MMM vs MTA vs incrementality is not a contest. Which one to trust depends on the decision. Use attribution to steer inside a channel day to day, an incrementality test to ask "does this spend cause sales?", and MMM to split a budget across channels over time, if you have the data for it.
Many guides end with "use all three". Fair, but not much help when you have one analyst and a budget meeting on Thursday. So here is the version with a rule attached.
Multi-touch attribution (MTA) splits credit for a sale across the ads and touchpoints a customer passed through. It is a rule for sharing credit. It is not proof that the ads caused the sale.
Incrementality is the effect that would not exist without the activity. You usually estimate it with an experiment: a holdout test, where some people or regions do not see the ads, or a geo test, where you compare regions.
Marketing-mix modelling (MMM) is a statistical model of total sales against spend in each channel over time. It also tries to separate out seasonality, pricing and promotions. It works on aggregates, not on individual users.
An agency guide sums up the roles roughly as speed for attribution, proof for experiments and a cross-channel view for MMM1. That is a reasonable starting split. The detail below is where it gets useful.
None of the columns wins. Each is good at one job and poor at the other two.
Because attribution measures where sales were standing, not what moved them. Researchers compared observational methods with large randomised experiments run on Facebook. They found that observational methods often failed to reproduce the experimental results2. Attribution models are observational too. They see who clicked, not who would have bought anyway.
That does not make attribution useless. The same agency guide calls it fast, granular and tied to automated bidding1. Just don't use it to answer the causal question.
You are adjusting bids, pausing ads or choosing between creatives inside one channel. The decision is small, frequent and reversible. Speed matters more than proof. Write down which attribution model and window you are using, and don't compare numbers that use different ones.
A plausible result would change the budget. For example: "Is retargeting adding sales, or collecting credit for sales we had anyway?" Tests put some revenue at risk in the holdout, and they take time. If no result would change your plan, don't run one.
You spend across several channels, you have a long and varied history, and the question is how to split next year's budget. If your spend has barely changed, the model has little to learn from. Whether you are ready is a separate question, covered in do I need marketing mix modelling.
None of the three fits your volume or budget. A MER trend, a before-and-after comparison or a post-purchase survey can still tell you something. Each has a known bias. Surveys, for example, record what customers remember, not what caused the purchase. See post-purchase attribution survey for how to combine one with MER.
Often. A MarTech column by a measurement consultant argues for confidence from methods converging rather than from any single output, and says conflicting results often reflect different sides of customer behaviour rather than a broken method3. We agree. If attribution says a channel is your best, and a test says it adds little, you have learnt something important about that channel.
Triangulation means using the methods to check each other. It does not mean averaging three numbers until the meeting ends.
The industry is still working this out: IAB's State of Data 2026 report is devoted to measurement across attribution, incrementality and MMM4. We don't claim to have settled it either.
For most brands, our default order is this:
The last step in every case is the same. Turn the answer into a budget move based on marketing budget allocation by marginal return: where the next euro earns most, not where the average looks best.
For ZFD: Pick the method by the decision, not by fashion. Then check the answer against your sales and your margin, because that is where the money ends up.
Neither is better. They answer different questions. Attribution steers daily work inside a channel. MMM estimates how the budget should split across channels over time.
Not at once. Start with the decision you face, then pick the method that answers it. In our view, many brands can go a long way with attribution, a MER trend and the occasional test.
No. A test answers one question about one thing at one time. MMM tries to describe all channels together. Tests can be used to check whether an MMM's estimates are plausible.
Using more than one method and comparing their answers. Where they agree, you gain confidence. Where they disagree, you have found something worth investigating.
Attribution gives credit to ads that were near a sale, including sales that would have happened anyway. A test removes the ads and measures what is left.
“Their engagement is impressive.”
See how far your own numbers drift. Put them through the Is your ROAS real? calculator, then book a call about the decision you need to make.
