Hotel booking API provider: judge it at the front desk

The first support ticket after go-live is rarely about search. It is a guest standing at a front desk at eleven at night, holding a confirmation number on their phone, and a receptionist who types it in twice and says the hotel has no booking under that name. The guest paid you. Your system says the booking is confirmed. Both are telling the truth, and that is the problem.
That ticket is what choosing a hotel booking API provider is really about. Most teams choose on the part they can see in a demo: how many properties come back for a city, how clean the availability endpoint is, how fast the SDK gets them to a first result. Those things matter. They are also roughly the same everywhere, and none of them decides what happens at the desk.
Why coverage is the wrong first question
A search that returns more hotels feels like a better API. But a property you can show and a booking the hotel will honour are two different promises, made by two different parts of the provider's business. The first comes from a content feed. The second comes from contracts, connectivity to the hotel's own systems, and a support desk that someone staffs on a Saturday.
If you are deciding how many providers to connect at all, that question has its own article, and the mapping problem it describes is real. This one assumes you are choosing the provider, and asks what its booking response is worth after the guest has paid.
The five properties of a hotel booking API provider that show up after payment
Strip a provider's documentation down to the responses your site will store once money has moved. These are the rows where two providers with identical search endpoints turn out to be different products.
| Property | What a good answer looks like | What it costs you when it is missing |
|---|---|---|
| Cancellation terms | Structured: a deadline, a penalty and the time zone the deadline is in | Free text that someone has to read, and a refund decided by a guess |
| Who collects the money | Stated per rate: prepaid to you, or paid at the hotel, and who is merchant of record | A guest charged twice, once by you and once at check-out |
| The hotel's own confirmation number | Returned, or delivered later through a documented channel | Only the provider's reference, which means nothing at the front desk |
| Price change between availability and book | An explicit new price or a clear refusal, never a silent difference | The gap between the shown rate and the booked one comes out of your margin |
| No record at arrival | A named procedure and a contact that answers outside office hours | Your staff on the phone to the hotel while the guest waits in the lobby |
You can disagree with the order. You cannot ship a site that takes cards without an answer to each row.
Cancellation is data, or it is a liability
Look closely at the first row, because it fails most quietly. Many responses carry the cancellation policy as a sentence: free cancellation until a date, then the first night is charged. That reads fine to a person. Your code cannot act on a sentence without parsing it, and parsing it is how mistakes reach production.
The detail that bites is the time zone. A deadline of midnight means one moment in the hotel's city, another in your server's, and a third on the guest's phone. If the provider does not say which, your site will one day tell a guest they can still cancel for free, refund them, and find the hotel charged the penalty anyway. Nothing threw an error. The money simply left.
Non-refundable rates deserve the same scrutiny. A provider that marks them clearly, in a field and not only in the rate name, has thought about the site that sells them. Ask to see both kinds in a real response before anything is signed.
Whose name is on the guest's card statement
The second row sounds like accounting. It is actually a customer-service question. When a rate is prepaid, the guest pays you, you or the provider pay the hotel, and the hotel should charge the guest nothing beyond extras such as the minibar or a city tax collected at the desk. When a rate is paid at the hotel, your site takes no money for the room at all, and the guest's card is charged by the property, often at check-in or check-out.
Both models work. What does not work is a checkout that cannot tell them apart. If the provider returns the two kinds of rate in the same shape, with the payment model buried in a description, your site will sooner or later charge a guest for a room the hotel then charges again. The guest sees two lines on a statement and calls the name on the first one, which is yours.
So the answer you want is per rate, in a field, and it includes who is merchant of record. That last part decides whose descriptor appears on the card and who handles a chargeback.
Questions to put in writing before you sign
Demos show the happy path. Written answers show whether the provider has met the other ones. Send these and keep the replies:
- Show me one booking response for a non-refundable rate and one for a rate with a free-cancellation deadline.
- In which time zone is that deadline expressed, and where in the response does it say so?
- When does the hotel's own confirmation number arrive, and how do I receive it if it comes after the booking call?
- For a pay-at-hotel rate, what exactly does the guest see on their card, from whom, and when?
- When a guest arrives and the hotel has no record, who do they or I call, and at what hours?
A provider that answers the last question with a procedure has handled it before. One that answers with a feature name has not. The same holds on the flight side, where what stands behind the API matters more than the API.
What getting it wrong costs
Each failure is small and repeats. A refund paid out of your margin because a deadline was read in the wrong time zone. A guest charged twice because a pay-at-hotel rate looked prepaid on your checkout. A member of staff who spends the start of every week phoning hotels to reconfirm bookings by hand, because the provider never returned a number the hotel recognises. None of it shows up in a coverage count, and all of it shows up in the support inbox within the first month of real traffic.
The work to avoid it is an afternoon: read five booking responses with the table above beside you.
Where we sit
Tekravel exposes hotel and flight APIs. Access starts as a conversation rather than a self-serve key, services are switched on per tenant, and supplier content reaches a tenant through supplier groups rather than one-to-one deals. We do not publish property counts on this page, and we would rather you ask us the five questions above than take a number on trust. If you are at that stage, the API page is where the conversation starts.