What Can Marketing Mix Modeling Really Do — and Where It Fails

Q: What can marketing mix modeling do?

Marketing mix modeling (MMM) is meant to show which parts of the marketing budget actually drive sales, enabling fact-based budget reallocation. It can estimate the contribution of individual channels over time and simulate scenarios for how to split the budget. What matters most, though, is whether the underlying model is causal: only then does it capture true effect rather than mirroring correlations in the historical dataset.

Q: What does causal marketing mix modeling do differently?

Causal MMM models the full system of effects between media investment, perception and market performance — including indirect effects, delays, saturation (inverted-U curves) and interactions. SUPRA uses Universal Structure Modeling (USM) for this. In a published study, the causal model explained around 20% of the change in brand strength from advertising pressure, versus 1% for conventional methods.

Q: What data does marketing mix modeling need?

Conventional MMM uses aggregated time series of media spend and sales. Causal MMM can additionally incorporate perception and tracking data to make the indirect pathways of effect visible. More important than a particular volume of data is that the relevant intermediate variables (such as brand perception) are modeled as well — otherwise the actual advertising effect stays invisible.

Marketing mix modeling promises to show which budget works. Conventional linear MMM often fails — the effect of a strong channel like TV can appear as 'zero'. Here's why that happens, and what causal MMM does differently.

Dr. Frank Buckler Founder, SUPRA · 7 min read

The short answer

Marketing mix modeling (MMM) is meant to show which parts of the marketing budget actually drive sales — and so enable fact-based budget reallocation. It can do that, but only under one condition: the underlying model has to be causal. Conventional, linear MMM misses indirect and non-linear advertising effects — and as a result measures the wrong thing, systematically.

Why the TV effect disappears in a linear model

Advertising rarely works directly and instantly. It works indirectly — through brand perception, with a delay, in interaction with other channels, and with saturation (past a certain point, more pressure buys little). A linear MMM can't represent those pathways. The consequence: the effect of a strong channel like TV shows up in the model as near zero — and a team reallocates budget on the basis of a false signal.

That exact moment — "this can't be right" — is the heart of the problem. The model is computed cleanly and is significant, yet it's wrong. It's the same confusion of correlation and cause that undermines conventional market research: measuring what is visible and stated, not what actually causes behaviour.

Case study · from "The End of the KPI-Illusion" (Dr. Frank Buckler)

An automotive advertising study: 20% vs 1%

20%

explained by the causal model

explained by conventional methods

100–200%

effectiveness upside on the top ad days

In a study drawing on data from Thomson Media Control, YouGov BrandIndex and the University of Münster, the causal model — built with Universal Structure Modeling — explained roughly 20% of the change in brand strength from advertising pressure, versus about 1% for conventional methods. On the most effective 50% of advertising days, the effectiveness upside ran 100–200% higher. Michael Trautmann, the former Audi marketing director, commented on the findings. It is the data-based answer to Henry Ford's old line: "Half my advertising is wasted — I just don't know which half."

The principle

"A statistically significant model can be very significant and very wrong."

Dr. Frank Buckler, 10x Insights

What causal marketing mix modeling does differently

Causal MMM models the entire system of effects between media investment, perception and market performance — including indirect effects, delays, saturation and interactions. SUPRA uses Universal Structure Modeling (USM) for this, a causal AI method. In practice, that delivers:

The true contribution of each channel — including the ones that work indirectly, like TV or the brand itself.
Saturation curves instead of linear assumptions — the point at which extra pressure stops paying off.
A budget simulation that shows which reallocation actually lifts sales.

The same foundation carries our work on Decision Intelligence for Pricing, and the reasoning is developed at length in Dr. Buckler's book The End of the KPI-Illusion.

Frequently asked questions

What can marketing mix modeling do?

It's meant to show which parts of the budget drive sales and enable fact-based budget reallocation. It can estimate channel contributions over time and simulate scenarios — what's decisive is whether the model is causal, so it captures true effect rather than correlations in historical data.

Why does conventional marketing mix modeling fail?

Because it usually models effects linearly and misses indirect and non-linear advertising effects. The effect of a strong channel like TV can then appear as near "zero" — and teams reallocate budget on the basis of a false signal.

What does causal marketing mix modeling do differently?

It models the full system of effects, including indirect effects, delays, saturation and interactions (USM). In a published study, the causal model explained around 20% of the change in brand strength from advertising pressure, versus 1% for conventional methods.

What data does marketing mix modeling need?

Conventionally, aggregated time series of media spend and sales. Causal MMM additionally draws in perception and tracking data to make the indirect pathways visible. What matters is that intermediate variables like brand perception are modeled — otherwise the real advertising effect stays invisible.

Find out what your MMM is missing

Bring your media and tracking data. We'll show you whether a causal analysis reveals advertising effectiveness your current model reports as "zero".

Get my AI Diagnostic →