---|---|---|
| META_FACEBOOK_SPEND | 0.2796 | ±0.0392 | 7.14 | 0.000***● |
| GOOGLE_PAID_SEARCH | 0.7300 | ±0.1552 | 4.70 | 0.000***● |
| EMAIL_CLICKS | 0.0265 | ±0.0086 | 3.09 | 0.002**● |
| BLACK_FRIDAY | 1719.06 | ±209.72 | 8.20 | 0.000***● |
§ 02
OLS or Bayesian — your call
Start with OLS you can read line by line; graduate the same project to Google Meridian when you want priors — MCMC sampling and full posterior uncertainty on every estimate. The method is a decision you make, not one made for you.
§ 03
Decomposition you can defend
Per‑layer contribution curves, period by period, down to the variable. Diagnostics run real tests — Durbin–Watson, Breusch–Pagan, VIF — with p‑values and stars, not a green checkmark.
Project team
RACI · sign-off gates · audited
5 roles
- Lead A
- approves gates, promotes to production
- Modeler R
- builds and runs models
- Data steward R
- signs the data gate
- Reviewer C
- reads everything, comments
- Stakeholder I
- published reports only
DATA ● signed
PUBLISH ○ in review
PRODUCTION ◆ hard gate
Figure 2
An MMM that outlives its author.
Most models die when the person who built them leaves. openMMM builds the governance in — RACI roles per project, sign‑off gates from data to production, an audit trail on every decision — so knowledge transfers instead of walking out the door, and your MMM keeps earning trust year after year.
§ 04
Open source, genuinely
The Streamlit app runs the whole modeling workflow on your laptop — Apache‑licensed, yours forever. Prefer the cloud? Local projects import in one click. Want full control instead? Stay local. Both make us happy.
§ 05
Bring your own AI
openMMM ships an MCP server: connect Claude, your IDE, or any AI client to your workspace. Ask why TV’s contribution went negative, or have it build and run the next model — the built‑in copilot uses the exact same tools.
§ 06
Plan forward, not just backward
Reallocate budgets on a finished model with scenarios, then go weekly: the media planner rewrites last year’s flighting or plans next quarter’s — with adstock carryover and saturation doing the math.
Appendix AOffice hours
Talk to a human.
Whether you run the Streamlit app locally or build in the cloud, we help you get your MMM right — and keep it right.
20 minutes · free
Intro call
A quick sanity check: your data, your channels, and which path fits — before you invest the time.
Book 20 minutes2 hours · paid
Working session
Hands‑on consultancy on your model — data prep, specification, diagnostics, governance. Streamlit or cloud.
Book a sessionSee it refit in real time.
Load the sample dataset and start toggling variables.