Travel inventory is described by people whose job is to make it sound appealing. That produces prose, and prose is the right medium for persuasion and the wrong one for comparison.
The result is a results page where price and star rating are filterable because they are numbers that arrived structured, and everything a traveller actually needs to check is a paragraph.
Why the filters appear broken
A filter can only exclude inventory that has been tagged. Most systems, faced with untagged inventory, keep it in the results rather than dropping it, because dropping it would hide bookable stock. So a filter that almost nothing is tagged against excludes almost nothing.
From the traveller's side this is indistinguishable from a broken filter, and they stop using them. Which means the one mechanism you have for helping them narrow forty options is now something they have learned to ignore.
Accessibility and family requirements are the ones travellers cannot compromise on, and usually the ones nobody made filterable.
What a traveller can filter on, against what they need to know
illustrative of the shape of travel inventoryThe attribute recorded almost everywhere is the one almost nobody decides on. The measured figures on this site are the laptop shelf's, on the method page.
View as table
| Value | |
|---|---|
| Star rating | 94% |
| Late check-in | 44% |
| Cot available | 31% |
| Step-free access | 22% |
The cost lands after the booking
In retail, an unfilterable attribute costs you a sale you never see. In travel it frequently costs more than that, because the traveller books anyway and finds out on arrival.
Someone who needed step-free access and got three flights of stairs generates a complaint, a refund, and a review that outlives both. The attribute would have been cheaper than any one of those.
Most of it is extraction, not acquisition
The encouraging part is that the facts are usually already written. "Lift to all floors" is in the description. "Cots available on request" is in the description. They are sentences rather than fields, which is exactly the case that is cheap to recover from text you already hold.
- Take one property set and list the constraints travellers cannot compromise on: access, family, late arrival, parking
- Check which of those exist as structured values rather than as sentences
- For the sentences, that is extraction, and it is usually the largest and cheapest share
- Where a fact arrives from one supplier and not another, that is mapping, which is a different cost again
Whether a recovered fact is accurate enough to publish is a judgement only you can make, particularly for accessibility, where being wrong is worse than being silent. What a measurement can tell you is which absent facts are costing the most, so the verification effort goes somewhere useful.