Proof
Proof: what our models actually scored
Everyone publishes their wins. These are the working notes — including a validation failure that would have shipped if we had not held data back.
Fit on data it could see4.85%
Predicting 3 months it never saw3.87%
Our 10% ceiling — above this we do not ship, and you get your money backReal client model, 14 markets, 22 channels
Predicting months it never saw came in below the fit on months it did. That is the signature of a model that learned the business instead of memorizing its history.
3 pages
- What a 3.87% error on unseen data actually provesHiding recent months from a model and making it predict them tells you whether it learned your business or just memorized your history.3 min
- The model that failed, and what it cost to find outWe took a validated model — 3.87% error on unseen months — and split it in two to get finer detail.3 min
- Why we don’t use click-based toolsClick-based tools cannot see offline media, cannot tell a sale your ads caused from one that was coming anyway, and let each platform grade its own homework.3 min