Omio’s bet on conversational travel is more than just a chatbot

Travel booking is one of those industries where everyone claims to be “reinventing the experience,” but most of the time you end up clicking through a dozen screens to find a train that leaves 15 minutes after you wanted. Omio, the multi-modal travel platform, is trying to change that with conversational AI. And unlike some of the half-baked attempts I’ve seen, they’re actually putting real engineering weight behind it.

The core idea is straightforward: instead of digging through filters and calendars, you just tell Omio what you need. “I want to go from Berlin to Munich next Tuesday afternoon, preferably by train.” The system parses that, pulls options, and handles the booking in a back-and-forth conversation. It’s powered by OpenAI’s models, and Omio has been integrating them into both customer-facing features and internal workflows.

What caught my attention is how deep this integration goes. Omio isn’t just slapping a GPT wrapper on their search bar. They’ve rebuilt parts of their backend to be AI-native, meaning the models aren’t just generating responses—they’re actually executing booking logic, checking availability, and handling edge cases like cancelled trains or sold-out seats. That’s a lot harder than it sounds. Most travel APIs are a mess of inconsistent data formats, and getting an LLM to reliably call them without hallucinating a non-existent route takes serious engineering discipline.

I’ve tested the conversational booking feature myself. It works, but it’s not flawless. On a simple query like “flights from London to Paris next Friday morning,” it handled everything smoothly. When I tried something more ambiguous—”I need to get to a meeting in Lyon by 3 PM, but I want to leave from Zurich after 10 AM”—it stumbled a bit. It suggested a train that required a 45-minute sprint between stations in Geneva, which anyone who’s been there knows is a terrible idea. The system didn’t account for transfer time properly. That’s the kind of nuance that separates a demo from a daily driver.

To Omio’s credit, they’re transparent about this. They’ve been running the feature in beta with a subset of users and iterating based on real feedback. The product team told me they’re focusing on improving context retention—so the AI remembers that you said “window seat” earlier in the conversation without you having to repeat it. That’s table stakes for any good conversational interface, but it’s surprising how many travel bots fail at it.

The internal use case is less flashy but probably more impactful. Omio has deployed OpenAI-powered tools for customer support agents, helping them quickly look up booking details, process refunds, and handle complaints. The support team’s response time has dropped significantly. One agent told me they used to spend 10 minutes navigating internal dashboards for a single refund request; now it’s under two. That’s real productivity gain, not just hype.

There’s also a developer angle. Omio’s engineering team uses GPT-based code assistants to speed up feature development. They claim a 30% reduction in time spent on boilerplate code and API integrations. I’m always skeptical of such numbers—companies tend to cherry-pick metrics—but even if it’s half that, it’s meaningful. Travel tech is notoriously complex, with dozens of suppliers, each with their own APIs, rate limits, and data quirks. Anything that reduces the friction of stitching those together is a win.

The bigger question is whether conversational travel will actually catch on. I’ve been burned before. Remember when every travel site had a chatbot in 2018? They were all terrible. But the underlying tech has improved dramatically since then. LLMs are genuinely better at understanding messy human language than the rigid intent-classification systems of the past. The risk now isn’t accuracy—it’s expectation management. If Omio’s bot handles 90% of queries perfectly but fails spectacularly on the remaining 10%, users will remember the failures more than the successes.

Omio seems aware of this. They’re not marketing it as a magic solution. The conversational interface is optional; you can still use the traditional search if you prefer. That’s the right approach. Let people opt in when they want speed, and fall back to the old way when they need precision.

I’d like to see them open up the conversational feature to more complex itineraries—multi-city trips, mixed modes of transport, overnight journeys. That’s where the real value is. Anyone can book a single flight with a chatbot. The hard part is planning a trip that involves a ferry, two trains, and a bus across three countries, all while keeping the total travel time under eight hours. If Omio can crack that, they’ll have something genuinely useful.

For now, they’re on a solid path. The engineering is thoughtful, the rollout is measured, and they’re not pretending the tech is perfect. That’s more than most AI travel startups can say.