"Close enough" is no longer good enough: what the line between search and booking teaches us in the AI era

At the TravelTech Show in London a few weeks ago, Andrew Jordan, chief technology officer of Travelport, gave an example that stuck with me more than many of the AI slides I've sat through in recent months: in the United States there are fourteen cities called Birmingham. In a world of form-based bookings, with origin and destination typed in by hand, this was never a problem. In a world where you ask an AI assistant to book your trip in a spoken sentence, the risk of ending up in the wrong Birmingham becomes very real.

The point raised by Jordan, alongside Harry Herbert of Anthropic and Jason Cincotta of Kismet during the panel on agentic commerce in travel, runs deeper than a simple bug to fix. The industry, Jordan argues, is moving from a deterministic world to a probabilistic one. Until recently, booking systems always responded the same way to the same input — same request, same answer, every time. With conversational AI, the same question can generate different responses, and the range of ways a traveller might express a desire — from "I want a fantastic holiday in Thailand" to "the least risky way to get a stranded employee home while staying within company travel policy" — is enormous.

The distinction Jordan proposes is the one I find most useful to bring into our own sector, that of independent Italian accommodation: there is a clear line, he says, separating everything that belongs to discovery, itinerary-building and inspiration from the moment of commitment — that is, payment. Before that line, a certain amount of ambiguity and error is acceptable; it's part of the conversation. After that line, "you can't be almost right." Either you book the right room, on the right date, at the right price, or you fail — no grey areas.

This distinction strikes me as the most honest lens through which to navigate the mix of enthusiasm and anxiety surrounding AI as applied to our work. For months I've heard hotelier colleagues worried that a chatbot might "get a booking wrong," and others convinced that plugging an AI assistant into their booking engine will solve everything. Both positions, to my mind, muddle two entirely different jobs. A conversational engine can — and indeed should — be given the freedom to be creative, to suggest, to inspire, to respond in slightly different ways depending on how it's asked: that's where a guest decides whether our property feels like them, and it's a space where a touch of imperfection does no serious harm. The management system that controls actual room availability, the live price, the payment confirmation, must remain rigorously deterministic — always, no exceptions — because there, a mistake is not a stylistic detail but a guest who arrives at the wrong property, or a room sold twice.

What I notice when I talk to technology suppliers pitching to small properties is that this distinction is often deliberately blurred, because "AI that books on its own" sells better than "AI that helps you write nicer replies." But the more responsible choice today, for a property like mine, is precisely the opposite of what tech marketing pushes us to do: let artificial intelligence work freely on the discovery side — content, responses to questions, the guest's first spark of inspiration — and lock everything concerning availability, price and confirmation inside systems that leave no room for ambiguity.

A second point from the panel that I find equally honest is Jason Cincotta of Kismet's position: the real brake on agentic AI in travel is not the technology, which now works, but traveller behaviour. Fully agentic booking is already possible on many hotel websites, Cincotta points out, but that's not how people still plan a trip. It's a healthy reminder for those of us who risk investing time and energy in features our guests, in practice, aren't yet ready to use. Before chasing the latest conversational function, it's worth asking ourselves honestly how many of our guests would actually use it today — and how many still want to speak to a person, or at least see a real photo before they pay.

The lesson I'm taking home from this London conversation, as an independent Italian property, is to resist the temptation to treat AI as a single block to be adopted or rejected wholesale. It makes sense to let it run free where the cost of an error is low and the value of a natural response is high, and it makes sense to keep it on a very short lead where an error costs a lost booking or, worse, a disappointed guest at check-in. The clear line Jordan talks about is not a technical problem to be solved; it is a responsibility that belongs to whoever runs the property — not to whichever supplier happens to be knocking at the door.

FAQ

Q: What does it mean that travel is moving from deterministic to probabilistic?

A: It means that traditional booking systems always return the same response to the same input, whereas conversational AI can generate different responses to the same question, introducing a margin of ambiguity that simply didn't exist before.

Q: Where should AI ambiguity stop in a hotel?

A: Before the moment of payment, ambiguity is acceptable — for instance during inspiration or recommendations. After that point, meaning availability, pricing and confirmation, systems must remain rigorously precise and deterministic.

Q: Is the real obstacle to agentic AI in travel the technology itself?

A: According to the experts who spoke at the TravelTech Show in London, the main brake is not the technology, which is already mature, but travellers' willingness to trust it and change their booking habits.