For a decade, Marketing Mix Modelling was the thing you did before digital gave you "real" data. Slow, expensive, backward-looking — it was the methodological equivalent of dead reckoning in an era of GPS. Then digital attribution arrived and everyone stopped thinking about it.
The problem is that digital attribution was never the GPS it claimed to be. It was a really confident-sounding dead reckoning.
Why digital attribution broke down
Last-click attribution was always a simplification. Multi-touch models like linear, time-decay, and data-driven attribution were marginally better but still working within a closed system — they could only model channels they had visibility into. The walled gardens got higher. iOS 14 gutted signal quality. Cross-device journeys became opaque. What was left was a measurement framework that confidently allocated credit to a shrinking subset of actual touchpoints.
The platforms were happy to fill the gap with their own attribution — which unsurprisingly showed their own channels in a flattering light. Advertisers ended up in a measurement environment where everyone was simultaneously claiming credit for the same conversion.
What modern MMM solves
MMM doesn't depend on user-level tracking. It works at the aggregate level, correlating marketing investment inputs against business outcome outputs — controlled for seasonality, pricing, competitive activity, and external macroeconomic factors. Done well, it produces channel-level elasticity curves that tell you not just what drove growth historically, but what the marginal return on the next dollar of spend will be across each channel.
Modern implementations — using Bayesian inference, better priors from industry benchmarks, and tighter integration with incrementality test results — have addressed most of the accuracy criticisms of legacy MMM. A well-calibrated model can now be updated on a rolling basis, making it a live planning tool rather than an annual retrospective.
The hybrid approach
The most robust measurement setup isn't MMM or incrementality testing — it's both, calibrated against each other. MMM provides the long-run strategic view: channel saturation curves, diminishing returns thresholds, budget allocation optima. Incrementality testing validates specific tactical decisions and feeds coefficients back into the model to improve accuracy over time.
Together, they close the loop between what happened and why, and between short-term results and long-term efficiency. The businesses using both are operating with a material information advantage over those still relying on platform-reported ROAS as their primary decision input.
MMM isn't back because it's fashionable. It's back because it was always the more honest approach — and the alternatives have now been shown to be significantly less reliable than they claimed.