Marketing mix modeling (MMM)
Marketing mix modeling ignores individual users and looks at the whole business instead: when spend on each channel went up or down, what happened to results?
Updated
MMM started in consumer goods, where TV and print had no clicks to track. It has returned to app marketing because privacy changes, especially on iOS, have made user-level attribution patchy. A model that needs only daily or weekly totals per channel is unaffected by missing identifiers.
How marketing mix modeling works
An MMM is typically a regression model. The outcome, for example daily new subscribers, is explained by spend on each channel plus other factors that move results, such as seasonality, pricing changes, app releases, featuring and promotions. The model estimates how strongly each input moves the outcome.
- Baseline. The part of results the model attributes to no paid activity at all, which includes organic demand and brand.
- Adstock. Advertising effects decay over time rather than stopping the day spend stops, so models spread each day’s spend over the following days.
- Saturation. Returns diminish as spend rises; the model estimates where each channel starts to flatten.
- Contribution and marginal return. The outputs show each channel’s share of results and what the next dollar is likely to return, which is what budget allocation needs.
When MMM makes sense for an app
- You spend meaningfully across several channels, including ones that are hard to attribute, such as TV, influencers, podcasts or out-of-home.
- You have enough history: the model needs many weeks of data with real variation in spend per channel. Flat budgets teach it nothing.
- User-level attribution is weak or inconsistent across platforms, and you need one comparable view.
- The decision you face is budget allocation between channels, not which keyword or creative to bid on.
For an app spending mostly on one channel, or at an early stage with little history, MMM rarely adds much. Direct attribution plus a few well-designed incrementality tests usually answers the questions faster.
Limits of MMM
- Correlated spend. If you always raise every channel at once, the model cannot separate their effects.
- Coarse output. MMM works at the channel or campaign-type level. It cannot say which Apple Ads keyword or ad group to bid up.
- Slow feedback. Models need weeks of new data to update, so they guide quarterly planning better than weekly optimization.
- Assumptions. Adstock and saturation choices shape the results. Different reasonable settings can produce different answers, so calibrate against experiments.
MMM vs multi-touch attribution
| Aspect | Marketing mix modeling | Attribution and MTA |
|---|---|---|
| Data | Aggregate spend and outcomes over time | User or device-level touchpoints |
| Granularity | Channel or campaign type | Campaign, ad group, keyword, creative |
| Privacy impact | Largely unaffected by identifier loss | Weakened by identifier loss |
| Best for | Budget split between channels | Day-to-day optimization within a channel |
The two are complementary. MMM decides how much should go to Apple Ads relative to other channels; keyword-level attribution joined with trials, subscriptions and revenue decides where that budget goes inside Apple Ads.
Related terms: mobile attribution and ROAS are covered in the adapty.io glossary; incrementality has its own entry here.