Why we don't score hotels or vacation packages
The pull to add hotels is real — better commissions, an obvious next question from users. We turned it down anyway, and the reasons are structural, not a matter of not getting to it yet.
The most common question we get after "does this work for my route" is "will you do this for hotels too." It's a fair question. Hotel affiliate commissions run several times what a flight click pays, and the honest answer is still no — not later, not once we have more data, but no.
The core claim is flight-shaped
The score is built on a directional bet: prices climb as departure approaches, so the right move is usually to book now rather than wait. That is true often enough for flights to be worth saying in a subject line.
It is not reliably true for hotels. Distressed inventory means a property with empty rooms two weeks out will often drop its rate rather than hold it, and for a large share of stays the honest advice is the opposite of ours: wait. A score that tells people to book now on a vertical where waiting is frequently correct isn't a smaller version of the same product. It's a different, less honest one.
The unit you'd need to score doesn't exist at usable volume
Our confidence gate already refuses to score a route until we've recorded enough fares in the same route/month/booking-window cell — 20 to produce a number, 60 to send an email about it. Flights make that threshold reachable because demand pools: everyone watching JFK to Lisbon in March shares the same cell.
Hotels don't pool that way. Everyone watching "Lisbon in March" spreads across a few hundred individual properties instead of one comparable fare. Score at the property level and almost every cell sits permanently below our threshold — the gate would refuse nearly everything, forever, which is a worse product than not building it. Score at the city level instead and the number pools fine, but it can no longer tell you whether the specific place you want is a good price, which was the entire point of asking.
It would break the one metric we're actually watching
Right now we track a single number to decide whether any of this is worth continuing: the open rate on the third alert someone gets. That number only means something if every alert is answering the same kind of question in a comparably reliable way.
Mix in a second vertical with a different hit rate and a different failure mode, and the open rate stops being readable. We would not be able to tell whether the score itself is any good, or whether we'd just diluted a real signal with a weaker one from somewhere else. That is not a cost we're willing to pay to add a second vertical before we know the first one works.
Packages fail even before that
Vacation packages don't get as far as the pooling problem. A bundled price is opaque by construction, and the mix of flight, room and whatever else changes between quotes even for the "same" itinerary. There's no stable unit underneath the price to build a percentile from, which means there's nothing honest a score could say about it.
What we're doing instead
None of this means we think flights are as deep as this gets. The additions on our list — nearby-airport substitution, date-shift savings, eventually alerting on a price drop after you've already booked — are all still flights. They extend the one claim we can actually stand behind rather than adding a second claim we can't. See how the score works for what that claim is today.