How Marketing Mix Modeling Is Making A Comeback

Deterministic attribution models map user touchpoints but do not reveal how those touchpoints influenced conversion behavior. Certainty about user behavior and collateral value is overstated, especially when user-level signals disappear. Third-party cookie removal and Apple’s App Tracking Transparency reduce cross-app tracking, with initial data showing U.S. users opt out of tracking 96% of the time. Clean rooms can share user data anonymously and securely to generate insights without compromising privacy, and machine learning can create modeled conversions to infer behavior and patterns. These approaches remain partial because they depend on platform rules and assumptions that may not match business reality. Modern MMM uses aggregated outcomes like sales and conversions, focuses on incrementality, and captures offline, brand, and long-term effects without relying on personalized data.

"Deterministic attribution models looked like the ideal solution. They show us when users hit touchpoints and are supposed to tell us which of these touchpoints causes a conversion. They show us that a user saw an ad, but they don't tell us how it influenced them. They propose certainty of user behavior and collateral value when they are anything but certain. Now that the signal is gone, the illusion is broken."

"The move away from third-party cookies isn't the only cause: Apple's App Tracking Transparency framework, launched in 2021, requires apps to ask users for permission before tracking them across other companies' apps. Initial data suggested that U.S. users opt out of tracking 96% of the time, stripping advertisers of the user-level data they'd relied on for targeting and measurement."

"There are, of course, methods and techniques that can help bridge the gap. Clean rooms allow advertisers and platforms to share user data anonymously and securely to get insights without compromising data privacy. With more access to machine learning, we can also create modeled conversions to infer user behavior and learn patterns. But they are only partial solutions. Both remain constrained by platform logic, with assumptions and limitations based on the rules imposed by the platform you're using. Neither is tailored to the reality of your business."

"MMM may look limited compared to attribution models, but its simplicity is what makes it work. It assesses the impact of your marketing based on the outcomes: sales and conversions. It was designed for uncertainty, in the sense that it does not rely on personalized data. It works with aggregated data, looking at the outcomes of each channel to make logical assumptions about their value. It focuses on incrementality and captures the offline, brand and long-term effects of what you're doing."

Read at Forbes