MMM or attribution? They answer different questions. A plain-language guide to what marketing mix modelling and attribution each measure, their blind spots, and when a UK brand needs which.
Marketing mix modelling is having a revival. Privacy changes have made user-level tracking leakier every year, and MMM, which never needed cookies in the first place, suddenly looks modern again. That has produced a wave of content pitching MMM against attribution as if one must replace the other. They will not, because they answer different questions.
Two different questions
Attribution asks: which touchpoints did this customer interact with before buying? It works at the level of individual journeys, stitched together from tracking data, and its natural habitat is day-to-day optimisation inside channels: which campaign, which keyword, which audience.
Marketing mix modelling asks: when we spent more here, what happened to total sales? It works at the level of aggregate statistics, correlating spend patterns with revenue outcomes over a long period while accounting for seasonality, pricing and everything else that moves sales. It never sees an individual user, which is exactly why privacy changes cannot break it.
What each one hides from you
Attribution's blind spots are well documented, and we cover them properly in our attribution guide: it over-credits channels that touch users near purchase, under-credits everything upstream, and cannot see anything that happens outside its tracking, from offline sales to dark social. Fundamentally, it records correlation in journeys, not causation, which is why platform-reported numbers always flatter the platform reporting them.
MMM has quieter blind spots. It needs meaningful variation in spend to learn from, so a channel run at a steady budget for two years teaches the model almost nothing. It is slow, updating monthly or quarterly rather than daily. It struggles with small channels whose signal drowns in noise. And a badly specified model produces confident nonsense, which is worse than no model at all.
The hierarchy that actually works
For most mid-market UK businesses, the practical stack has three layers. Attribution handles daily and weekly optimisation inside channels, treated as directional rather than gospel. Incrementality testing, which we explain in plain language for CFOs here, provides the causal ground truth that calibrates everything else, because a holdout test measures what actually happened rather than what a model believes. MMM sits on top for businesses spending enough across enough channels to justify it, informing quarterly and annual budget allocation.
The three disagree constantly, and that is fine. When attribution says a channel is your best performer and an incrementality test says half of that revenue would have arrived anyway, the test wins. When MMM says TV is driving search volume that attribution credits to PPC, you have learned something no single tool could tell you.
When is MMM worth it?
Honest thresholds: meaningful spend across four or more channels, at least two years of clean sales data, and the willingness to act on quarterly answers. Below that, most businesses get more value from disciplined tracking and measurement plus periodic incrementality tests. The worst outcome is buying an MMM because it is fashionable, then continuing to allocate budget on last-click numbers anyway. Measurement only earns its cost when a decision changes because of it, the standard we hold all analytics work to.
The question behind the question
Teams asking MMM versus attribution are usually really asking: can I trust my numbers enough to move budget? That trust does not come from any single tool. It comes from triangulation: fast, imperfect attribution for tactics, tests for truth, and modelling for strategy, each correcting the others.
If you are trying to work out what that stack should look like at your scale, talk to us. We will tell you honestly which layer your spend justifies, and which expensive acronym you can safely skip for now.
