Friday, August 28, 2015
Interview with Devon Tivona of Pana, The AI Driven Travel Concierge
Can artificial intelligence software be the answer to better travel arrangements for you on your next trip? Denver-based Pana (www.pana.com) thinks so, and is developing AI software which works interfaces with human agents, to provide what it hopes is a much better--and more efficient--way to book travel for users. We spoke with co-founder and CEO Devon Tivona this week, to hear about the startup, which just raised $1.35M in a funding round.
How exactly does Pana work?
Devon Tivona: We're an on-demand travel concierge, for things like flights, hotels, and food. You can download our app, and you're introduced to a concierge, which can book you a trip to New York, find you a hotel in Paris, all through an easy to use chat interface. That creates an awesome experience, of having a superhero in your pocket wherever you go.
We understand there's a bunch of artificial intelligence software that is part of this process?
Devon Tivona: The actual Pana concierge is fifty percent artificial intelligence, and fifty percent actual, real humans. When a message comes into our system, we do a lot of analysis, and figure out exactly what our users are talking about. What are they asking for, where do they want to go, when do they want to come back. That analysis makes our real, human concierges hyper efficient. A traditional travel agent might be able to handle fifty clients at once, but with our software, we can handle 1,000 active users per concierge. To do that, we're doing a lot of natural language processing, and using machine learning and traditional artificial intelligence on the hundreds of thousands of messages coming through our platform. That help us understand what those conversations look like, and propose a particular response to those conversations, and things like that.
What is your prior background and how did you get into this?
Devon Tivona: We have three cofounders, two of them technical. Lianne and I met at CU, where we were both studying computer science. We both had worked at a bunch of startups. She came from Twitter, I came from a startup called Everlater. After that, I was building stuff in the travel and adventure travel space, as well as a blogging platform for adventure travelers to post their stories, which we sold to Mapquest in 2013, before we did this.
How much of this is technology, how much of this is people?
Devon Tivona: It's a 50-50 split between what the technology handles, and what real humans handle. That split will grow, and become more and more automated over time. One thing we talk about, is when a user reads a message from us, we constantly are trying to pass the Turing Test, so that they experience something that really feels like they are chatting with a real person. We always want to be passing that test. Siri, for example, is definitely not a real person. It doesn't understand you. That's why we'll always have people involved in this process. For a very complex request, the technology is never going to figure that out, even ten or fifteen years out. On the other hand, there are things like flight searches, where we can figure out where you are going, and if you prefer economy or first class, and that kind of stuff AI can handle. That makes our approach different from the Siris of the world, where you know you are talking to a chat bot. Instead, we want to make it feel like you're always talking to a real person.
How do you handle the booking end?
Devon Tivona: We have integrated our back end with a bunch of different booking engines, that come with negotiated rates from hotel and airline partners. Because we go through booking engines, you'll often see prices lower than you might see on Expedia or Kayak. Of course, fares are always different, and some will be higher, as it is on every search engine.
What's the most interesting result you've seen out of using your AI software so far?
Devon Tivona: I think it's been some of the sticky situations we've been able to get people out of, because we've seen things coming, and been able to solve for them. For example, our systems noticed that there was a train strike in Germany a couple of weeks back, and that allowed us to reach out to a traveler five or six hours before she needed to get to the airport, and help her find an alternative to the train, so she would not miss her flight. Without that information, and particularly since she did not speak German, she absolutely would have missed her flight.
Can anyone use the service now?
Devon Tivona: Our product is in private beta right now, for taste makers and early, early users. We'll be building things up for the next three or four months to a full, public launch, which is the next thing you'll see us doing. We'll continually be interating and making the product better, the technology better, so that when we launch there won't be a wait list. We want it to be an incredible experience.
Will software ever replace humans?
Devon Tivona: I can't imagine a computer doing all the complex stuff we need our assistants to do. However, I do know that we can make humans a hundred times more efficient than today, and that we have a long way to go.