The MMM Renaissance and Incrementality: Defending Budgets Without Last Click

Why last click is finally done

Last-click attribution was never precise, but it was convenient for a long time. With tracking restrictions, consent requirements and platform black boxes it has become not just imprecise but systematically biased: channels that harvest demand look brilliant. Channels that create demand look expensive.

Allocating budget on that basis consistently moves money from growth to harvesting — and two years later you wonder why new customer numbers stagnate.

The two tools that belong together

There is no single method that replaces attribution. There is a combination.

• Marketing mix modeling (MMM) delivers the top-down view: what contribution do channels, price, distribution, seasonality and external factors make at aggregate level? MMM is privacy-friendly because it needs no personal data.
• Incrementality testing delivers the proof: does additional budget in a channel create additional revenue? Geo splits, holdouts and controlled switch-off tests answer that causally.

MMM without tests tends toward model faith. Tests without MMM produce isolated findings without an overall picture. Only together do they form a robust steering logic.

What a viable setup needs

• Clean spend data across all channels, including production and agency cost, not just media.
• Consistent outcome metrics — ideally contribution margin instead of revenue.
• Sufficient history, usually at least two to three years on a weekly basis.
• External variables: price, promotion, competition, seasonality, distribution, weather or holidays where relevant.
• A test calendar: a fixed rhythm of planned experiments, not spontaneous one-offs.

The typical mistakes

• Treating the model as truth. An MMM is an estimate with an uncertainty range. Communicate ranges, not decimals.
• Too many channels, too little variance. If budgets have been allocated identically for years, no model can separate effects.
• Underpowered tests. Runtimes that are too short or regions that are too small produce noise.
• Platform numbers as counter-evidence. Self-reported platform conversions are no measure of incrementality.
• No link to decisions. A model that changes no budget decision is expensive reporting.

The build-out in four phases

• Phase 1 – Data foundation (4–6 weeks): consolidate spend, outcomes and external factors, align definitions.
• Phase 2 – Base model (4–8 weeks): first MMM with uncertainty bands, focused on major channels instead of maximum granularity.
• Phase 3 – Calibration (ongoing): planned incrementality tests whose results feed back into the model as priors.
• Phase 4 – Decision routine (ongoing): quarterly budget reallocation based on model plus test evidence, documented and tracked.

How to sell it internally

Finance leaders rarely care about method details but care a lot about decision quality. Argue via risk: last click leads to systematic misallocation. MMM plus tests reduces that risk and makes marketing spend discussable in the same language as other investments — with contribution, range and proof.

Conclusion

The return of MMM is not nostalgic regression, it is the pragmatic answer to a world with fewer individual data points. Combined with real incrementality tests it creates steering that does not just allocate budgets but defends them.