Google Meridian: an open-source marketing mix model
Meridian is Google's open-source Marketing Mix Modeling (MMM) framework. It models the real effect of spend by channel on sales statistically, on aggregate data, without relying on cookies or user tracking.
- Made by
- Price
- Free, open source
- Language
- Python
- Approach
- Bayesian statistics
What is Google Meridian?
MMM brings weekly sales, channel spend, price, seasonality and other factors into one model and calculates each channel's contribution. Meridian does this the Bayesian way: you can feed prior knowledge and test results into the model.
The business question is simple: how should we split the next budget across channels? The effect of TV, YouTube and brand channels that last-click attribution misses becomes visible here.
What Meridian makes possible
Channel contribution
Each channel's share of sales.
ROI curves
The saturation point of a channel.
Budget optimisation
More results from the same budget.
Scenario planning
The expected result when the budget changes.
Using test results
Feeding lift tests into the model.
Geo models
More precise estimates at regional level.
What Meridian works with
- Open the guide
Google BigQuery
A serverless data warehouse. GA4 raw data and large-scale SQL analysis.
- Open the guide
Google Ads
Search, YouTube, Display and Performance Max campaigns.
- Open the guide
Meta Ads
Facebook and Instagram ads, Pixel and Conversions API.
- Open the guide
Google Data Studio
A free dashboard tool. Shows GA4, ad and spreadsheet data in one place.
My experience
Projects where I used Meridian:
- [Project: brand, what I built, result %]
- [Project: brand, what I built, result %]
Meridian resources
Questions about Meridian
How much data does Meridian need?
At least two years of weekly sales and channel spend data is recommended. Regional data improves accuracy.
Does it replace attribution?
No, it complements it. Attribution is for daily optimisation; MMM is for strategic budget allocation.
Can it be used without a programmer?
The model is built in Python; data preparation and interpretation need experience. Anyone can read the result in a dashboard.