Question 1

What does MMM stand for in marketing?

Accepted Answer

MMM stands for marketing mix modeling (sometimes media mix modeling). It is a statistical method that estimates each marketing channel's incremental contribution to revenue from aggregate weekly data, no user tracking involved.

Question 2

Is MMM still relevant in 2026, or has attribution replaced it?

Accepted Answer

The opposite happened. User-level attribution weakened when Apple's App Tracking Transparency made tracking opt-in, and MMM, which needs no user data, moved back to the center. The modern consensus runs MMM alongside experiments and platform attribution rather than picking one.

Question 3

How long does an MMM take to build?

Accepted Answer

In-house first builds typically consume a quarter from data assembly to a defensible model, most of it data cleaning rather than modeling. Managed vendors quote faster onboarding since data integration is their pipeline; the model still needs your two years of history either way.

Question 4

How often is an MMM refreshed?

Accepted Answer

Legacy agency engagements refreshed quarterly or annually. Modern managed vendors and in-house stacks refresh weekly to monthly. Refresh cadence is one of the twelve questions we suggest asking every vendor.

Question 5

How much spend do you need before MMM makes sense?

Accepted Answer

There is no official floor, but the practical pattern is clear: with under roughly $5M annual paid spend or fewer than four channels, geo experiments answer your questions more cheaply than a model. Our when-not-to-use-MMM page treats this honestly.

Question 6

Is Google Meridian really free?

Accepted Answer

The license is free and open source. Running it requires a Python data science environment and someone who can defend Bayesian modeling choices; the US median data-scientist salary is $120,230 per BLS. Free software, salaried operator.

Question 7

What data does an MMM need?

Accepted Answer

About two years of weekly spend and revenue history, spend across four or more channels with real variation, one finance-grade revenue series, and a log of non-marketing events like price changes and promotions. Geo-level splits improve precision materially.

Question 8

Can MMM measure brand marketing?

Accepted Answer

Yes, within limits. MMM captures brand channels through their lagged, decaying effect on revenue (adstock) and through halo effects on other channels. What it cannot do is measure brand equity itself; it measures revenue movement.

Question 9

What does MMM cost?

Accepted Answer

Three anatomies: open-source (license $0 plus a data scientist), self-serve SaaS (published entry points at $2,990/mo Sellforte and EUR 2,700/mo Cassandra), and managed programs (Paramark publishes $100k to $220k+ per year; nine of twelve managed vendors publish nothing).

Question 10

MMM vs incrementality testing: which do I need?

Accepted Answer

Both, at different frequencies. Experiments give causal ground truth for a handful of big questions per year; the model allocates across all channels continuously and gets calibrated to the experiments. They are complements, not rivals.

MMM, frequently asked