Attribution relies on tracking people across devices and websites, which is getting harder. MMM takes a different approach. It uses aggregated data, such as weekly spend by channel and weekly sales, alongside factors like seasonality and price changes, to estimate each channel's effect. It does not need cookies or personal data.
How MMM works
- Collect data: weekly or daily spend and activity by channel, plus sales.
- Add other factors: seasonality, promotions, prices, weather, competitor activity.
- Build a model that explains sales from these inputs.
- Estimate contributions: how much sales each channel drove.
- Simulate scenarios: what happens if budget shifts between channels.
What MMM needs
| Requirement | Why |
|---|---|
| Two or more years of data, ideally | Captures seasonality and trends |
| Variation in spend | The model learns from changes |
| Reliable sales data | The outcome being explained |
| Several channels | To compare contributions |
| Statistical expertise | To build and check the model |
Strengths
- Privacy-friendly, using aggregated data.
- Covers offline channels and offline sales.
- Captures long-term and delayed effects.
- Supports budget planning with scenarios.
Limitations
- Needs a lot of data, so it suits larger advertisers.
- Less granular: it rarely tells you which post or ad worked.
- Can be wrong if important factors are missing.
- Takes time and expertise to build and update.
Open-source tools
Several open-source MMM tools have been released by large technology companies, such as Meta's Robyn and Google's Meridian, making MMM more accessible. They still require data preparation and statistical understanding. Many businesses work with analysts or agencies to use them.
What MMM outputs look like
Typical outputs include the estimated share of sales driven by each channel, the return on each pound spent per channel, response curves showing diminishing returns and recommended budget splits. Good outputs show uncertainty, such as a range of likely returns, and explain which factors were included. Ask for a short written summary alongside any charts, so decisions are clear.
Where social fits
MMM often shows social media contributing more than last click attribution suggests, because it captures influence that does not produce a trackable click. It can separate paid social from other channels, but organic social is harder to model because there is no spend to vary. Some models use organic reach or posting volume as an input instead.
Combining methods
| Method | Best for |
|---|---|
| Attribution | Day-to-day optimisation within channels |
| Incrementality tests | Validating specific campaigns |
| MMM | Budget allocation across channels |
Using them together gives a fuller picture. The Academy's guides to attribution models and incrementality testing explain the other two.
Is MMM right for you?
For small businesses with modest budgets and short histories, MMM is usually overkill. A simpler approach, combining UTM tracking, customer surveys and occasional tests, gives useful insight at far lower cost. MMM becomes valuable as spend, channels and complexity grow. A common threshold is when several channels each take a meaningful share of budget and decisions between them carry real financial weight.
Platform signals and modelling
MMM looks at spend and outcomes, not platform ranking. Still, creative quality affects how efficiently spend turns into results. Ads and posts that meet platform signals, such as Instagram's documented watch time and completion, likes, shares (sends per reach), saves, relationship history✓, tend to deliver more for the same budget, which shows up in the model as stronger channel performance.
Reading MMM results
Treat results as estimates with uncertainty. Look for ranges rather than single numbers, check whether results make business sense and test big recommendations with experiments before acting fully.
A simple illustration
Imagine a retailer with three years of weekly data on sales and spend for paid social, search, email and radio. The model finds that sales rise in the weeks after paid social spend increases, even when search clicks stay flat, and that the effect lingers for about three weeks. It estimates paid social contributes around a fifth of sales, compared with less than a tenth in last click reports. The retailer tests a modest budget increase in social and confirms the effect with a geo test before shifting more budget.
Adstock and saturation
Two concepts appear in most models. Adstock describes how marketing effects linger after spend, because people remember ads for a while. Saturation describes diminishing returns, where each extra pound produces less extra sales than the last. Together, they help MMM recommend how much to spend in each channel before returns fall off, which is often its most practical output.
How often to update
Models should be refreshed regularly, often quarterly or twice a year, as markets, channels and creative change. A model built on data from before a major platform change or a new competitor may give outdated advice. Many teams rebuild annually and update inputs more often.
Working with an analyst or agency
If you commission MMM, ask what data is required, how uncertainty is shown, which factors are included, how results will be validated with tests and how often the model will be updated. Ask for results in plain language with clear budget recommendations, not just charts. Good partners explain limitations as clearly as findings.
Common mistakes
- Running MMM with too little data.
- Ignoring seasonality and promotions.
- Treating outputs as exact.
- Expecting post-level insight.
- Never updating the model.
Learning advanced measurement
The Academy's advanced measurement module in the fourth level covers MMM, incrementality and reporting to leadership, with a focus on understanding results rather than building models from scratch. The free assessment shows where you would start.
