Sabre's 95%: Airline Offers Cached From Both Ends

Eric Marketts

Sabre says its cached airline offers are up to 95 percent accurate. It publishes that as a win. Read it the other way. As many as one answer in twenty may not be what the airline would have said. And that's the product working as designed.

Steph Nell

Welcome back to The V1 Airline Retailing Report. I'm Steph Nell.

Eric Marketts

And I'm Eric Marketts. Last week we ended on one line. The number is right, and the value is somewhere else. This week, three more numbers that are right and point the wrong way. Two narratives and a scoreboard check. First, the airline's live personalized offer is being cached from both ends of the distribution chain at once, and Sabre has already published the number that proves it. Second, new Phocuswright research says the AI front door for travel is not consolidating into one chokepoint. It's splitting three ways, and trust hasn't caught up with use. And then a quick check on the industry's own scoreboard for NDC, the standard airlines use to sell their own offers through agencies, which got better this year and still counts checkboxes, not orders.

Steph Nell

The through-line is this. At every layer, the metric the industry watches is accurate and beside the point. And the number that actually decides the outcome is the one nobody publishes.

Eric Marketts

Let's get into it.

Steph Nell

Narrative one. And Eric, this follows on from last week, so set it up for anyone who missed that episode.

Eric Marketts

Start with one word, because the whole story turns on it. Cache. A cache is a saved answer. Instead of asking the same question again, you keep the last answer and reuse it. Think of a printed price list taped to a shop counter, versus phoning the supplier every time a customer asks for today's price. The price list is fast and cheap. It's also only as current as the last time somebody printed it.

Steph Nell

And why does that matter for airlines?

Eric Marketts

Because the whole promise of modern airline retailing is the opposite of a price list. The idea is that an airline builds its offer at the moment you ask. The price, the bag, the seat, the bundle, possibly tailored to who you are. That's what we mean by a live personalized offer. Last week we covered one side of the squeeze on it. Anthropic, a company that makes AI models, cut the price its AI agents pay to reuse a saved answer by 75 percent, and left the price of asking a fresh question where it was. That made asking the airline for a fresh answer four times more expensive, relative to reusing an old one. That's the buyer's side of the chain choosing the saved answer. This week is the other side. The systems that sit between the airline and the people selling its tickets have been doing the same thing, and one of them put a number on it back in February.

Steph Nell

Who, and what's the number?

Eric Marketts

Sabre. One of the three big GDSs, the companies whose systems travel agents and booking sites use to find and sell flights. In February Sabre announced a product called Cache-powered Intelligent Shopping, inside its Sabre Mosaic platform. In plain terms, Sabre predicts what the airline is likely to say and keeps those answers ready, instead of asking the airline every time. And it published the results as a selling point. Up to 28 percent reduction in look-to-book, the ratio of shopping requests to actual bookings. Answers in under half a second. And up to 95 percent accuracy versus live polling, meaning versus actually asking the airline. One of Sabre's customers, the CEO of the travel search company Wego, describes what comes out as pre-generated offers served from a single data cube. One big pre-built table of answers.

Steph Nell

So read me the 95 back the other way.

Eric Marketts

Ninety-five percent accurate sounds great. Flip it. As many as one answer in twenty may not match what the airline would actually have said at that moment. It could be a different price, or a seat that's already gone. And that isn't a bug. That's the product working as specified. One more detail. Sabre's headline says 95 percent alignment. The body of its release says up to 95 percent accuracy. "Up to" means best case. So one in twenty is the best case, not the average. And the release never says what the mismatch is, what causes it, or who absorbs the cost when the saved answer and the airline's real answer disagree.

Steph Nell

And Amadeus, the biggest of the three?

Eric Marketts

Amadeus does the same thing without using the word. It has something called the NDC Offer Repository, which hosts airline fares, in its words, outside the core systems. Saved answers instead of live questions. And Amadeus tells you why it built it. Forty-eight thousand requests to one airline, for a single departure point. Up to 400,000 shopping requests per airline. Automated shopping, from booking sites and more and more from AI agents, creates enormous volume, and that volume can overwhelm an airline's pricing system. So Amadeus stops asking live. It just never calls it a cache, and like Sabre, it says nothing about how fresh the saved answers are, how accurate, or how personal.

Steph Nell

So the bull case, because saving answers isn't a sin.

Eric Marketts

It isn't. And here's the distinction that matters. You can save an answer for a group. Travelers booking from London. Business travelers. Gold-status members. Customers buying one type of fare. In our read, a lot of what gets sold as dynamic pricing is really pricing by group, not by individual, and serving a group answer from a saved copy in under half a second is good engineering. When AI agents are firing hundreds of thousands of shopping requests, asking the airline live every single time doesn't hold up. Saving the answers that rarely change, and asking live for the ones that do, is sensible triage. Ninety-five percent right and instant beats a hundred percent right and crashed.

Steph Nell

And the bear case.

Eric Marketts

You can't save an answer for one person in one moment. What makes an offer personal is the request itself. Who you are, your loyalty status, what you've bought before, what you're searching for right now. You don't have any of that until the request arrives, so you can't have saved the answer in advance. The usual defense is, fine, use the saved answer while people browse, then personalize when they're ready to buy. But that misses where the damage happens. The saved answer decides which airlines the traveler sees in the first place. If the shortlist is built from saved, one-size-fits-all answers, personalizing at checkout is too late. You're tailoring an offer for someone who already chose from a list the cache picked.

Steph Nell

So the critical take.

Eric Marketts

Go back to where this started. In 2012 the airline industry adopted IATA Resolution 787, the resolution that launched NDC, the modern standard for how airlines sell. It says the point is to let an airline give, quote, an intelligent response based on who is asking. The two biggest players in distribution have now built systems whose whole job is to answer before they know which individual is asking. That's not a betrayal. It's a confession. At the volumes AI shopping creates, the personal offer NDC was built for is too expensive to build live every time. So the middle of the chain saves answers, the AI agents at the buyer's end reuse answers, and the airline's in-the-moment offer gets squeezed from both sides. Each side made a perfectly reasonable call. And here's the part that should land. Only one of those decisions was ever the airline's to make. An airline can shape how its own answers get saved. It has no say at all over whether an AI agent reuses an old answer. We've been watching since Episode 8 for any airline to publish rules for how AI agents may shop it. In Episode 17 we found one partial answer. Lufthansa Group's terms cap how many searches a seller can run for each booking, at 500 to 1. But rules on saving answers, how fresh a quoted price has to be, what it costs to ask? As far as we can find, no airline has published them. So right now, how fresh an airline's offer has to be when an AI agent quotes it is being decided by everyone except the airline.

Steph Nell

Narrative two moves from the offer to the front door. And Eric, I want to slow this one down, because "front door" is doing a lot of work. What do we actually mean by it?

Eric Marketts

The front door is wherever a trip starts. The first place a traveler goes to work out where to fly, when, and on which airline. For decades in business travel, that was the travel agent's booking screen. That screen was run by three companies, the GDSs, the global distribution systems. Amadeus, Sabre, and Travelport. If an airline wasn't on that screen, agents couldn't sell it. That's what we mean by a chokepoint. One door everyone has to walk through, so whoever owns it can charge a toll. And they do. Airlines pay the GDSs a fee on every booking.

Steph Nell

And for everyday travelers?

Eric Marketts

For most people it became Google, or an online travel agency like Expedia or Booking.com. Same pattern, different owner. Now the industry's assumption is that AI becomes the next front door. You tell an AI assistant, get me to Chicago on Thursday for under four hundred dollars, and it does the shopping for you. The fear is that one assistant wins. One ChatGPT, or one Google, that every airline has to reach or disappear behind. That fear sits under most of the arguments we've covered this year. Phocuswright's research on European travelers, released September 4, says that single door isn't forming. At least not yet.

Steph Nell

So where are people actually starting?

Eric Marketts

Three places, and people use more than one. First, the standalone AI assistants. ChatGPT, Claude, Gemini. You open the app and ask. Roughly half the travelers using AI for trip research use one of those, 44 to 51 percent across the UK, France, and Germany. Second, AI built into travel sites. The chat assistant inside an online travel agency, or inside a price-comparison site, what the industry calls metasearch. That's somewhere between a fifth and a third, 21 to 33 percent. Third, the AI answer that now shows up at the top of an ordinary search engine result. About a quarter to a third, 25 to 34 percent. Those add up to more than a hundred because people use more than one. The point is simple. Three doors, and none of them is winning.

Steph Nell

And the trust part?

Eric Marketts

That's the gap. Using AI for trip research has gone mainstream. Trusting it hasn't. Only about one traveler in five, 17 to 20 percent, says they really trust what AI tells them. Compare that with roughly half who trust friends and family, 42 to 55 percent, and about a third who trust a travel agent, 33 to 42. People use AI to do the research. They still check with a human before they buy.

Steph Nell

So the bull case, and this one cuts against the industry's fear.

Eric Marketts

For airlines, three doors is good news. One dominant AI door would be a new chokepoint, with the power to charge airlines a toll the way the GDS screen did. Three competing doors means none of them can dictate terms. An airline that shows up clearly in all three gets found, whichever app the traveler opens. And the trust gap helps too. If people still check with a human before they buy, the airline's own brand and the travel agent still matter at the moment someone says yes. The AI does the research. It hasn't earned the purchase yet.

Steph Nell

And the bear case.

Eric Marketts

Three doors also means three times the work. Three sets of technology to connect to. Three sets of rules. Three different ways of deciding which flights to show first. The airline can't pick one and focus. And high use with low trust doesn't usually last. It's often the stage before one player pulls ahead. Whichever door earns trust first could take the market quickly, and it may be one the airline doesn't control, or can't even see inside.

Steph Nell

So the critical take.

Eric Marketts

Here's what I think the industry is missing. A lot of energy right now goes into what we've called the protocol war. That's the fight over which technical standard AI assistants will use to connect to airlines and pull their fares. Everyone is picking a side, on the assumption that one AI buyer wins and you'd better be plugged into it. This data says that's wrong, or at least early. If discovery stays split three ways, the advantage isn't being plugged into the winning assistant. It's being easy for every assistant to understand.

Steph Nell

Easy to understand how?

Eric Marketts

Think of a restaurant menu. If the menu is clear, priced, and written the same way everywhere, any waiter can take the order. If it's a mess, it doesn't matter which delivery app you signed up with. For an airline, the clear menu is its fares, bag rules, and seat options, published in a clean, structured form that a machine can read without guessing. We call that - OFFER LEGIBILITY. Almost nobody measures it. Everyone measures which assistant is winning. But legibility is the thing an airline can actually build, and it pays off whichever door wins. Own the clarity of the offer, not the pipe it travels down.

Steph Nell

Before we close, a quick scoreboard check, because the same pattern shows up in the industry's own numbers.

Eric Marketts

Every year a travel technology consultancy called AltexSoft publishes a datasheet of what NDC airlines can actually do. Quick reminder. NDC is the standard that lets an airline sell its own offers, built its own way, through agencies and booking sites. The 2026 edition came out at the end of July. It covers eighty-two airlines, up from seventy-five last year, and scores each one against seventy-three capabilities that IATA, the airline industry association, defines. Picture a big grid. One row per airline, one column per capability. Can it show fares? Take a booking? Sell a bag or a seat? Take payment? Change a booking after it's made? Every box is a yes or a blank. Nearly six thousand boxes.

Steph Nell

And the news is where the yeses are growing.

Eric Marketts

In payments, and in what's called order servicing, meaning everything that happens after the sale. Changes, refunds, cancellations. That matters, because selling the ticket was never the hard part. Handling it afterward is. So it's real progress, and a better thing to count than who can display a fare. Two caveats, and both come from the source itself. It says not every NDC airline verifies its capabilities with IATA, and some profiles aren't kept up to date. And it doesn't put a number on the growth.

Steph Nell

And the bigger problem?

Eric Marketts

A yes in the payments box doesn't mean the money flows the new way. Here's the background. For decades, an airline booking has rested on two old records. The PNR, which is the booking file, and the ticket, which is the proof you paid. Modern airline retailing is supposed to replace both with one record, called an order. A settled order means the money has moved and the sale is closed out on that new record. We showed the gap in Episode 18, when we looked at Finnair's and Riyadh Air's launches. Finnair's first launch on Amadeus's new order platform still settles through BSP, the IATA system agencies have always used to pay airlines, and it still asks sellers for a ticket number. Riyadh Air, an airline built on orders from day one, still creates a PNR when a partner airline is involved. The yes in the box is real. The old ticket is still underneath it. The number that would prove modern retailing works is how many orders clear with no ticket underneath, airline by airline. As far as we can find, nobody publishes it.

Steph Nell

Two narratives and a scoreboard. One argument. Close it out, Eric, and I'll take the last word.

Eric Marketts

The caching pincer sets the pattern, at the layer where the money is. Sabre publishes up to 95 percent accuracy as a win, and its release never says who eats the other five. Amadeus publishes the query volume, and its material never states how fresh the stored answers are. No carrier we can find publishes how fresh and how personalized its offer actually is by the time an agent quotes it. That's the number that now decides the outcome.

Steph Nell

The discovery data says the same thing one layer up. The industry watches discovery share, which agent is winning the front door, and the front door split three ways with no winner and trust that never caught up. The number that actually decides who gets booked isn't discovery share. It's whether your offer is legible, easy for all three of those doors to read at once. One is measured obsessively. The other isn't measured at all.

Eric Marketts

And the scoreboard closes the loop. Eighty-two airlines, seventy-three capabilities, growth right where it's needed. All accurate. It counts what an airline says it can do, not orders that settle without a ticket. The cached offer, the front door, the capability grid. Right number. Wrong thing. And in all three, the number that decides the outcome is the one nobody publishes.

Steph Nell

If this episode was useful, share it with someone making decisions in distribution, retailing, or travel technology. And if your organization has an answer to any of these, published terms for how agents may shop you, a real measure of offer legibility, or a count of orders that settle without a ticket, we want to hear it. These conversations are better when more informed people are in them.

Eric Marketts

We'll be back next Monday. I'm Eric Marketts.

Steph Nell

I'm Steph Nell. Thanks for listening.

Eric Marketts

Stay sharp out there.