Why AI's biggest disruption in travel is your org chart, not your tech stack

Most travel companies think the hard part of AI adoption is scaling the pilot. Getting the model into production, wiring it to the GDS, making it reliable enough to put in front of a customer. That part is real, and it’s hard. But it’s not the part that decides who wins.

The hard part comes after the pilot works. It’s the decision nobody wants to put on a slide: restructuring the organization around what AI actually does.

Travel was built on handoffs

I spent years inside travel’s distribution machine. Tourico, then Hotelbeds, leading a team of 56 engineers. From the inside you stop seeing the technology and start seeing the shape of the thing. And the shape is layers.

Travel distribution is one handoff after another. A request moves from the traveler to an agent, from the agent to a desk, from the desk to a supplier connection, from there to a fare rule somebody has to interpret, then back up the chain. Every layer exists for one reason: to translate information so the next layer can act on it. Someone aggregates. Someone routes. Someone explains the supplier exception to the person who can’t read the raw response.

That structure made sense when machines couldn’t read context. They can now.

AI doesn’t optimize those handoffs. People keep saying it does, and that’s the comfortable version. The uncomfortable version is that AI makes a lot of those handoffs pointless. When a model can read the raw GDS response, apply the fare rule, and hand the traveler a clean answer, the three roles that used to sit in between are not faster. They’re redundant.

The companies that already crossed the line

This isn’t a travel theory. Other industries already paid for it, in public.

Amazon cut roughly 13% of its managerial layer — the people whose job was to aggregate decisions and pass them up and down. Meta announced 8,000 jobs going, around 10% of the workforce, while posting record revenue near $201B. Read that twice. Not a struggling business cutting to survive. A profitable one deciding to rebuild around AI pods instead of management tiers. Amex GBT, inside our own industry, committed to $155M in synergy savings with explicit language about reducing headcount.

Then there’s Klarna, who went furthest and got burned. They halved their workforce, hit real service-quality problems, and started rehiring. It would be easy to file that as a cautionary tale and stop there. I read it differently. The idea wasn’t wrong. The cultural transition was harder than the technology — and they moved before they were ready for that part. That’s not a reason to avoid the move. It’s a map of where the landmines are.

What I actually see in travel

My conversations look different from all of this.

The travel companies I talk to know AI is real. They’re not running pilots to find out whether it works. They already believe it works. They’re running pilots to keep a hand on it. A touchpoint. A way to stay close to the thing until they’re ready to make the decision that actually costs something.

The pilot isn’t a commitment. It’s a hedge.

And that gap — between knowing and deciding — is where most of travel is sitting right now. Knowing is cheap. Half the industry knows. Deciding means looking at your own org chart and naming the boxes that exist only because, until now, one layer couldn’t talk directly to another.

There’s no single playbook, and that’s fine

The companies getting this right are not copying each other. Some build an AI-native team next to the old structure and let the old layers quietly go redundant. Some start with a plain audit: which roles exist to decide, and which exist only to translate. Some let attrition do the work, slowly, without a memo. Culture, leadership, technical maturity — all of it bends the path.

What they share isn’t a method. It’s one decision: the org chart is no longer a fixed thing.

This is the part I keep coming back to while building Bitravel. We didn’t take an existing travel agency and bolt an AI layer on top. We built the agent — Alex — as the structure, not as an assistant to the structure. No translation tier underneath it, because there’s nothing for that tier to translate. That choice was easy to make from a blank page. It is genuinely hard to make when you already employ the people in those boxes, and I don’t pretend otherwise.

The next phase of AI in travel doesn’t need better models. The models are ready. It needs a leadership team willing to look honestly at the layers AI makes redundant, and start somewhere.

So here’s the question worth sitting with: which layers of your organization exist only because the layer above can’t yet talk directly to the layer below?

This is the kind of question we’re answering in practice while building Bitravel — an AI travel agent built as the structure, not bolted onto one. Book a 30-minute call for a live look, no deck.