Big question
When attribution, experiments and mix models disagree, which one moves the money?
The position
Experiments outrank models and models outrank attribution, and you write that order down before the numbers arrive rather than after. An uncalibrated mix model is an argument rather than a measurement.
Why it is live
The three methods answer different questions over different horizons, and no published work grades them against each other. Meridian and Robyn now both take experimental results as calibration input, which is the industry conceding the ordering rather than proving it.
The case for the ordering
DocumentedAn experiment produces a counterfactual and the other two do not, and both major open source mix model tools are built to be calibrated by experiments.
The case against
UnknowableThe best published blind cross-advertiser evaluation this subject has is from 2002, and no evaluation of that kind has ever been published for a mix model, so the ordering is reasoning rather than a tested result.
Where the weight sits
For the ordering, on reasoning rather than on evidence. Use it and say which of those two it is, because the distinction is the credibility.
What would change this
A blind cross-advertiser evaluation of mix models, run by a party selling none of them, with the designs and the results published.
Where this is taught
Connected Evidence
Sources
- IAB measurement leadership summit recap, 23 Jul 2026secondary
- Google Meridian documentation, accessed 5 Sep 2026primary
- Meta Robyn documentation, accessed 5 Sep 2026primary