Winning Travel's AI Race | Ariel Cohen, Navan and Molly O'Shea, Sourcery | RAISE Summit 2026

Amazing to be next to the Mona. Like we are basically it's the first time speaking in a place like this. This is amazing.

Oh, yeah, I forgot that. But my CFO is sitting here so he didn't approve. So we thought of doing it. But you're all here. So I don't think you need the trip. But maybe the next one. Well, okay. So, we're at Raise. This is a huge conference. So many people flew in internationally to be here.

Before that, I have to ask how many of you are using Navan because I saw that a lot are using corporate travel. So, not a lot. Oh, okay. We're getting hands. Okay, cool. Thank you. That's amazing. So AI agents, right? That's what you're asking. I think, you know, we'll talk about technology in a second and what we do with AI, but I think, you know, we are sitting in a conference and everybody came together and there were so many events here in the conference, and it's this thing that we at Navan have the belief that the in-person connection, being together, no matter what you talk about, what you want to do, is so, so, so important. And actually, that's what drives us, that's what fuels us, this is how we are thinking about stuff. So when we talk about agents or our AI infrastructure and all of this, we never, never, never forget that it's actually about people, and it's about people that need to use travel to be here, but it's also the people that we will all and are all employing, and what does it mean, and what all of this AI transformation mean to everybody, and I think it's really, really important to remember that.

So from day one we started the company 11 years ago. Recently took our public. Always, always thought about how can we make it super efficient, super fast, super easy to use. So from day one it actually meant machine learning because you really, really need to understand the user. You need to understand what they want. You need to kind of organize the search correctly. How do you think about cost? Saving money. All of these things. We actually save money for companies that are using us. On average, it's 15% and it's for the entire travel budget. So the companies that will spend $100 million a year on travel, we're going to basically make it 85, and it's all machine learning. So this is kind of the early days. But then, you know, as we thought more and more on the service component, you know, stuck in an airport, we started to develop probably six years ago kind of early stages of conversational bots to do that, but they were kind of covering, I don't know, 10% of the problem with the satisfaction of 60%, right? So kind of a good experiment but definitely something that doesn't really work. When GenAI came, obviously it was early, so we couldn't just deploy it, because if you think about travel, travel is super important. I cannot send you to the wrong flight or I cannot. So hallucination was very, very serious back then. So we had to develop our own platform. So we basically developed something that we call internally Navan Cognition, which is an orchestration platform. It's basically an agentic platform, but what's so unique about it is that it's actually orchestrated between AI agents and actually travel agents, and both are super, super important. I know that everybody has this opinion like in five years nobody will be employed, maybe nobody will ever need travel agents, nobody, whatever. I actually think it's bullshit. I think travel is so, so complex, so fragmented, so personal, that the only way to make it work is really to start with an AI agent that can book your flight, change your flight, I don't know, take care of a receipt, all of the things that if you're using Navan, you're actually seeing in the platform, but then knows when somebody else needs to intervene. An example: in the 24 hours before the flight is happening, there is no way, API, AI to do anything. You can try to call with some bot to the airline, good luck. So you need to really know how to orchestrate it, start a discussion here, move it there, and create a perfect, perfect experience. So just to give a statistic here, today 60% of all of the support interaction — this means I'm stuck in the airport, I want to change my trip, I want to do complex stuff — are actually done by Eva, our chatbot. A big number of the transactions are based on our model. We are also using frontier models. But the way that it works, we brought satisfaction to almost the level of a human agent. But it's because of this orchestration that I've talked about, because of the fact that when it fails, we actually know how to move it, continue the discussion, finishing it. So again, the human element here is super important.

Actually, any type of AI problem that everybody here is probably dealing with, you cannot afford in the context of travel. So kind of if you think about everybody showing demos around AI, right? It's kind of the — and especially about travel because it's such an obvious use case: I'm going to plan my trip automatically, I'm going to do everything automatically. It is super complex, and the first complexity even before hallucination, any type of latency — it's not something that you will agree with reasoning while I'm kind of trying to change my flight. It's not a thing. It's not something that you can do. So you really need to handle that. But then I remember the early days, it would send — and we back then we used one of the frontier models and it didn't know what to do. So it would send you to some random travel agency, you know, will the travel agency, but it will tell you, okay, I'm going to send you to some random travel agency to fix the flight for you. And obviously, it's completely kind of fucked up. So this is where we decided to develop our own platform. This is where we said it's so specific, it's so — the need that we understand — to your point, you cannot hallucinate anything. You cannot hallucinate an upgrade, right? And by the way, you can't trick a model to give you an upgrade. So you cannot do that. Think about it. Think like I've upgraded to business class because you kind of tricked me to do that. It's not something that the system can do. And in the early days you could actually trick the system to do that. You cannot miss the time to the airport. You cannot — it's serious stuff. So early days we really scripted the guardrails. Then we started to evolve. We've created models that are supervising other models. In a way, it works extremely similar to the way that a call center would work. Very, very similar. Like if you think about a call center, you'll have in our case travel agents and some operations people and some supervisors and some angels or humans, right? And in a way, our system works very, very, very closely to that. So basically, I think that was the need, that's what inspired us, that's what we've developed.

So just to give you, we really accelerated our growth rate in the last few years. Just to give you an idea, the usage in the system in the last quarter, the one that we reported, grew up by 50%. So GBV grew up by 50% year-over-year. We continue to accelerate, the revenue grew up by 40%. And we became profitable and cash flow positive — all of these things that you want to see at this stage. But a big, big, big reason for that is that because of AI, because of this infrastructure that you're talking about, we could make the experience better. But the entire flow from the time that you sign up as a company, the type of ad that we're actually giving you in Instagram, right, which is highly related to what will convert you, reminding you as a user — although your company told you to use us, it doesn't mean anything — so reminding you as a user to continue using Navan, but then all the way to book your trip, to support you, everything around that. So I think we are kind of in the infrastructure era, to your point, and it's okay, but somebody will need to create some applications on top of it, and a lot of new companies will be created around this infrastructure that eventually will take the value chain maybe from the CPU layer all the way to the application layer, because people will actually need to do something with all of this.

So first of all, I think it's kind of funny. We started in 2015 and we thought that we're gonna make it mobile. And we quickly discovered that in business travel back then — and it's still a thing — people are calling an agent and emailing an agent. And I remember we got funded by a famous VC, and we had this discussion and we talked about mobile, machine learning, all of these things. And I remember that I told him, hey, we are basically trying to bring it to the browser, right? So that's how, like, you know, we are basically disrupting the phone — and I mean like the phone, like people are calling — and this is 2015. So it quickly moved when we started to introduce to the market the idea that you don't have to call an agent, that you don't have to email back and forth, that you don't need to spend 30 minutes to book a trip. Back then it took seven minutes to book an entire business trip in Navan, which is a fairly complex task. So we started to introduce that it could be fun, it could be easy. We introduced instead of a policy, rewards: every time that you book something below policy, we can reward you as an employee to later book your personal trip in Navan. And it's important because bleisure — which is this thing that I'm doing, the weekend in Paris but I'm doing the business trip in Paris — is a big, big, big, big use case for us. And it came out of this. So we really took it from something super boring to something fun and still in a corporate environment. So I think with all of the new technology, it gives us the ability to take it from a certain level of satisfaction to a really, really good satisfaction. I think that's what drives us. I personally used to do products. I stepped back to do products two years ago because of this opportunity. And this opportunity opened up to basically our recent release, which is Navan Edge. It's a completely AI end-to-end product that is basically our personal assistant on everything: planning your trip, booking your trip, knowing the right prices, figuring out your restaurants, figuring out the event that you want to do with your team at night. And we're going to bring everything through this platform. So whether it's regular travel, VIP travel — which is more of how do you support execs — and meetings and events, all of these things we're going to kind of drive through this platform. The pace of development went up, but also the capabilities went up. So for me as a product person, it's actually really exciting to get back into that kind of thing.

I never do that. I never do that. No, I just think that the best thing, always, always, always, is to think what is the customer problem, and customers are in different places in their journey. I mentioned the search algorithm, right? And we really nailed it. Like, we're going to show you the flight or the hotel that you want. You don't need to sort it out. And yesterday I had dinner here with some of our enterprise customers, and one of them brought up, 'I actually want just to sort it out based on the price. I don't need all of this fancy thing.' And it's important to understand it. Customers are in different places of this journey. There are still people that are using legacy systems, right? And you sometimes ask why, but it's a fact. So I think I'm actually never start from the technology. I never read research papers. I'm starting from what is the user need, what is our customers' needs. It's the thing that drives us in this context of bringing people here.

If they're going to go away? I think how we're going to beat — we are beating them. So we are taking massive, massive, massive share from the main competitors in the play in this space. Whether it's usually a combination of a travel agency and Concur, so we are taking massive share from them. We're going to continue to do that. But we are also — and that's unique for Navan — we are really good also at bringing companies forward that are managing travel for the first time. And actually, this AI created a lot of new companies, and a lot of them were not going to go to a travel agency and Concur to book their or manage their travel. So once they go to a certain stage, they usually come to us. A lot actually through our website, PLG. Anthropic actually came through our website, PLG, went all the way to their size, and obviously they travel a lot. So that's an example. But a lot of the AI companies in this space — I think the world is changing. I cannot imagine the legacy players keeping up. You cannot build a model if you are having live agents; it's kind of very hard to do. And if you are kind of a tech company, so that's our advantage, and we're going to continue to use the advantage to take this entire market.

I think we actually — it's funny. We kind of sometimes joke about it because our business model: we get paid only when you use us, and we get paid by our customers and also by the airlines and hotels. So you are kind of being paid for task, which is not that far from the new business model. So in a kind of nice way, we didn't have a SaaS model to begin with, so we don't have to kind of die with SaaS, which is a good way if you are a public company. But now seriously, I think our agents — I'm talking about our human agents — they are proud of what they do. They're doing something very important. They're very professional. You're talking about people that are doing it for 20, 30 years. And the combination of the two works extremely well.