Media mix modeling: why Mastercard and Uber are betting on an old tool in a new era

Media mix modeling, or MMM, spent roughly a decade being treated as a relic. Digital marketing arrived with click-through rates, last-click attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition →, and the seductive promise that every dollar could be tracked to a conversion. MMM, which uses statistical regression to estimate how each marketing channel contributes to sales over time, seemed slow, backward-looking, and frankly unnecessary. Then cookies started disappearing, walled gardens stopped sharing data, and performance marketing began delivering diminishing returns. Now MMM is back on the agenda, and the companies leading its revival are not academic researchers or consultancies. They are Mastercard and Uber.

Why this matters specifically for CMOs

The CMO's job has always involved justifying marketing spend to a CFO who wants proof. For a decade, digital attribution seemed to solve that problem. You could show that a paid search ad generated a click that led to a purchase, and the case was closed.

The problem is that this model is structurally incomplete. It overstates the contribution of lower-funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.View full definition →, easily tracked channels (search, retargetingretargetingShowing ads to users who have previously visited your site or interacted with your brand, to bring them back and drive conversion.View full definition →) and understates the contribution of brand-building activity (TV, sponsorships, out-of-home) that generates demand weeks or months before anyone clicks anything. A CMO who relies exclusively on last-click or even multi-touch attributionmulti-touch attributionA method that distributes conversion credit across all marketing touchpoints in the customer journey, rather than crediting only the first or last interaction.View full definition → will systematically underinvest in brand and eventually pay for it in declining conversion rates as the brand weakens.

This is precisely the tension that Mastercard and Uber advertising leaders addressed in Adweek's "The Innovation Playbook" series. Both companies are operating in markets where performance marketing alone cannot sustain growth. Mastercard competes on brand perception in a category where products are technically similar. Uber operates in a marketplace where both sides of the platform, riders and drivers, need brand reassurance, not just promotional incentives. In both cases, the investment in brand activity is substantial and impossible to attribute through conventional digital tracking. MMM is the only credible tool they have to model that contribution.

For a CMO at this level, the stakes are specific: being able to defend a $50 million brand sponsorship in front of a board that expects attribution data, or making a media reallocation call across ten channels without waiting six months for a controlled experiment to conclude.

How MMM actually works

The core mechanism is regression analysis. You feed the model historical data: weekly sales figures, spend by channel, pricing, seasonality, competitor activity, macroeconomic indicators. The model then estimates the marginal contribution of each variable to sales, isolating how much of the revenue lift in any given period can be attributed to TV spend, digital display, paid social, and so on.

A simplified example: Uber in the UK runs a brand TV campaign across four weeks in February 2026, combined with paid social and search. Rides completed rise 12% over that period. But February is also Valentine's Day, there is a cold snap, and Uber increased driver incentives. MMM separates each of these drivers statistically, assigning a coefficient to the TV campaign while controlling for weather, seasonality, and the incentive program. The output tells Uber's marketing team roughly how much incremental revenue the TV campaign generated, and at what cost per incremental ride.

This is different from last-click attribution in one fundamental way: it accounts for channels that never produce a direct click but still move the needle. It also captures carryover effects, the fact that TV exposure this week may influence a purchase three weeks from now, through what modelers call the adstock or decay function.

The historical weakness of MMM was data granularity and compute time. Traditional models ran on monthly aggregates and took weeks to build. What has changed in 2026 is the combination of better first-party datafirst-party dataData collected directly from your own customers and prospects through your own channels: your most reliable and privacy-compliant source.View full definition → infrastructure, cloud computing that runs regression at scale, and AI-assisted model calibration that can cross-validate MMM outputs against geo-based incrementality experiments in near-real time. Mastercard's analytics teams, operating across dozens of markets simultaneously, can now run MMM at a frequency and resolution that would have been prohibitively expensive five years ago.

When to use MMM and when not to

MMM is the right tool when you have a diverse media mix that includes channels where direct attribution is impossible or unreliable. If you run TV, out-of-home, radio, sponsorships, or influencer campaigns at meaningful scale, last-click attribution is not giving you an accurate picture. Full stop. MMM is also the right tool when you need to make portfolio-level budget decisions across channels, rather than optimizing within a single channel.

It is not the right tool if your marketing is concentrated in two or three trackable digital channels and your sales cycles are very short. A direct-to-consumer brand spending 90% of its budget on Meta and Google, with a three-day purchase cycle, can extract more actionable signal from platform-native experiments and incrementality testing than from MMM.

There are also honest limitations to acknowledge. MMM requires a substantial history of consistent data, typically two to three years minimum to detect seasonality patterns reliably. It is also only as good as the data inputs: if your offline sales data is patchy or your spend records are inconsistent, the model will reflect those gaps. And MMM outputs are probabilistic ranges, not precise figures. Presenting an MMM result to a CFO as "our TV campaign generated exactly $4.2 million in incremental revenue" misrepresents what the tool actually produces.

Uber and Mastercard are not using MMM because it is simpler than digital attribution. They are using it because their businesses have outgrown the explanatory power of digital attribution alone. Brand-building investment at the scale these companies operate requires a measurement framework that can hold brand and performance together in the same model, and MMM, updated with better data pipelines and AI-assisted calibration, is currently the most credible way to do that.

The CMO who masters MMM in 2026 gains a specific advantage: the ability to defend brand investment with a rigorous financial argument, not just a creative one. That is what keeps the brand budget off the table when the CFO starts looking for cuts.