Ariel Cohen: How Navan Leveled Up with a Pivot to AI

0:05Hey, everyone, and welcome to Generative Now. This is the show where we talk to the builders who are creating the world's most exciting AI products and companies. I am Michael Magnano. I'm a partner at Lightspeed. And for today's episode, I spoke with Ariel Cohen, co-founder and CEO of Navon, the all-in-one travel, corporate card, and expense management solution. Now, in this conversation, I was really interested to talk to the co-founder and CEO of a larger company. We've obviously talked to a lot of startups on this show, but in this interview with Ariel, I wanted to hear what it was like for a really big company to run hard at AI after ChatGPT emerged about a year ago.

0:43So in this conversation, Ariel shared with me what that shift was like and how they made it happen. Take a listen. Ariel, good to see you. Good to see you. Amazing to be here. Yeah, thanks so much for doing this. You are a busy person because Navon is just, it seems like it's on top of the world. I feel like every time I'm in an airport, I see a huge billboard. I see something talking about Navon. I feel like I'm using it more and more. I feel like I'm running into more and more people that are using it. And I want to talk to you about that. I want to talk to you about AI. But first, maybe, I think it would be great to give the listeners some context for how you got here, how you got to this point.

1:20Tell us about you and your background and the story of Nirvan up until this point. Of course, of course. So, you know, the story of Nirvan is actually pretty cool. Both me and Elon, my co-founder, we had a startup before, which we've sold. but we also worked in pretty big companies like HP, Mercury, Jive. And in all of these companies, we both managed global teams, which when you manage a global team, it means that you travel a lot. And we did that for years. And as Elon and I were thinking about what do we want to solve next, like what's our next startup, travel became really obvious and fast.

1:58And there were several reasons for that. One, it's a really, really big market. So companies are spending a lot of money on travel. It's usually 5 % to 7 % of their entire OPEX. It's also a big number. Companies are spending more than$1 trillion annually on travel. But then when you look at this as a traveler and as a manager, there is a huge gap between what you think that you should get from a technology perspective and also a service perspective. than what you're actually getting. So as a traveler, it took me, it used to take me a very, very long time to book a trip. Just this simple thing, like in consumer life, it's obvious, you click here, you click here, you click here, eventually your trip to Hawaii is ready.

2:49When it comes to business travel, there is this entire thing, you try to use a system like Comcare. And in most of the cases, it will not give you what you wanted from a travel perspective. so you'll end up emailing a travel agency or you're going to call someone and all of these things. So you are ending up spending a lot of time on booking a trip, which is a simple thing. If you want to change it, and in business travel, I know now 30 % of the trips will get changed. So a lot of trips will get changed. If you think about the amount of time that you've changed your plans in an existing trip or before or after, and also the amount of things that can happen to you if you travel a lot, it's just getting changed a lot.

3:31So if you want to change something, it's like an entire process, emailing, back and forth. That's why a lot of people are eventually ending up employing EAs that will do that for you. And then as a manager, so nothing makes sense as a traveler. As a manager, there is always that meeting that a CFO shows up and telling you, hey, your team is spending a lot of money on travel. And you, okay, but can you tell me what to do? Okay. and you don't know what to do. So you're getting this feeling that you're doing something wrong while you maybe just ended the quarter, right? And to end the quarter, you needed some people to fly around the world and close it.

4:13Or you just ended the planning of the year and so on. So on both sides, it didn't make sense. And Elon and I both being in technology for years, like now it's probably, I don't know, 25 years. we knew that the technology can solve a lot of it. We knew that. We knew that, you know, it should take no more than five minutes to book a trip. It should take a few minutes to change a trip. You know, if you take it farther to the other side of travel, 30 % of the budget is actually going to entertainment, to the meals, to hosting customers, all of the things around it. I don't know, you are in an Uber or Lyft, a cab, all of the things around it, it's 30%.

5:00So if travel is complex, then you do the expense. That doesn't even make sense because that's kind of somebody is telling me, okay, check out from the hotel, take a receipt. If you use the mini bar, now it's an entire thing. If you ate something in the restaurant, it's an entire thing. So now you are employing your best employees, the employees that cost you the most, the sales team, the execs and all of these things to describe a receipt. you know, that's the minibar that's the, it's called itemization right, that's the minibar that's the restaurant, here are taxes, describe the three levels of taxes that every hotel has and after 30 minutes you'll finish describing this receipt and you can repeat it with restaurants, with cars with all of these things, you know, people are having this pile of receipts and basically taking something that was digital and bring it back to be digital with a lot of metadata, with a lot of so that didn't make sense either So Elon and I basically said, you know, we should have this best solution that's super easy to book, super easy to change, super easy to get back from the trip and do the expense management.

6:12Just easy. Just easy. And if something happens to you, you're going to get really good service. That's what we wanted to do. And that's what we've been doing in the last eight years. It's incredible. I mean, what you just described, itemizing expenses, it's so true. I mean, I'm something like, you know, maybe similar to you, almost 20 years into my career in technology. And to this day, I'm still doing some form of itemization. It's getting better, obviously, thanks to products like Novan. But it is such a massive, massive time suck. So talk a little bit about, like, how exactly does the product enable companies to get this time back?

6:49Yeah. So what we've discovered, Ilan and I, is that there is a reason for all of this complexity around travel, less about expense, but around travel. The reason is that the actual travel infrastructure is very, very old from a technology perspective. So I still remember we started a company and we looked for the API to book a flight. So spoiler, there is no API like that. To actually book a flight, you need to have agreements with a lot of different entities obviously the airline but also some middleware GDS and other things so you need to kind of go one after the other per geography so the agreement that you have with I don't know United Airlines in the US will be different in Europe will be different in the Far East so you really need to invest a lot of time and effort on this pretty much business development or what we are today calling suppliers relationship.

7:49So driving content into the system is really, really complex. And some will not even have normal APIs. So you will need to kind of find ways to hack it and to bypass stuff. Some will have APIs, you know, if you're connected to an OTA or something. But mashing it all together into, you can search, I want to fly there. Mashing it all together, it's actually really super complex. So this is actually where machine learning, you know, I know that we're going to talk about generative AI, but this is where machine learning became very, very important to Novan. And why? Because we, from the get-go, valued the idea of bringing a lot of content from everybody.

8:32If you kind of look at a traditional travel agency, they will trim it to what you would call business content. We knew that we want to bring OTAs like Expedia, Priceline, bookings, but also the traditional. And why is that? Because the prices are better and availability is better. So we wanted to mesh it together. We also knew that we are not going to limit it just to traditional airlines. But in Europe, low-cost carriers, super important. Like EasyJet is important. Reiner is important. But to connect to them is different. So let's say that we've managed to connect to them. But how do you mash it together so it will not take you forever to select a flight or to select a room in the hotel because I'm bringing all of this content in?

9:20And this was about getting to know you. After you do one or two bookings in the system and assuming that the rest of the employees in your company are also booking, we will know in more than 90 % accuracy what you're going to book. From the time that you get into the system, you said, you know, I want to be in London, we will know what you're going to book. And it's important because then it allows us to put the search results from a search perspective really, really up there. No matter how many times you're going to play with the filters and the sorting, it's not important. Actually, you'll book what we kind of thought that you're going to book.

9:56So in average, it takes five minutes to book a trip on a van. And some of the trips are complex. These are business trips, you know, multi-city, a lot of different places. so that was really really important in terms of the experience there was another thing that we realized early why people still call an agent because the agent knows them which allows personalization so if you think about agents agents will tell you you're telling me that you want to go to london i'm gonna tell you stay in this hotel i know your office address i know your preferences so in london there is one hotel and it's great because it's very personalized but it's very, very much limiting you.

10:37It's like always stay in the Marriott. I, for example, I really like the experience. I like boutique hotels. I like the soul house. I like these kind of hotels. I actually like to experiment with different hotels when I'm staying in cities. So I don't want my EA to select my hotels or the agent to select. I want to see the vast kind of thing, but I want a system that will make some sense to me. So this is, again, when machine learning was super important and a big, big part of. So I want to summarize the magic was connect to everything and full machine learning makes some sense into it. Yeah, that makes sense.

11:06I mean, you're entering a very fragmented market. So you're basically providing a personalization layer on top of lots of fragmentation to give back time and give people really the experience that they want without having to invest a lot of time in it. That makes a ton of sense. And what about for the expenses? So, you know, we always knew from the, if you guys have our first pitch deck, so you can see it. We said that we want to do T &E, travel and expense, from day one. And that was the pitch. The pitch was actually T &E. We very quickly realized that it's just too complex, that travel is really complex on the infrastructure level and expense management is really complex.

11:47And we also have the rewards concept. We are basically giving rewards for employees to save company money when they choose stuff under policy, which that has its sheer complexity of its own. So it's almost like you're talking about like three different problems to solve or products. And we basically decided to trim it, to only focus on travel. And even then, we really went MVP. We really, really trimmed it. We started the company in May 2015. In November 2015, we had a product out there in the market, kind of alpha beta stage, but we had a product in the market. We actually had customers. We started to make money, actually, nine months after.

12:31And to do that, we needed to trim it and trim it and trim it and trim it and really have a really good MVP. So it took us several years to actually really go to expense management. One of the reasons that it also took us time, I looked a lot on Expensify. and I started to call it a yellow Conquer because I couldn't find a difference. Like, you know, it's kind of easier to sign up, so I get the PLG thing. But at the end of the day, you'll find yourself doing the same thing. You're itemizing, you're uploading receipts. The mobile app is not that better than Conquer. So I was probably better brand play, probably better PLG play.

13:15So that's why newer companies will use Expensify and not Conquer. But to be honest, as a product guy, I couldn't find a difference. And I was saying, so what? The world needs another. Like we will be a blue or maybe red and then our colors or that. We will build another Conquer and we'll call it Nivan. I didn't see the innovation. I thought that there is nothing to do there. So I was kind of reluctant to jump into expense management. And instead, we integrated to Expensify, to Conquer, to Coupa, to others. because I couldn't think about an innovative kind of thing there. And then, like everything in life, you know, something happened.

13:55We were blocked by some airline. And in that process, we went deep into payments, really, really deep. And we realized what Marketa is doing and what Stripe is doing and other companies. And we had this Sahar moment that through the credit card, you know a lot, you know, of what's going on when you're swiping your card. Like, I remember it was surprising to me to see that, you know, the seat on the airplane. You know, you bought something in the United Airlines website and you see, you can see through the credit card, you know, on the transaction. Yeah, you can see on the data where the person was sitting.

14:37Or you can see that somebody added another espresso shot in a Starbucks. and I was, oh, wow, that's really interesting because if you know that, then expense management could be different. So it was actually accidental that we discovered it through a problem that we had and something else and then we said, okay, so if I'm controlling the payment mechanism, basically I'm issuing the credit card, I can do expense management and I can do expense management not by creating another Expensify but I can basically tell you, swipe the card and I know everything and I don't need everything from you. I'm going to give you a policy feedback online.

15:17This was out of policy, in policy. Online, you'll know that, which is very different than expense management. And the expense was done right now online. There is no after. There is nothing to give me after. When we realized that, we started to go deep into payments and expense. And interestingly enough, you know, COVID happened and we kind of lost our entire revenue overnight. of travel. So we kind of needed to accelerate something because we still had a lot of employees and we needed to do something with ourselves. And so we really, really accelerated our expense management and payments. So if you think about our business, we are kind of around half a billion run rate.

15:59Kind of 25 % of this is expense management, which is pretty big considering that we started it only two and a half years ago. What's cool about the expense side which I just gathered from this, is that your point about there being information, there being data in the credit card transaction, I imagine there's a bit of a flywheel that happens because you probably learn a bunch of things from the expenses that then feed back into the personalization of the travel services. Is that accurate? 100%. We today, you know, companies that are managing travel has leakage. This is an employee that decides to book outside and creates a lot of problems for the company.

16:34You know, we can talk a lot about that, But it's the reality. Are you really going to fire your head of sales because he booked outside? No way. Or your top employees and so on. Or your CEO. A lot of this would be the CEO, the EA books outside. So companies have this goal to have full adoption because of safety, control, policy, a lot of different things. Because of the control of the credit card, we can report to the company adoption. Basically, companies that are using us getting into a dashboard in the morning, and they see how much money they save for the rewards program. They see, obviously, how much they spend.

17:15They see NPS. So they see the NPS of their employees. Like, are their employees happy or not? They see CSAT. So, basically, are they getting good service from us? They see adoption. How do they see adoption? Because we're basically crossing it with the credit card information, and we are telling them, hey, all of these transactions were booked outside. That's awesome. So let's jump into AI. Navon is a large company. I mean, you just mentioned you're doing roughly a half a billion dollar run rate. So this is a big company. This isn't a tiny little startup. But from the outside looking in, from my perspective, it seems like you guys have jumped right into AI.

17:55Almost, I don't necessarily want to call it a pivot unless maybe you do. But it seems like you've really gone deep in AI and you've done it very, very quickly. You know, I was not aware at all of generative AI, you know, and we had a pretty big AI team here. And I was completely blind to it. And one day I see Chad GPT. And I remember I did the morning after, I think they came out on Wednesday. On Thursday, we had an emergency meeting here because I saw it. And I was at the beginning, I was scared. I was, oh, my God, we are fucked. like nobody will need what we do. So that was my immediate reaction.

18:37And the second reaction, wow, the opportunity is amazing. Like we can really take this personalization that I was talking about, but also service to a completely different level if we'll embrace it. So it was basically within 24 hours, you know, we had a meeting with me, with Ilan, with Nina who is running a travel here with the relevant people. And we said, we need to get very serious about it. We need to learn what it is. We need to hire the relevant people. We need to get very, very serious about it. Let's first map it really quickly. Let's learn as fast as we can, all of us, and figure out what does it mean for us.

19:19And it was almost like, for me, it was an emergency. It was, I really thought that we can lose it all or actually become the winner. It's interesting that you said pivot. We did talk about it in pivot terms. We did. I was actually saying that up until now, I never thought of pivoting the company. Since the beginning, we're basically running the plan that we've communicated to you on our seed round. We were basically running the plan. And that was the first time that I was actually using this term. And I was saying, this is a pivot. We will need to think about the entire thing differently. And I'll give you one example.

19:57We have this chatbot that covers a pretty big part of our support calls, which that thing really helps us to have the economics of a tech company in a business that is basically service, you know, if you think about it. So it comes down to gross margins and so on. But it also allows us to do another thing. Now it's really happening. It allows us to commoditize. we are now, you know, for the first time in this market ever, one of our versions is free. So you can become an Avant customer by signing up and not paying us a dime. And a lot of this is from the efficiency of this chatbot. But this chatbot on its first tier was based on Watson, right?

20:40Because it's before, you know, before you have different models out there. And the switch to, at the beginning, OpenAI, Now it's actually two models. It's an Anthropic and OpenAI, but at the beginning, the switch to OpenAI was massive in terms of what we saw of the satisfaction. We measure C-SAT for the bot as well, and we compare it to the agents. The satisfaction that you are getting out of this interaction with the bot and the understanding of what the hell did you tell me, right? So what do you want? Can I handle it or not? And if I can, should I pass it to an agent? And this thing, I actually think that we are really at the beginning there.

21:26We completely pivoted the bot to do that. So we actually can get really good service from the bot. And it also saves a lot of money for us. And therefore to our customers, right? That's why they are now getting a free version of us. But then, and this is more a beta, but eventually I think it will become the main thing. Then we took it farther and said, we are doing better on hotels. Eventually it will be for everything. Is it a new UI? Like, is it really a new UI? Do you really need to click on stuff to book stuff? Does it even beneficial for you? So today, and I don't know if you guys are part of the beta or not, but today, when I kind of go and book something, I'm not booking it through our UI.

22:10I'm going to our bot, and I'm basically telling the bot, hey, I'm flying to London next week. Where should I stay? And the bot will tell me you have, based on all of my past, right? So we basically took all of the past and we match it with basically all of the knowledge that we have. We recommend that you'll stay in one of these three hotels and it will tell me why. It will tell me that in previous chats or selections, I really optimized for the gym or the restaurant or the location or whatever. Now, here is the bridge. We thought that maybe it's too much to ask people to just chat in order to book something.

22:50So we've put the chat here and the UI here and they're interacting. So once it tells you, hey, here are the three hotels, it opens in the search, the three hotels, and even circulating things that we thought we should highlight for you. And when you book, you can either tell the bot book for me or you can actually go to the web UI and click. and so we kind of created by the way I don't know if that will be eventually in three or three years the UX that we will have we are experimenting there that's why it's also a beta but there is something there again of like what would be the future UI right is it just chat is it chat with some objects I actually don't know but this is massive for us this is a pivot to your point this is massive for us in terms of how people will interact with us yeah it's really interesting to imagine the future of these ai native products right i mean per your point who knows if this will be the ui of the future i mean we were all we were all typing into terminals you know 30 years ago and the ui has obviously evolved to be about buttons and tapping on things because it's easier for our brains to understand but now it does really feel like there's a new wave of interfaces and applications that are chat first right it's a little bit counterintuitive.

24:07You know, in ChatGrid, but I would say it's even Generative AI for us because there is one system that I, one product that I'm using that I think that they are doing really, really good job on combining Generative AI. It's actually superhuman, you know, the mail app. Yep. You know, they have this button that they added. And, you know, before that, I would sometimes take the email that I want to send and bring it to ChatGPT and ask, can you do it like more engaging or more better English or whatever, right, you want to do. but it will kind of write it in a different way than my language. And I think that superhuman are using a different model and also using my kind of what I was also interacting there.

24:51And just by doing an AI there and rewrite it, it made my emails significantly more effective. And this is really, it's not a chat. It's not, it's like just making me better writer. So I think we'll see a lot of embedded kind of things like this is the experience. You've got one level of value without AI or without generative AI. And now we are getting something else, kind of what we saw with machine learning, that the search is better, that Netflix knows what I want to watch better and all of these things. So I think it's kind of the same thing. It will be, I think, more embedded. Yeah, I think I agree with you.

25:29So you talk to the team and you say, hey, this is an emergency. And again, you're not a small opportunity, actually both. Right. An emergency and an opportunity. And again, you're not a small team. Like, how does the team react to that? And how do you turn that conversation into action and into building and into shipping? And how long does that transition take? You know, we're not a small team, but we are. I think one of the things that I always thought about Nirvana is kind of Amazon. like again you can find it in our first pitch deck we said Amazon for travel. Why? Because in travel you have endless amount of needs.

26:06It's not just book hotel flight whatever you can have VIP services you can have meeting and events you can have local service that you can have endless amount of needs. Payments we've talked about it other things. So can we keep adding more businesses can we innovate innovate innovate innovate and it was always part of the culture. The problem to your point we have we have 300 3 000 employees worldwide and it's very global so how do you stay nimble with 3 000 employees you can't right how do you even steal this ship like you want to do one change and you need to have five whole hands right and you know and explain to everybody and all of these things so our solution for this is first of all culturally to act fast to have this sometimes brute force of like we're gonna act fast and it doesn't work for everybody, by the way, there will be people that you'll use brute force to steer the shift and they will not stay on the ship.

26:58But for the type of employees that are successful here, it works really well. These are employees that are actually getting excited by, hey, and now we are coming with something new. We are excited about it. And so that's one kind of mechanism. It's just part of the culture. The other thing, we went more and more in the last couple of years, kind of in post-COVID for us, to smaller and smaller team. Basically, we came with these CEOs models. So there is a CEO for Europe. There is a CEO for our expense and payments business. There is a CEO for PLG and their growth. and by that we are really trimming the team to be small teams really trying to have the teams less than 150 people and also trying to have their own culture so each of these teams have their own culture and we keep doing it by the recently we really kind of split the travel team which became really big to another kind of fruits and logic to another thing I think that allows us to move fast because it's not just with generative AI.

28:06You want to do stuff. You want to innovate. You want to introduce new stuff to the market. You want to solve problems. By the way, sometimes you have problems and you want to solve problems. And to steer a 3 ,000 people organization, remember that I've talked with a team and I described a bank to them, like a really big financial institution and some process that we had with them. And this bank, I think, is doing like$80 billion revenue a year. So basically 160x of what we are doing. And their level of complexity when you deal with them, some is regulation, some is just the sheer complexity of the organization, is really, really high.

28:47And I was telling the team, you know, when we'll do$100 billion of revenue, I don't want to be 200x more complex, right? We'll not be able to move. So how are we going to solve it? How are we going to solve it? And for me, it's like keep splitting the team, keep trimming it. So we'll have a better ability to move fast. When you think of AI, you know, everyone's talking about it as a platform shift. You know, the tech industry has certainly been waiting for a platform shift. So everyone talks about it in those terms. Is that how you think about it? Or do you think about it more as a resource akin to cash or people?