Analytics & Measurement
Google Meridian
Also known as: Meridian, Meridian MMM
Google Meridian is an open-source marketing mix modelling framework that estimates how much each media channel contributes to sales or another business outcome using Bayesian methods. It works on aggregated data, models results at geographic level and can be calibrated with the findings of incrementality experiments. Teams use it to produce channel ROI estimates, response curves and budget scenarios without relying on user-level tracking.
What it is
Meridian is a Python library released by Google so advertisers and agencies can build and run their own marketing mix models. It uses Bayesian inference, so each estimate comes with an uncertainty range rather than a single point number, and it supports hierarchical modelling across regions. Outputs include channel contribution, ROI, diminishing returns curves and budget optimisation scenarios.
Why it matters
As cookies and device identifiers become less reliable, aggregate modelling has returned as a practical way to judge the value of each channel. Meridian matters because the modelling approach is documented and controlled by the advertiser, rather than being a vendor score that cannot be interrogated. It also pushes teams towards calibrating models with experiments, which makes the resulting numbers easier to defend in a planning meeting.
How it works
Practitioners assemble consistent weekly, geo-level data covering spend and exposure by channel, the outcome measure, and controls such as price, promotions, distribution and seasonality. They set priors, fit the model, check diagnostics and posterior intervals, then run scenarios to compare budget splits. Results are normally reviewed with finance and channel owners, and refreshed as new data arrives.
When it applies
It applies when you have roughly two to three years of clean history, real variation in spend across time or regions, and cross-channel budget questions to answer rather than day-to-day campaign tuning.
Examples
- A retailer models three years of regional sales against TV, paid search, social and catalogue spend to set next year's budget split.
- A subscription brand uses the result of a completed video geo lift test as a prior so the model's video ROI reflects measured evidence.
- An agency rebuilds a client's spreadsheet based mix model in Meridian to add credible intervals and reach and frequency inputs.
How it is measured
- Out of sample prediction error on a holdout period, for example MAPE
- Width of the posterior credible interval on each channel's ROI
- Gap between modelled ROI and lift measured in geo experiments
- Forecast versus actual outcome after a budget reallocation based on the model
Insights on Google Meridian
Related terms in Analytics & Measurement
- A/B testingA/B testing is a controlled experiment that shows two or more versions of a page, email or feature to randomly split groups of users and compares how each performs against a chosen goal. It isolates the effect of a single change so improvement can be attributed rather than assumed. It is also called split testing.
- AI cost per taskAI cost per task is the total cost of completing one unit of useful work with an AI system, divided by the number of tasks completed to an acceptable standard. It includes model tokens, tool and retrieval calls, retries, and any human review needed before the output can be used. It is the unit economics measure behind an AI workflow.
- AttributionAttribution is the practice of assigning credit for a conversion to the marketing touchpoints that preceded it. It covers the models, rules and data joins used to decide which channels, campaigns or content get counted. In AI search, attribution is harder because many assistant-led journeys leave little or no referral data.
- Click-through rateClick-through rate is the proportion of people who saw something and then clicked it, expressed as clicks divided by impressions. It is used to judge search listings, ads, emails and internal links. A higher rate usually means the message matched what the audience was looking for.
- Consent ModeConsent Mode is a Google framework that lets tags adjust their behaviour based on the consent choices a visitor has made. Instead of tags being blocked outright, they receive signals about whether analytics and advertising storage are allowed, and act accordingly. Version 2 added parameters covering the use of personal data for ads and for ad personalisation.
- Conversion trackingConversion tracking is the practice of recording the actions you care about, such as purchases, form submissions or calls, and connecting them back to the channel, campaign or session that led to them. It gives advertising platforms and analytics tools the outcome data they need to report performance and optimise bidding. Accuracy depends on correct tag implementation, consent handling and clear conversion definitions.
Primary research · August 2026
How ChatGPT Shortlists Software Brands
An audit across 10 categories and 60 buying questions. I recorded what ChatGPT reads, throws away and links to when a buyer asks it which software to buy, and what that decides.
- 60
- Questions asked
- 10
- Software markets
- 2,680
- Results read
- 367
- Links shown
Free download
Get the full report
35 pages · PDF · 536 KB. Enter your details and it downloads straight away.