Proof
Does Qloo's taste data actually predict new routes?
We tested it the hard way: train on one year, predict the next, and compare against the same model without Qloo. Every number here is out of sample. The limits are at the bottom.
How it works
Every US route, every month
US DOT T-100 data for all 61 airlines (2022–2026) shows which airport pairs gained a brand-new nonstop each year.
Who already makes the trip
US DOT DB1B, a 10% sample of every ticket, shows people flying each pair through connections, and what they pay.
Where the taste already is
For 158 leisure destinations, Qloo's taste graph returns a US-wide heat map of about 10,000 cells. Each airport gets its area's affinity, its rank among all towns, and how much it prefers that place over its other options.
One honest model
A logistic regression scores every possible route using only the previous year's data; the airline matchmaker learns each leisure airline's habits from its own launches.
Limits, stated plainly
- Qloo's data is a 2026 snapshot. Some of the taste it measures may have formed after the routes it scores, which could flatter the backtest. The 2026 forward check is the clean test.
- Taste predicts where leisure airlines launch. It does not predict which new routes survive; we tested that and found no signal, so Nonstop makes no claims about route cuts.
- Qloo improves each airport's ranking but slightly lowers precision when all routes nationwide are pooled into one list (average precision above), so Nonstop is built around the per-airport view.
- Qloo has no destination entry for about 30 small Midwest and Plains airports (for example Green Bay, Fargo, Jackson Hole), so they can't be ranked as destinations yet.
- Results describe group tastes, never individuals. No personal data is sent to Qloo.