Why Marketing Mix Modeling Works Without Cookies
Privacy changes cut user-level tracking. Why marketing mix modeling still works, how to act on its results, and how it fits with other measurement.
For years, much of digital marketing measurement rested on following individual people: cookies on the web, device identifiers in apps, and attribution rules that split credit between the touchpoints a tracker could see.
That foundation has weakened. Apple's App Tracking Transparency, introduced with iOS 14.5 in 2021, required apps to ask permission before tracking users, and many people said no. Safari and Firefox block third-party cookies by default. Google ultimately kept third-party cookies in Chrome, but the broader trend is clear: user-level data is less complete than it was, and the numbers in ad platforms and mobile measurement dashboards have become harder to trust.
Marketing mix modeling (MMM) doesn't depend on any of that. This guide explains why, what it can and can't do, and how it fits with other measurement methods.
What marketing mix modeling measures
MMM is a statistical method that uses regression to estimate how much each factor contributes to an outcome, usually sales. The inputs are aggregate: sales over time, spend or impressions by channel, and outside factors that also move sales.
That lets a model answer questions such as:
- How much revenue does TV drive compared with search and social?
- Where does each channel hit diminishing returns, the point at which more spend adds little?
- How much did promotions, weather or a disruption like a pandemic affect results?
Because it never needs to identify a person, the model works the same way whether or not cookies or device IDs are available. For a full walkthrough of the method, see our guide to marketing mix modeling.
Why MMM faded, and why it came back
MMM dates to the 1960s. As programmatic advertising and cookie-based tracking grew, it took a back seat. User-level data seemed more precise and more immediate.
The promise didn't fully hold. Rich tracking data still had to be turned into credit through attribution rules, and settling on sound rules proved hard. Many teams defaulted to last-click attribution, which gives all credit to the final touchpoint before a conversion. That approach ignores everything earlier in the journey and can't say whether a sale would have happened anyway. Our piece on the problem with last-click attribution covers this in detail.
As privacy changes reduced tracking and modeling techniques improved, marketers returned to MMM as a way to measure performance without relying on user-level data at all.
What MMM does well
Cross-channel measurement. MMM puts every channel on the same basis, including ones people don't click, such as TV, radio, podcasts and out-of-home. That makes channels comparable in a way platform reports, each grading its own results, can't.
Budget planning and forecasting. Once the model estimates how each channel responds to spend, you can test "what if" scenarios and forecast the effect of a new plan before committing to it.
Durability. Because it uses aggregate data, MMM isn't affected when a browser, operating system or platform changes its tracking rules.
How an MMM gets built
Good results start with clean data. That means gathering a solid run of historical data and mapping campaign names and structures to consistent dimensions, so that spend and results line up correctly across platforms and over time.
The modeling team then fits the model, using techniques such as adstock to capture how long advertising keeps working after it runs. Only once the model is delivered and checked does the analysis of marketing impact begin.
Acting on the results
The basic logic is simple. If a channel drives more than its share of results relative to its share of spend, it is a candidate for more budget. If it drives less, it is a candidate for cuts.
Two cautions apply. Make changes gradually, because MMM results often differ from what teams are used to seeing in platform dashboards, and a large sudden shift adds risk. And when a finding is surprising, test it. A controlled experiment is a good way to confirm what the model shows before moving significant money.
The limits of MMM
MMM isn't the answer to every question:
- Less granular. It estimates contribution at the channel or campaign level, not for individual ads or users.
- Slower. It needs a meaningful run of history and isn't built for real-time optimization.
Traditional models made both limits worse by being rebuilt once a year. Models refreshed more often, with more granular channel data, narrow the gap, but they don't replace in-platform tools for daily decisions.
How MMM fits with other measurement
MMM is one of three common approaches, and each answers a different question:
- Platform attribution reports results quickly and at any level of detail, so media buyers can optimize campaigns on recent performance. It fills MMM's gaps in speed and granularity, but each platform measures only its own ads and tends to favor itself.
- Lift or incrementality testing checks whether the conversions a platform reports were actually caused by the marketing.
- MMM provides the cross-channel view used to plan budgets.
Many brands use all three. What matters most is a feedback loop between them, and agreement across the organization on how results will be used, so that conflicting numbers lead to tests and decisions rather than arguments. Our comparison of multi-touch attribution and marketing mix modeling goes further into how the approaches differ.
Measurement that doesn't depend on tracking
Privacy rules and platform policies will keep changing. A measurement approach built on following individuals will keep losing ground with each change. One built on aggregate data won't. That is why marketing mix modeling, a method more than half a century old, has become central to how marketers measure today.
Frequently asked questions
Does marketing mix modeling use cookies?
No. MMM works from aggregate data such as sales, spend or impressions by channel, and outside factors like seasonality and promotions. It never needs to identify or follow an individual person.
Did Google end third-party cookies in Chrome?
No. Google dropped its plan to remove third-party cookies from Chrome. But Safari and Firefox block them by default, and Apple's App Tracking Transparency, introduced in 2021, sharply limited tracking in iOS apps, so user-level measurement is far less complete than it once was.
How should I act on marketing mix modeling results?
Channels that drive more than their share of results are candidates for more spend, and channels that drive less are candidates for cuts. Make those shifts gradually, and use experiments to check findings that differ sharply from what other measurement shows.
What is the difference between MMM, platform attribution and lift testing?
Platform attribution gives fast, detailed reporting inside each ad platform, useful for day-to-day optimization. Lift tests check whether a specific channel or campaign actually caused conversions. MMM gives the cross-channel view used for budget planning. Many brands use all three.