By Emmanuel Mounier, Secretary General, Global Travel Tech.
A quick trip to the basement
Imagine walking into a shoe shop. You ask for a pair in your size, and the assistant disappears downstairs to check the stockroom. They come back, you change your mind, you would prefer them in black, and off they go again. Every question you ask sends someone down to the basement.
That is more or less what happens every time you look for a flight or a hotel online. Behind the clean interface, your query runs through a long chain of systems, checking live prices, availability, rules and conditions. Every time you filter for only direct flights or family friendly hotels the whole query starts over from scratch.
For years this worked, because of a very human limit. Not that we look a little: we circle a destination and weigh options and prices for weeks before committing. Expedia has clocked the average traveller viewing around 141 pages of content before a single booking. But a person eventually gets tired and books their ticket. The economics were quietly built around that ceiling. AI is now removing it, and the industry needs to think about what comes next.
Now fill the shop with robots
Picture the same shop, except most of the customers are now AI agents, each firing off thousands of requests a second. There are not enough assistants in the world to keep running to the basement. Hire more, and the shop goes under. Ignore the requests, and customers leave. And here is the uncomfortable part: you can no longer tell which customer is a real person checking out a handful of options for every purchase they make, from a machine able to request millions of options before buying anything at all.
This is not far-fetched. Someone has already put it to the test. To find a single fare, a technology founder recently pointed an AI agent at one airline’s website. It came back with 881,076 options: every date, every stopover, every combination. He wanted one ticket. He shared it because it was so striking.
Every one of those checks drew on bandwidth, computing power and energy somewhere down the line. He had done nothing wrong. He was deliberately showing what AI now lets any of us do: ask for everything at once and let the machine sort it out.
When looking stops leading to booking
The whole model rests on a quiet bargain. A single trip can mean weeks of looking before anyone books, and that one booking pays for it all. None of this is new. Two decades ago, airlines saw roughly ten searches for every ticket sold; today the industry’s own estimate runs into the tens of thousands to one, and it calls the trend unsustainable. AI is what turns a steep climb into a cliff.
For airlines the squeeze is sharpest, because their live fare data is richer and heavier to serve. Large metasearch engines can already field up to a hundred billion requests a day, and AI agents will only push that higher. For the platforms and distribution systems our members operate, it lands as a real and rising cost that nobody agreed to take on.
So the industry is reaching for the familiar tools: caps on searches, smarter filtering, dynamic pricing. It is even renaming the problem, proposing measures like “CPU to order” that count the computing burned to turn a search into a sale. These help. But they are patches on a model that always assumed searching was cheap. The harder question sits underneath. How do you build a distribution business that works when the searcher never gets tired?
I do not believe the answer is to slam the door on AI agents. Used well, they can be genuine co-pilots for travellers, and locking them out would only send people to whoever lets them back in. The answer has to be a new bargain, one where the cost of all that searching is visible and fairly shared.
Better for whom?
AI is sold to us as a way to make life lighter and more efficient. Yet here it is doing the opposite, spending vast amounts of energy to throw up more options than anyone needs. After hundreds of thousands of searches, the traveller still books one ticket. Just one.
We talk a great deal in this industry about sustainability, about the carbon cost of the journeys we help people take. We should be just as honest about the cost of how those journeys are searched and sold. A system that multiplies effort without multiplying outcomes is not progress. It is waste dressed up as convenience and unsustainable in the long-term.
None of this is a reason to fear the technology. It is a reason to be clear about what we want from it. The point was never searching for its own sake. It was helping someone get where they want to go. If AI can do that with less friction, wonderful. If it only does it with more noise, more cost and more energy, then we are allowed to ask a very simple question. Better for whom?