What is MMM? Preparing for budget decisions with Meridian
Marketing mix modeling (MMM) uses time-series business and marketing data to estimate channel contributions statistically. It supports questions about contribution, spending response, and allocation. Google's Meridian is an open-source framework for this analysis. Official introduction
Start with the budget decision and available data, before building a model.
MMM, attribution, and incrementality experiments
| Method | Main question | Dependency |
|---|---|---|
| Attribution | Which touchpoints receive credit for measured outcomes? | Coverage and attribution rules |
| MMM | What contributions and budget responses can be estimated? | Data, model, and causal assumptions |
| Incrementality experiment | What outcomes would not have occurred without the activity? | Group or geographic comparison design |
These methods answer different questions. Read outcome definitions, periods, and included costs before comparing platform ROAS with modeled ROI. See incrementality measurement.
AI assists more of the preparation
On September 10, 2026, Google announced agentic support for Meridian data checks and model building, brand-signal extensions, and general availability of GeoX experiments. Measurement update
Easier preparation does not establish causality merely by pasting spend and sales into AI. Agree with the analyst on inputs, assumptions, and which results can support decisions.
Prepare consistent time and geographic data
Meridian uses spend, media exposure, KPIs, and controls for confounding. Align time and geographic granularity. Data collection guide
Use this handover worksheet. The analyst should determine which variables belong in the actual model.
| Data | Record | Check |
|---|---|---|
| Business outcomes | Revenue or purchase counts | Returns, booking dates, product scope |
| Media costs | Spend by channel | Currency, fees, geographic scope |
| Exposure | Available impressions or clicks | Comparable time and geographic coverage |
| Activity conditions | Pricing and promotion changes | Dates, scope, and duration |
| Demand changes | Available market or search records | Relationship with spending decisions |
| Data status | Missingness, definition changes, retrieval dates | Zero versus missing |
Meridian requires summable KPIs across time and geography. Retain counts rather than relying solely on rates such as CTR. KPI requirements
Reconcile taxes, returns, and revenue recognition across sources. A polished model does not resolve inconsistent definitions.
Data sufficiency depends on the model
There is no universal sufficient sample count. The guidance considers geographic and time information, model complexity, spending variation, and credible intervals. Data sufficiency
More detailed channels do not necessarily yield reliable separate estimates. Explain which activities moved together and what changed historically.
If data is insufficient, first organize measured performance and change history. The monthly report workflow can help.
Avoid adding every sales-related variable
When expected demand rises, a team may increase advertising. Demand and spend then move together. Explain the factors behind the spending decision so the analyst can assess attribution of that change.
Conversely, adding every correlated metric is not appropriate. Google's guidance emphasizes confounder selection and avoiding unnecessary variables. Control variables
Marketing teams can provide valuable records of why budgets and promotions changed, rather than independently prescribing a model.
Read uncertainty and assumptions before changing budgets
Google advises against judging MMM only by reduced prediction error. Review causal assumptions and model validity. Model assessment
Ask the analyst:
- Which outcomes and costs does the estimate cover?
- Which factors and causal assumptions are included?
- What uncertainty accompanies the central estimate?
- Do alternative assumptions or periods change the decision?
- How far can the result extend beyond historical budgets?
- Where does it agree or disagree with experiments and business records?
Combine allocation proposals with inventory, sales capacity, asset preparation, and spending limits before applying them.
An analyst handover template
Assess whether MMM can support this decision.
Budget decision:
Products, markets, and channels:
Outcome definition and included costs:
Available history and time/geographic granularity:
Missing fields and definition changes:
Advertising, pricing, and promotion change dates and reasons:
Budget, inventory, and operating constraints:
Separate supported decisions from questions we cannot yet answer.
Identify extra data, assumptions, and result validation needed.
Do not fill data gaps with invented numbers.
Complete the decision and data worksheet first. Shoryaku AI can discuss priorities using connected advertising and GA4 data. It does not build MMMs or run Meridian; work with an analyst equipped to design and assess the model.