Skip to content
Blog

Travel API comparison: judge the axes, not the vendors

Tekravel Editorial
Written with the help of AI

You have three vendor decks open in three tabs. Each one has a slide with a long list of airlines, a slide with a hotel count, a sandbox, REST and JSON, and a support email. You build a spreadsheet, put a tick in every cell, and the spreadsheet ends in a tie. That is the usual end of a travel API comparison done as a feature checklist, and it is the wrong end, because the differences that will decide your next year only show up after go-live: on the booking that succeeded at the supplier but timed out on your side, on the refund nobody can process, on the airline that was in the list but cannot be ticketed through this route.

This piece is for the developer or technical lead who has to make that call and defend it later. It does not rank vendors. It gives you axes to compare on, a trial plan that spends its time on failure, and a short set of questions that tell you quickly whether you are talking to a supplier or to a reseller of somebody else's supplier.

Why the feature checklist always ends in a tie

A checklist asks whether a capability exists. Almost every travel API can say yes to almost every row, because the rows are written at the level of the marketing page: search, price, book, cancel, ticket, webhooks, sandbox. The question that separates them is not whether cancel exists but what happens when you call it on a ticketed booking twelve hours before departure, on a fare with a penalty, through a route where the content came from a third party. A checklist has no column for that.

There is a second problem. The person filling in the checklist is usually the vendor. You are scoring their self-description.

A travel API comparison built on axes, not vendors

Replace the vendor-by-feature grid with axes you score yourself, from evidence you collected in the trial. The table below is the one we would use. You may weigh the rows differently — a hotel-only business cares less about ticketing — and you should argue with it before you adopt it.

AxisWhat a checklist asksWhat to actually find outWhat a weak answer sounds like
CoverageHow many airlines and hotels?How many can be searched, booked AND ticketed or confirmed through this API, for your marketsA single big number with no split
DepthIs there a cancel endpoint?Which post-booking actions are automated per content source: void, refund, reissue, name correction, ancillaries"Our ops team handles that"
Price integrityIs pricing real-time?How often the price at book differs from the price at search, and how the API tells you"Prices are always live"
Error semanticsAre errors documented?Whether an error tells you what state the booking is in, not just that something failedOne generic error code for every failure
Idempotency—Whether a retried book call can create a second bookingSilence, then "we have not seen that"
Supplier of record—Whose contract, whose IATA number, whose BSP the ticket goes through"We work with all the major GDSs"
Access and supportIs there a sandbox?Who answers when a live PNR is stuck, and how fastA ticket form and no named contact

Two of those rows have no checklist question at all. That is not an accident: they are the rows that cost the most when they go wrong, and the ones a vendor deck is least likely to volunteer.

Coverage is not depth

The airline list is the most misleading slide in any deck. A list of several hundred carriers can mean several different things, and only one of them is what you want.

Some of those carriers come through a GDS and can be booked and ticketed end to end. Some come through NDC connections, where shopping may work well but servicing after the booking — changes, partial refunds, adding a bag — may still be manual for that airline. Some are low-cost carriers whose content is aggregated from another reseller; you can display them, sometimes you can book them, and then the ticket arrives by a route you do not control, with a timeline you cannot promise your customer. And some are on the list because a fare was once returned in a search response.

A carrier you can show but cannot ticket is not coverage. For a retail product it is worse than absence, because the customer picks it, pays, and then waits.

Hotels have the same shape with different words. A property count includes properties the vendor has mapped and properties it has not; static rates that were cached last week next to live availability; rooms on allotment with a release period next to free-sale inventory. Ask for the count that is live-bookable in the cities you sell, and ask how duplicates across their own suppliers are merged — the same hotel appearing three times at three prices is a support ticket waiting to happen.

The practical test is narrow. Write down the twenty routes and twenty hotel cities that make up most of your volume, and ask the vendor to show those specific ones working end to end in the sandbox: search, price, book, ticket or confirm, cancel. A vendor who has the depth will do this without fuss. A vendor who has the slide will start talking about the slide.

What to test in a trial: error paths, not happy paths

Every sandbox books a one-way economy ticket on a quiet route successfully. That is what the sandbox was built to demonstrate. Your production traffic will spend a meaningful part of its life on the other paths, and a trial is the only cheap moment to see them.

Use the trial to run this list, and record what the API returned next to what was actually true afterwards:

  • Price change between search and book. Does the book call fail, succeed at the new price, or return a distinct response asking you to confirm? Can your UI tell the difference without parsing a message string?
  • Expired offer. Search, wait longer than the vendor's offer lifetime, then book. Is the expiry a specific error, or the same error as "supplier down"?
  • Sold out at book. Inventory that disappears between price and book. What state is the booking in, and is any payment authorisation released?
  • Retry after timeout. Send a book request, drop the connection before the response, then send it again. Do you now have one PNR or two? This one test tells you more about a vendor than the rest of the trial put together.
  • Booking created, ticketing failed. Can the API report a PNR that exists but is not ticketed, and does it tell you the ticketing deadline?
  • Partial failure. A multi-room hotel request where one room confirms and one does not. A round trip where the outbound and return come from different sources.
  • Cancellation outside the free window. Is the penalty quoted before you confirm, or discovered on the invoice?
  • Supplier outage. Ask the vendor how to simulate one. If they cannot, ask how their clients learned how it behaves.

For each case you are scoring two things. First, does the API tell you the truth about the state of the booking? Second, can your code act on that truth without a human reading free text? An API that fails clearly is easier to live with than one that succeeds ambiguously.

We wrote in more detail about the flight side of this in what flight booking API docs leave out, and about designing a client for slow hotel suppliers in the hotel availability piece. Both are useful companions to a trial plan.

Questions that reveal a thin aggregator

There is nothing wrong with an aggregator. Most travel content reaches most sellers through one. The risk is the thin one: a layer that resells another layer's API, adds a margin, and owns neither the supplier contract nor the operational desk. When something breaks, the answer has to travel two companies upstream and back.

Five questions bring this out quickly:

  1. Who is the supplier of record for this booking? Whose contract with the airline or hotel does it rest on?
  2. Whose IATA number and BSP does the ticket go through? If the answer is "a partner's", ask who receives the ADM when a fare rule is broken.
  3. When I call cancel, what happens on your side? An automated call to the supplier, or a queue that a person picks up?
  4. Can I see the upstream error? A thin layer tends to flatten every failure into one code, because it does not fully understand the errors it receives.
  5. Who do I call about a stuck live booking, and who do they call? Count the hops.

None of these answers is disqualifying on its own. A partner's IATA number can be fine if the contract says clearly who carries the debit memo. What matters is that the vendor answers precisely and quickly. Vague answers here are the answers you will get later, in production, at a worse hour.

What getting it wrong costs you

The integration is the cheap part. The expensive parts arrive later and do not appear in any estimate.

A duplicate-booking bug means a customer charged twice and a PNR you have to find and cancel by hand, possibly after the free void window has closed. A booking that succeeded upstream while your side recorded a failure means a seat held, then auto-cancelled, and a customer who believes they are booked. A carrier you displayed but could not ticket means a refund, an apology and a customer who does not return. An ADM for a fare rule the vendor's route broke lands on whoever owns the ticket — make sure you know in advance whether that is you.

Then there is the switching cost. Every vendor quirk you absorb — a status code that means two things, a field that is sometimes a string — ends up in your data model. Two years in, the choice of API is no longer an integration you can replace in a sprint; it is the shape of your booking table. That is the real reason to compare carefully now.

How our own API fits this framework

We expose flight and hotel APIs. Access starts as a conversation rather than a self-serve key, which some developers find slow; the reason is that the conversation is where the questions above get asked in both directions — which routes and cities you need, which content you will actually be able to ticket or confirm, and who handles what after the booking. If you are running a comparison, bring this list to it. The travel APIs page is where that conversation starts.

Whatever you choose, choose it from the trial log rather than the deck. The deck tells you what a vendor wants to be. The error paths tell you what it is.

Tekravel Editorial

Travel technology desk

The Tekravel travel-technology desk writes for the trade: agency owners, consolidators and the developers who integrate them. Every article is checked against the platform it describes before it is published.