The Rise of Marketing Mix Modeling

Marketing mix modeling is old enough that it predates the internet entirely. Its roots go back to econometric work on advertising response in the 1960s, and by the 1980s and 1990s it had become standard practice at large consumer packaged goods companies deciding how to split budgets across television, print, radio and promotions. Nielsen and a handful of specialist consultancies built entire businesses around running these regressions for brands that spent tens of millions of dollars a year and needed a defensible answer for where the next dollar should go.

Then digital marketing arrived and, for the better part of a decade, MMM looked like a relic. Why run a slow statistical model on aggregate data when you could track an individual click straight through to a purchase? Google Analytics, launched in 2005, made last-click attribution the default assumption of an entire generation of marketers. Multi-touch attribution vendors through the early 2010s promised even finer-grained credit across an entire customer journey. Against that backdrop, a technique that only produced a channel-level estimate with error bars looked slow and imprecise. Digital budgets and digital talent moved almost entirely toward click-based measurement.

The turn started with privacy, not technology. Apple’s Intelligent Tracking Prevention began cutting cookie lifespans in Safari starting in 2017. GDPR took effect across Europe in 2018 and made tracking without consent a legal liability. And then April 2021 landed the heaviest blow: Apple’s App Tracking Transparency prompt, shipped with iOS 14.5, asked every iPhone user directly whether apps could track them, and most said no. Meta later put a number on the damage, roughly ten billion dollars in lost 2022 revenue attributable to the change alone. Attribution models that depended on seeing an individual’s cross-app behavior suddenly could not see most of it.

MMM had never depended on seeing an individual at all, which is exactly why it started looking relevant again. It only needed aggregate spend and aggregate sales, data that no cookie policy or consent prompt could touch. Meta open-sourced its own MMM tool, Robyn, in 2021, a clear signal that even the platforms most invested in individual-level tracking saw where measurement was heading. By 2024 the momentum was unmistakable: Google released its own open-source MMM framework, Meridian, that November, putting a modeling approach it had spent two decades treating as secondary to click tracking into the hands of any marketer willing to build with it.

What makes 2024’s version of MMM different from the P&G-era original is accessibility. Running a mix model used to require a specialist consultancy and a budget that only the largest advertisers could justify. Open-source tooling, cloud compute and a wave of software built specifically to lower that barrier meant a mid-sized business could run a reasonably sound model without hiring an econometrics team. The technique did not change much. Who could use it did.

The irony is hard to miss. A method built for an era of three television networks and a handful of magazines turned out to be better suited to the privacy-constrained internet than the click-tracking infrastructure built specifically for digital advertising. Attribution chased ever-finer individual precision right up until regulation and platform policy made that precision impossible to sustain, and the aggregate approach that never needed it in the first place was still standing when the dust settled.