NDC API vs GDS: what changed for the people integrating it

The ticket was correct. The price was correct. The customer still opened a complaint, because the fare they bought on your site came with no checked bag, and the same flight on the airline's own website, for nearly the same money, came with one. Your search had returned the GDS fare. The airline's site sold its NDC bundle. Nobody on your team had done anything wrong, and the ticket still landed on someone's desk.
That is the practical shape of the NDC API vs GDS question. It is rarely a debate about XML schemas. It is a question about which version of an airline's shelf your customer sees, and what you can still do for them after they have paid.
This piece is written for the developer or technical lead who has to choose — or inherit — a distribution route for flights. If you have never integrated a flight API at all, read what a flight booking API integration actually involves first; the search, price, book and ticket sequence is assumed below.
Three routes to the same seat
There are three ways airline content reaches a booking platform today, and most comparisons collapse them into two. The difference between NDC through an aggregator and NDC straight from the airline matters more to your codebase than the difference between NDC and GDS.
| Classic GDS content | NDC through an aggregator or GDS | Direct NDC connection to one airline | |
|---|---|---|---|
| Who builds the price | Filed fares, assembled by the GDS | The airline, passed through a normalising layer | The airline, in its own flavour of the standard |
| What one integration buys you | Hundreds of carriers in one message format | Several NDC carriers behind one schema | One carrier |
| Bundles, branded fares, ancillaries | Partial; varies by carrier and by how they file | Usually richer, sometimes flattened by the aggregator | Everything the airline chooses to expose |
| Servicing after booking | Mature, uniform, decades of tooling | As good as the weakest carrier behind it | Exactly what that airline implemented, no more |
| Ticketing and settlement | Ticket issued on a plate you or your consolidator hold | Depends on the aggregator's model | An order with the airline; still settled through the usual agency channels in most markets |
| Cost of a new carrier | Close to zero | Low, if the aggregator already has it | A new project |
You can disagree with a row — the servicing row is the one people argue about — and you should, for your own carriers. The table is a starting position, not a verdict.
What NDC actually changed
Before NDC, an airline filed fares and the GDS built the itinerary price from them. The airline did not know who was asking or what else they might buy. NDC inverts that. You send a shopping request to the airline (directly or through someone), the airline returns offers it constructed for that request, and when the customer buys, you accept an offer and the airline creates an order.
For an integrator, three things follow. Offers carry their own expiry, so a price is a short-lived object rather than a lookup you can cache. Ancillaries — seats, bags, meals, flexibility — arrive inside the offer instead of being bolted on after the PNR exists. And the airline can price differently for different requesters, which is the commercial point of the whole exercise.
What it did not change: you still need someone with ticketing authority, you still need to reconcile payments, and the standard is a standard with dialects. Two airlines on the same schema version can disagree about which fields are mandatory, how a child passenger is expressed, or whether a seat map is a separate call. Plan for per-carrier adapters even when the documentation says they share a spec.
The content differences that come back as support tickets
Nobody files a ticket saying "your distribution strategy is mixed". They file these:
- "The airline's website is cheaper." Often true. Some carriers keep their best fares or bundles off the GDS and on NDC, and some add a charge to bookings that come through the older channel.
- "My bag disappeared." The customer compared a GDS fare and an NDC bundle with the same name on screen and different contents. If your result card does not show baggage per fare, you will hear about it.
- "I saw two prices for the same flight." You are receiving the same flight from two routes and showing both. Deduplication across routes needs a rule for which one wins, not just a key to match on.
- "The price changed when I clicked book." An NDC offer expired between search and booking, or a GDS fare was re-priced. Either way the customer sees one thing.
- "I can't pick a seat." The seat map exists in one channel and not the other, or it is a paid ancillary that only the NDC offer exposes.
Each of these is a data problem dressed as a customer problem. The fix is usually in how you normalise fares into your own model, not in which route you picked.
After booking is where NDC still hurts
Shopping and booking get all the attention in NDC launches because that is where the airline makes its case: richer content, better offers, more conversion. Servicing is where integrators spend their next year.
With classic GDS content, the post-booking toolset is old and uniform. A schedule change arrives in a queue. An agent can reissue, exchange or refund a ticket with commands every ticketing desk has known for years, and your back office can reconcile it through the usual settlement reports. Ugly, but predictable.
With NDC, every one of those operations is something the airline has to implement in its API, and many implemented the sale first. In practice that means:
Voluntary changes may be supported for some fare families and not others, so the "change flight" button in your customer area has to know, per order, whether the airline will accept the request — and fall back to a human when it will not.
Involuntary changes — the airline moved your customer's flight — may arrive as a notification you have to subscribe to, as a change you only discover when you next retrieve the order, or not at all. If your platform's idea of a booking is a row that is written once and never re-read, it will drift from the airline's version of the truth.
Refunds can be a single call, a multi-step quote-then-confirm, or an email to the airline's agency desk. Your finance team will find out which before your developers do.
Mixed itineraries make all of this worse. A trip with one carrier on NDC and another on GDS content is two orders with two servicing models, sold to the customer as one booking.
None of this is a reason to avoid NDC. It is a reason to design your booking record around "retrieve and reconcile" from day one, and to scope servicing per carrier before you switch that carrier's traffic over.
What getting the route wrong costs
The expensive mistakes here are not technical. Pick GDS-only for a market whose main carriers have moved their best content to NDC, and your prices lose to the airline's own site on the routes your customers search most — a margin problem you will misread as a marketing one. Pick direct NDC for a carrier whose servicing API is thin, and your operations team absorbs every change and refund by hand, at the moment customers are least patient. Pick an aggregator without checking what it flattens, and you pay for NDC content you cannot actually show.
And switching later is not free. Your booking records, your deduplication rules and your servicing screens all quietly encode the route you started with.
Why most platforms run a mix
Because the choice is per carrier and per market, not per platform. A sensible default for most flight sellers is broad GDS or aggregator coverage, NDC where a carrier's content clearly differs or a surcharge makes the older channel uncompetitive, and direct connections only where the volume justifies an adapter you will maintain yourself.
Before you add a route or a carrier, check:
- Which fares and bundles does this route show that the others do not, on the routes your customers actually search?
- Can you change, cancel and refund through it, for the fare families you will sell — tested, not read in documentation?
- How do schedule changes reach you, and what happens in your system when they do?
- When the same flight arrives from two routes, which one wins, and why?
- Who holds the ticketing authority, and who absorbs a debit memo if something goes wrong?
If you are comparing providers rather than routes, the evaluation framework for travel APIs covers how to test error paths instead of happy ones. And if you would rather integrate one flight API than maintain three routes yourself, the flight and hotel APIs start with a conversation rather than a self-serve key — which is the right place to ask which of these routes sit behind them, and how servicing works for the carriers you care about.