AI Takes Flight
How Artificial Intelligence is Driving the Travel Industry’s Online Boom
How can you meet search request growth at a rate matched only by Moore’s Law, whilst reducing costs year on year? That’s the challenge facing the travel industry, and AI is driving the solution. Lucy Ingham speaks to Mike Croucher, chief architect of Travelport, to find out more
Since the 1980s, the cost and availability of travel has changed dramatically. Once the preserve of business customers and the wealthy, it has become available to almost everyone, with many of us spending hours browsing everything from serious future trips to dream destinations.
This has been made possible by technology, and primarily automation, with advances over the past few decades bringing changes to how the industry is run, managed and presented to customers. This ultimately began in the 1980s, with computer systems for the first time allowing not only digital seat allocation, but the connection of travel agencies to airline’s IT systems, enabling one company to book flights across a whole range of providers.
“It automated the industry, so it took cost out by automation, and I think that allowed airlines, travel to just expand,” explains Mike Croucher, chief architect of Travelport.
With these computerised systems in place, the industry was then able to build in efficiencies, cutting costs further and increasing the availability of travel in the process.
“The early part of the nineties saw the move from automated to optimisation, so the cost structures of airlines and travel was heavily automated but not optimised. Optimising routes, optimising dates at airports, optimising congestion at airports like Heathrow; so you saw optimisation hit,” he says.
“And then I think the 2000s saw the internet, and you saw distribution, and what that saw was a consumerisation of travel that said people can book themselves, and I think with that you saw low-cost carriers come in. Low-cost carriers couldn't have existed in a no-internet world, because their cost structure would have been too expensive.”
However, much has changed since those early days of large-scale internet use, and now we don’t just browse to book holidays. Now travel is a given, rather than something to be aspired to, we surf travel options to dream about future destinations, to get inspiration and even just to pass the time. And with that, has come an incredible surge in the number of travel-related search requests being made.
High growth, stable bookings: the travel industry’s challenge
While the average traveller is faced with a vast number of options when booking, or even just browsing, holidays, the IT systems driving them are the domain of a very small number of organisations.
“Your Expedias, your Pricelines, your booking.coms, all of those are only fronts, and they need a booking engine behind it, and us or one of our two competitors are the engines behind the whole of the internet of travel,” says Croucher.
This means that Travelport alone now handles more search requests than would have been made across the entire internet just two decades ago.
“We're doing eight billion, nine billion searches a month at the moment, we reckon that will double again next year,” he says, adding that this number is already far higher than a few years ago.
“When I joined Travelport three years ago we were doing around one and a half billion search requests a month. Last year we did five. So it’s that Moore's growth of nearly doubling every 18 months.”
The reason for this search request growth is, according to Croucher, the increased time many of us are spending in the online world.
“Mobile and tablet has allowed people to browse travel without a desktop, and I think that means that a large group of people are bored during the day, and spend their time browsing,” he says.
“What they browse is not us but they browse the internet, so one of our customers like Expedia, but they also browse the meta searches from today or TripAdvisor and things like that. And they actually farm out to all these people to get an answer back.
“So imagine if you go on TripAdvisor and ask, they ask ten different questions to ten different OTAs – online travel agencies – and then they give you the results of all those back. Well they all hit us.”
“Mobile and tablet has allowed people to browse travel without a desktop, and I think that means that a large group of people are bored during the day, and spend their time browsing.”
What’s more, even if the flights, hotels and other services are the same, each OTA requires a different answer based on their own business position and offerings.
“Each of those OTAs are trying to differentiate themselves in a different way, so they have some private fares, different negotiated rates. They may order flights differently depending what they're trying to push,” he explains. “So we give a different answer back for each of those ten based on their business value and then TripAdvisor puts them up.”
Each of these search requests requires infrastructure to respond to it, meaning that as the number of requests rise, the cost of providing answers to them does too. Which would be fine if more requests meant more bookings, but this isn’t the case.
“The number of bookings that are being made on our system isn't going up significantly in line with doubling the number of search requests,” explains Croucher.
“When we designed these systems we were seeing travel agents tend to do about 20 to 30 searches per booking. Today a travel agent would do about 50 searches per booking, on average. If you look at an OTA, they're doing about a thousand [searches] to book, and in some of the worst cases ten thousand to one.”
Lower costs year-on-year: the AI-driven solution
With a rise in search requests disproportionate to the number of bookings, Travelport is tasked with doing more for less every year.
“We have to actually constantly provide a unit cost of search cheaper year-on-year, because we can't let our cost base go up,” says Croucher. “So the big thing with technology: how do we provide the same type of expertise, the same quality of service at the same time as keeping our costs down?”
The solution has come in the form of artificial intelligence, supported by cloud technologies.
“We're using all-modern technologies, lots of cloud distribution, lots of artificial intelligence and all things like that to actually really predict in advance what somebody is going to ask for,” he says.
An approach that has been developed in only a few years, it has seen Travelport transform its IT infrastructure.
“Four years ago, all Travelport's IT was centered in our Atlanta data centre. We've got what's called z/TPF, it's the IBM latest mainframe technology with an operating system on it which is exceptionally fast, exceptionally secure and exceptionally reliable. We use it, banks use it as your base system of record,” he says.
“It's greatest in-out, so you get read and you get data out really quickly, but it's not good at what you call orchestration. In other words, sitting on a computer a long time trying to work out what to do next.”
While this central server is still used for key tasks such as keeping customer data secure, it has been augmented with a network of cloud servers.
“We built that out; when I came in three years ago we moved from the theory of having it in-premise to the theory that actually you ought to put it in the cloud for two reasons. One is proximity to customer: you put it in the cloud, you take down the latency. So that gives us proximity and speed,” he explains. “And the other is that type of distributed capability says we can put things where we think it's best to use it.
“But we never went to the cloud in the way that I think most people think about the cloud. You can buy a piece of server farm in the cloud: that's fine. We've built up what I think is the first hybrid cloud, which is a serverless concept that sits between your data centre and the cloud; it's all integrated. And we started building it all in microservices, which are small elements of code and data together, which means we can operate the function in our data centre in the cloud without actually worrying about where we're operating it.”
“We've seen as we put data up into the cloud, and images, we've got small databases or image sets up there and we can get 95% to 96% cache hit rate, so only about 4% of our transactions are having to come back through to actually retrieve rich content.”
Where Travelport really differentiates in its use of the cloud is in its integration of artificial intelligence, which has involved a partnership with Microsoft and the use of some of the technology driving the Xbox gaming system.
“On the internet what happens is if you're playing a game, you've got this high-definition game unit but all they're controlling is the movement within that,” he explains, adding that the system predicts where you are likely to move next in order to load in the appropriate visuals from a remote server.
“We thought: how do you apply that to travel? So Microsoft worked with us last year to introduce the content distribution capability of Microsoft into our artificial intelligence that starts, say, from a high-definition picture,” he says.
“Imagine it: around the world you’re looking at hotel pictures, cars. The airlines are all trying to brand their stuff, it's no more than a tube and they want to say they're different so they all put different pictures of their seats up, things like that.
“So those high-definition pictures and videos, we're loading into the cloud intelligently and using the Microsoft artificial intelligence to decide where to put it. So in Australia obviously you've got a lot of Australian hotels, but if somebody started to hit that for American hotels you start uploading that.
“And we've seen as we put data up into the cloud, and images, we've got small databases or image sets up there and we can get 95% to 96% cache hit rate, so only about 4% of our transactions are having to come back through to actually retrieve rich content and pictures, because we start using intelligence to do it. When you first put it up there you get 50% and then it learns very quickly which pictures are the ones people are going to view.”
Whilst being able to show an individual customer cached images rather than ones retrieved from the central server will have a minimal impact, on the type of scale Travelport handles it has a significant effect.
“We can precalculate, precompute. And if you're doing that on that type of scale, say we do nine billion search requests a month, if you think what a search request is, we may turn up three hundred different options to somebody,” he says. “So if you think of individual itineraries, in other words pricing an individual itinerary, we're doing about six billion of those a day. One trillion transactions a year on our platform.”
But with request rates set to rise further, Travelport’s AI plans are only beginning to see fruition.
“So at the moment we've got heavy content out there, rich content, next year we start putting processing out there and putting processing next to the customer,” he says.
Using artificial intelligence to optimise the travel experience
While AI is proving vital to keeping Travelport’s costs down, it is also allowing the company to help improve the customer experience, which in part is handled by working with travel agencies to introduce AI.
“If you look at the large travel agencies, or the corporate travel managers, they have large call centres of people handling those change requests on flights. As world and economics are changing, they've got to [reduce] the costs of their business,” he explains. “We are working with them on artificial intelligence, agency efficiency, robotics, chatbots and things like that.
“So for the really simple case studies in travel, taking them away from call centres and doing them automatically. So linking that type of capability, take it out of the call centres, but also to put it into the mobile device so the traveller themselves is going to have it.”
This opens the door for a greater level of personalisation to be introduced to the industry, allowing customers to receive a more tailored experience.
“The interesting difference between a traveller on a mobile device and a call centre: the call centres can handle multiple pieces of data and make intelligent decisions, the individual on a mobile device wants the answer or wants two or three choices,” he says. “So you have to be a lot more sensitive to the person, the persona, the context of it and therefore using a mixture of artificial intelligence, predictive analytics, looking at next best action, all this coming your way via a mobile device.”
“I think you're going to see now not just datasets we know about, but the interaction of social media datasets, weather datasets and other things around you.”
This technology is already finding its way under the thumbs of travellers, as Croucher himself recently experienced when he was scheduled to take a business trip from London to Denver via Atlanta.
“That morning I get up and I'm heading out to the car, and my Heathrow to Atlanta trip is delayed by five hours, which means I won't get into Atlanta in time to make my Denver flight,” he explains.
“Now the predictive analytics that Delta has, it's all connected with stuff we do as well, said hold on a minute, you're going to be late. They sent me a message to say well it's not a problem, we booked you on the first flight out of Atlanta to Denver in the morning and we've got a hotel for you,” he says.
“But hang on a minute, two minutes later they put another option up to say actually we'll fly you via Minneapolis, and actually put you in first class there and back and get you there earlier than you would have liked. Press the button: it's an acceptance on my phone, got it.
“Now one of my friends travelling with me did the same. But somebody else didn't react quickly enough and of course by then that option has gone. So what you're looking for all the time is the next best action for that person, and you can use so much predictivity on that.”
While effective, Croucher sees this approach as just the start for the industry.
“I think that's going to go even further. We don't see it yet today, but we're going to see, for instance, you're in Chicago for a business meeting, you're on a ten o'clock flight out and there's a weather storm coming in,” he explains. “You get predictive analytics that tell you today that that flight has a 90% chance of being cancelled, but the five o'clock flight is likely to fly. So you could predict that type thing, give it to someone and say: do you want to leave the meeting early and get on that flight?
“I think you're going to see now not just datasets we know about, but the interaction of social media datasets, weather datasets and other things around you mixing into that type of ecosystem to give really good customer service to that end.”