Your MMM Is Probably Lying to You

There's a moment most marketing leaders recognize. The Marketing Mix Modeling report comes back, the outputs look legitimate. The budget recommendations seem on point. And then someone asks “What about our spend around CTV? What about creator spend? How does that affect our retail media spending budget?”  And the answer is: “Well, the model doesn't really capture that.”

Marketing Mix Modeling was built in an era when media was simpler. You had television, print, radio, some digital. The channels were measurable, the data was relatively clean, and the models reflected reality well enough to make good decisions from. 

That era is over. The media landscape has fragmented dramatically, and most MMMs have not kept up.

The result is a model that looks authoritative, produces confident outputs, but is missing a significant portion of what's actually driving your business.

What the Model Isn't Seeing

The IAB's State of Data 2026 report surveyed over 400 senior marketing and analytics decision-makers and found something that should give every brand pause. About 50% of marketers say commerce media and the creator economy is underrepresented in their MMM. 41% say connected TV isn't being captured properly.

Think about what this means in practice. A brand running meaningful investment across CTV, creator partnerships, and retail media is building a model that essentially ignores a significant chunk of its actual media activity. 

These aren't niche channels. CTV is where a growing share of premium video consumption happens. Creator and influencer marketing has become a primary acquisition channel for many retail brands. Commerce media, including retail media networks, sponsored placements, shoppable formats, is one of the fastest growing segments in digital advertising. If your model isn't capturing these, it's not giving you a complete picture of what's driving growth. It's giving you a partial picture dressed up as a complete one.

Why This Happens

The honest answer is that MMMs are only as good as the data that goes into them. And historically, some of the fastest growing channels have also been the hardest to measure cleanly. Creator content doesn't always have clean impression tracking. CTV measurement is still maturing. Commerce media data often lives in retailer walled gardens that don't integrate easily with brand-side models.

So the model takes the path of least resistance. It incorporates the channels with clean, accessible data, like paid search, display, social, and underweights or ignores the ones that are harder to capture. The output looks complete because it covers the channels you're most familiar with. But familiar isn't the same as comprehensive.

The deeper problem is that this bias compounds over time. If your model consistently underweights CTV, you'll consistently underinvest in CTV. The channel gets less budget, produces less signal, and the model has even less data to work with in the next cycle. You end up optimizing toward the channels that are easiest to measure rather than the ones that are most effective. The gap between your model and reality quietly widens.

What a Complete Model Actually Requires

Fixing this isn't just a data problem, although better data integration matters. It requires a different philosophy about what a model is supposed to do.

A marketing mix model shouldn't just reflect the channels you can measure easily. It should be built to reflect how your customers actually behave. Where do they spend their time? What influences their decisions? Which touchpoints matter across the full journey?

That means actively working to incorporate emerging channels even when the data is messier. It means running incrementality tests specifically designed to validate the channels your model is weakest on. And it means treating the model as a living system that evolves as the media landscape does, not a static output you refresh once a year.

The brands getting the most value from their MMM are treating model coverage as an ongoing discipline. They are constantly asking what the model isn't seeing, and building the infrastructure to see it.

A model that confidently tells you the wrong thing is worse than no model at all. You don’t want to work with a model that just produces an answer. You want a model that produces the right answers.

At M-Squared, we build MMMs designed to reflect the full complexity of how modern retail brands actually grow. See How it works.