400 €Attributed revenue÷50 €Spend=8
Incrementality
One question: what would have happened without this campaign? Incrementality compares exposed people with a control group, isolates the media’s real effect, and tests the MMM against reality.
What is incrementality?
Take a campaign. Of the people who saw it, 20% buy your product. The instinct is to credit all 20% to the campaign. Yet some unexposed people buy too: they already knew you, saw another ad, or would have come anyway.
The campaign’s real effect is therefore not the total observed among exposed people, but the gap versus a comparable control group. An incrementality experiment establishes this causal link. Its limitation: it measures one lever over a defined period, not the dynamics of the whole mix.
Exposed group4 observed purchases
Unexposed group3 purchases without campaign
4 − 3 =11 incremental purchase
ROAS or iROAS: the gap in one example
A campaign costs €50. Four €100 purchases are observed among exposed people. The control group shows that three of those purchases would have happened without the campaign.
400 € − 300 €Attributed revenue − Revenue without campaign÷50 €Spend=2
The number to use for decisionsWhy attribution alone is not enough
Attribution distributes conversion credit across touchpoints. It measures relative efficiency, not causal impact. Multi-touch does not solve the underlying issue: every platform claims conversions in its own reporting. Without an external reference, their combined claims can exceed reality.
Calibrated attribution →Claimed sales1 000 000
≠
Actual sales500 000
Four method families, two of which scale
All four answer the same counterfactual question through different designs. The last two rely more heavily on existing data, making them easier to repeat.
Geo lift testing
Compare two similar geographic areas, one exposed to the campaign and one unexposed.
A/B lift testing
Create two audience segments within one media platform and measure the gap between them.
AI-accelerated
Synthetic control
Pause one lever and use an algorithm to estimate what would have happened without the pause.
AI-accelerated
Spend variations
Move spend up and down at a constant total budget, then calculate the lifts caused by those changes.
How these results enter the model
mediaROI does not run the tests. Results you already have — produced by your teams, media platforms, or measurement partners — enter the Bayesian MMM as priors. The model is informed by an observed result for a lever and period without being forced to reproduce it.
The more a lever is documented through tests, the better the model can separate channels that move together. Past learnings stop being archived slides and become a statistical input to the model.
Incrementality results
Priors
Bayesian MMM
Calibrated attribution
Results supplied by your teams, media platforms, or measurement partners.
Three methods that check each other
None is sufficient alone. When all three converge, you can allocate budget without debating the method.
MMM
The holistic view and source of truth: every lever, the baseline, and external variables.
Calibrated attribution
The daily, campaign-level view, corrected by coefficients from the MMM.
Incrementality
The point-in-time causal check that tests the model against reality and informs it.
Discover the platform
30 minutes to walk through a complete model and ask your questions