Marketing mix modeling has had a genuine renaissance over the past two years. The discipline that was supposed to be replaced by digital attribution has instead come back as the centre of gravity for serious marketing measurement, while attribution has settled into a smaller, more honest role. The May 2026 picture is more interesting than the simple “MMM is back” framing suggests.

The reason MMM came back: the assumptions underneath digital attribution stopped being true at the same rate the privacy environment changed. Cookie deprecation, tightened mobile measurement frameworks, walled-garden measurement opacity, and the general fragmentation of consumer media consumption all eroded the foundation that digital attribution rested on. Marketers who had built their measurement entire stack on attribution found themselves with a measurement framework that had become directionally unreliable.

MMM doesn’t depend on the user-level identifier infrastructure that attribution does. It works with aggregated marketing spend data, aggregated outcome data, and statistical inference across them. The assumptions are different, and the assumptions held up better through the privacy transition than the attribution assumptions did. That’s the technical reason for the renaissance.

The commercial reason is also worth being honest about. Several of the largest digital advertising platforms quietly built their own MMM tools and offered them — sometimes free — to their largest advertisers. The platform-MMM tools are convenient and generally reasonable, but they’re also operating with a structural conflict of interest. The independent MMM has come back partly because chief marketing officers got uncomfortable about relying on platform-built measurement to assess whether platform spending was working.

What MMM actually does well in 2026: medium-frequency budget allocation across channels. The output is most useful at the planning level — how much should we spend on paid social versus paid search versus offline — rather than at the campaign-tactical level. The model gives marketers a defensible, auditable view of marginal returns by channel, which is the question that matters most for budget decisions.

What MMM still struggles with: short-cycle campaign optimisation, very granular sub-channel decisions, and any situation where the data history is short or the underlying business has changed structurally. Trying to use MMM to decide which creative variant to run is the wrong tool. Trying to use it to decide where to direct the next $5M of media spend is exactly the right tool.

The hybrid model that’s actually working in 2026: MMM for portfolio-level allocation, incrementality testing for specific high-stakes interventions, attribution (in its more honest 2026 form) for last-mile optimisation within channels. Each tool does what it’s good at, and the overall picture is closer to reliable than relying on any single one. The marketers who have built genuine measurement stacks rather than choosing one approach are producing better outcomes.

The technology around MMM has improved meaningfully. Bayesian MMM frameworks have made the modelling more practical to run frequently. Open-source tools (notably the Meta Robyn framework and Google’s Meridian) have lowered the barrier to credible MMM execution. The bench of analytics professionals comfortable with these tools has grown over the past two years. The discipline is in a better place operationally than it was when it was the dominant measurement approach in the 2000s.

The scepticism that’s worth keeping: MMM is a model. The output is sensitive to the assumptions, the input data quality, and the modelling discipline. Bad MMM is worse than honest “we don’t know” because it produces false confidence in numbers that aren’t load-bearing. The vendors selling MMM as a solved problem are oversimplifying. The marketers using MMM well are treating it as a serious analytical exercise that requires real expertise.

The ROI conversation has also improved. The 2024 marketing measurement conversations that frustrated CFOs — different attribution platforms producing different numbers for the same campaign, walled garden numbers that couldn’t be reconciled — have settled into something more disciplined. MMM-anchored measurement provides a single coherent view across channels that can be defended in a budget conversation. The CFO-CMO measurement conversation has gotten better in companies that have invested in MMM properly.

For marketing teams in 2026 thinking about measurement strategy, the practical observations are that MMM should be at the centre rather than at the periphery, that platform-built MMM is convenient but should not be the only view, that incrementality testing is a critical complement, and that attribution still has a role but a smaller one than it had in the cookie era.

The longer-term direction is clear. Measurement is moving toward portfolio-level statistical inference informed by occasional high-fidelity tactical experiments. The tooling is maturing fast. The teams that have invested in real MMM capability over the past two years are running rings around the teams that haven’t. The gap is widening rather than closing, and the next year of budget conversations will make that more visible.