Skip to main content

Season 1, Episode 2: A Conversation about the Future of AI in Travel: Part Two

Blue sky with white clouds image for PROS Travel Podcast

PROS is excited to continue our new podcast season, Ahead of the Curve. This month is part two of the conversation between PROS Travel President Surain Adyanthaya and PROS Chief AI Strategist Dr. Michael Wu about the exciting future of AI in the travel industry. Be sure to listen to Part 1 if you haven't already!

You can listen to Ahead of the Curve on Apple Podcasts and Spotify. Or listen to the full audio version below.

Full Transcript

Surain Adyanthaya:Welcome back to another episode of Ahead of The Curve. And joining us today is, once again, Chief AI Strategist of PROS, Michael Wu. Michael, welcome back....

Michael Wu: My pleasure.

Surain Adyanthaya: Michael, you've been very closely involved in many of the innovation initiatives at PROS, and AI is such a big part of our innovation and product strategy as a company, so it's really exciting for me to have you back, talking about some of the ideas that we've been working on. In the last episode that you joined us for, on Ahead of The Curve, you spoke more about the customer side of applying AI, how to improve the customer experience, and how that... The future will look with AI in that respect. With this episode, we're excited to talk about the business of airlines, of airline operations, other aspects of how you... How airlines get passengers from A to B, and how that can be better with AI. So I'm really excited to cover that topic with you.

Michael Wu: Yeah, me too. Yeah. [chuckle]

Surain Adyanthaya: I guess, Michael, just to kick it off: At the broadest level, AI can touch many things, we know, but how can it really impact the business operations of an airline? Where are some distinct areas where you see AI providing great value?

Michael Wu: Yeah. So I think AI can actually do a lot of things. Pretty much every single AI system out there is gonna help you... I would say, operate the business more efficiently and more effectively to some extent. There's almost kinda, I would say, infinite number a lot of [chuckle] variations of AI out there. But to make it a little bit more kind of digestible, I would say that there's actually three main kind of outcome that AI can drive. One is that it actually... Because AI automates a lot of human tasks that are kinda somewhat inefficient in the past, they help business operate with much greater efficiency. Okay, so improved efficiency is one big area. And also because AI is obviously, there's some investment... Initial investment in using this type of technology, but ultimately in the long run, AI can definitely help airlines or other business, I would say, cut costs, reduce costs. And finally, I would say, the last area is that AI can actually help you make money, [chuckle] it could drive revenue lift. So those are the three... In the broadest sense, these are the three area I would say AI can help. Yeah.

Surain Adyanthaya: Right. For sure, those are all very important to our airline customers around the world. Can you drill down a little bit on the efficiency side? How can AI help airlines operate more efficiently, and give us some more details?

Michael Wu: Yeah, so saying this from the last episode, we talked a lot about these consumer-facing type of AI technology. All those, I would say, would improve efficiency to some extent, but because obviously in the airline operation, I would say that the employees are the agents for their airlines. They are kind of the... I would say, at the bottom that... You can only have so... You can only hire so many people. So there's a limited number of them. There's always a lot more customer that they're trying to serve, than the number of agent that they have. So any kind of, I would say, these consumer-facing technology, whether it's chatbots or using facial recognition for check-ins and checking bags and all those, they... 'Cause since you don't... No longer need a human there, so this alleviate that bottleneck, of being... The number of agents that you have... You can actually employ. [chuckle] So that... All those would definitely make, I would say, a lot of the, I would say, airline or airport operation a lot more efficient.

Michael Wu: But I would say another big area is probably predictive maintenance. Predictive maintenance, is all about... You're gonna service a plane anyway, right? You're gonna service the plane at some kind of schedule, right? Why not serve... Service the plane, or service whatever machinery that you have, at the optimal time? It essentially introduces the least disruption to operation. Essentially, if you have... If you wait until something breaks to kind of service it, there is gonna be some disruption. There's gonna be some disruption, and then you're gonna have to find a replacement... A new part and then put... Get 'em in there to working, and you have to... So that is always gonna be a disruption, when you actually wait until something is broken. So... But if you're able to... Be able to do that in a proactive way, right before [chuckle], or maybe just before they... Things start to break, then you minimize the disruption. So as a result, things run more smoothly, and I would say that will improve efficiency as well. So, yeah.

Surain Adyanthaya: Right. Yeah like...

Michael Wu: I think one area that's actually... I heard this a while back, is that they're actually are doing this, what well would they call turnaround optimization. So when the plane... Between... The time between the plane land, and when they take off again, there's this turnaround time, lots of things have to happen: They have to refuel, they have to clean and disinfect, especially now with COVID, they need to load luggages, have catering, load the food and beverages into the plane. So lots of things have to happen. And I think that companies are actually using video analytic to actually see what's actually happening to this plane as they're... Between their landing and another... And the next take-off. And they can actually see, are there anything that's actually, I would say not coordinated well. Are they waiting on something, are they actually waiting on another cleaning crew that's cleaning another other planes? Should they dispatch another set of cleaning crew to service this plane? So they can actually see what's going on using video analytics. That's obviously AI based. You have to recognize what's happening. And to coordinate this process so that, again, this turnaround time can be minimized. So...

Surain Adyanthaya: Yep. No those are all very important aspects of the airline business, for sure, that have great value to airlines. I think the other point to the examples you brought up would be consistency of product and service.

Michael Wu: Exactly.

Surain Adyanthaya: Because I think that would dramatically improve with the application of AI technology, as you've described them, which I know airlines around the world are striving for. That's really great.

Michael Wu: Yes, yes, that's...

Surain Adyanthaya: So efficiency is very important, but one of the major levers that all airlines look at, and really focused on by LCC, is low cost carriers, etcetera. Or reducing cost-saving money. So how can AI help save money for the airlines?

Michael Wu: Yeah, so I would say that the there's... I know of two use cases that are... That's already probably in practice now in some airlines. One is actually forecasting kinda the in-flight kind of food consumption. Actually, a lot of people are... Based on their travel habits. Some... Like I said, some like to just sleep and not be disturbed, some like to enjoy a meal, but based on their kind of... Each individuals kinda travel habit or... And even like kind of the time of the flight, they know maybe certain people don't like to eat in the morning, and some people don't like to eat at... In the evening or... So based on all that kind of big data, they can actually forecast how much foods they should actually onboard a particular plane. And you can actually forecast it accurately, so you have just enough or maybe just a little bit more [chuckle], just in case. Then you minimize this kinda food waste. So you lower the kind of expected that amount, so you avoid this food waste, so you don't waste money. Every item that you throw away is essentially a waste.

Michael Wu: So I think that certainly could help save money a lot. Another use case I'm... I know of is certainly in this fuel optimization area. I know that... I don't know how much fuel costs for the airline. My guess is probably between... Quite high. [chuckle] Somewhere between probably 20% or 25%. So if he... AI could actually recommend, for example, a more fuel-efficient route, that could save... A number. I would say that like a large... A huge... I would say a... Fuel cost is actually a huge component of the airline's cost. So if they could actually save a little bit of fuel, that's actually a... Quite a... Quite significant for the airline. For example, I know that for a... Long time ago, I even hear this. Some planes could even get re-routed to take advantage of the jet stream so they get more tailwind, so they actually get to their destination faster and earlier, and also save fuel, and also to avoid storms by taking advantage of the weather system... Data that we can get, right? How do you route it... The plane? So obviously all this... Once you say fuel, when you save kinda on food waste, it's quite dramatic when you actually think about how big this... Airline... The scale on of which this airlines operate. So...

Surain Adyanthaya: Right. For sure. Well, I think cost-saving is certainly very important to airlines, and those are excellent use cases, but the other very important use case that we focus on quite a bit at PROS is how do... How can we help airlines make more money? Can you talk a little bit about that?

Michael Wu: Yeah, yeah, definitely. So we talk about this personalization idea last time. I think this is actually one area where AI can help airlines essentially monetize more, more efficiently and more effectively. 'Cause obviously if you're able to make an offer, for example, with a really hyper-relevant ancillary, that are more likely to be purchased, right? The person's probably gonna click that, and buy it, and use it. So and actually, we try more sales of those ancillary. So personalization would definitely, I would say help in this kind of revenue lift, I would say. I think that the second area is probably... Obviously, it's revenue management. It's something that we have been doing for the airline for the longest time, and it's... And obviously a core competency of PROS. So I think these are definitely the two area that AI can help, I would say, airlines to make more money. [laughter] I don't...

Surain Adyanthaya: I know. For sure. And we've discovered that with this new generation of travelers that started with the millennial traveler, because they're... They grew up with mobile devices and online purchasing, they expect a level of personalization and the selling entity, in this case the airline, to know them and to offer relevant, transactable offers. So it's certainly very important. You bring up a really good point. And I know for the last year and a half, Michael, you've been leading a team of scientists at PROS to innovate around revenue management. Can you tell us about some of these things you've been working on, and what your findings have been?

Michael Wu: Sure. Yeah, I know that's coming. [laughter] So yeah. So obviously, one huge area that we're innovating on is to improve these revenue management system by leveraging more different data sources. Okay. So I would say a lot of revenue management today, they use, I would say, traditionally they use kind of the historical kind of transaction, the seasonality and kind of capacity. These are data that the airlines have. They own them, they actually have 'em, so it's easy to get those data. But with COVID disrupting, I would say, the travel behavior of people, airlines have a harder... Much harder time to forecast the future travel kind of demand. Essentially, this is actually a big challenge. If you can't forecast the future travel demand accurately by using historical data, you gotta learn... You gotta leverage something else. You gotta... They gotta be able to... There had to be some other signals, that indicates people want to fly or their desire to travel. We need to get that data, get that signal, get that information somewhere. And so one of these data source that we've been exploring is shopping data. Obviously, it's common, I would say, that... For people to kind of look before they actually book like people shop around for a flight, and then eventually they make a booking. Yeah, so this is one data source that we use and we are able to kind of see some very promising performance lift by using... Leveraging this data source.

Surain Adyanthaya: Right. So I guess, in revenue management, we always were of the mind of the past is the best predictor of the future. So what you're doing with your scientists doesn't disregard that, but it enhances what that thinking with, "Let's take current information, what's happening right now and enhance our historical information to predict the future." And I think it makes all the sense in the world.

Michael Wu: Yes. Yeah.

Surain Adyanthaya: We've heard that very often from our airline customers that it's such a dynamic world that we have to be in the now, and this sort of captures that doesn't it, Michael?

Michael Wu: Yeah. Yeah, totally. And I think that, this is... Historically, this is has been very challenging, because... Think about this airlines actually don't own all the shopping data out there. So in fact, there's so many different channels that people could shop or kind of look for a flight. The data tend to be very, very fragmented. And airlines don't always have access to those partners. Or...

Surain Adyanthaya: Right.

Michael Wu: And because of that there's so many different channels. There's... These shopping datas also have, I would say, very different coverage. Some, like I would say, channels are really good in certain region, like say in Europe or in some... And some other channel maybe really good in other regions... So... And different people use different kind of channels in different places. So the...

Surain Adyanthaya: Right.

Michael Wu: The kind of consumer coverage for each one of these shopping data is different. So that makes it actually very difficult to leverage these shopping data as well. You may think that this is pretty obvious, but why haven't airlines do this in the past. So it is mainly because of this, datas are fragmented, they don't have access, and also like even if they have access, it's hard, because the coverage is very different, and they don't know how these channels are covered in different part of the world. And finally, I would say, one big challenge is that there's a actually huge, I would say, variability in kind of consumers shopping behavior. Some people would just look book for a flight for a couple of days before, and then they book right away, and then another one could look in a... Research for [laughter] a month before they actually make the booking, and some even could even set up, I would say, scripts and to monitor flights and... And so yeah. So this all makes it very, very challenging, yeah.

Surain Adyanthaya: So, Michael, can you tell us all about how well this data performs? I mean, what sort of benefit are you seeing from utilizing shopping data and other data sources?

Michael Wu: Yeah. Hugely. I think that by leveraging shopping data, we are able to see that during this COVID time where everything is kind of being disrupted, if you actually leverage shopping data, you can actually get up to 40% kind of error reduction in your forecast error reduction, which is very, very significant. When you talk about revenue management, you typically... Improvement are kind of single digit or if... When you get the double digit it's already kind of very, very, big, but this is 40%. It's almost unheard of.

Surain Adyanthaya: Very exciting. Very exciting.

Michael Wu: Yeah. So we actually kind of were little bit... The team was a little bit kind of cautious. You know what I'm saying? 40% improvement is... It's almost too good to be true. We were a little bit kind of suspicious of the results ourself, being good scientists ourself. So and... But after some investigation, we found out that in a like... Largely this is due to a, I would say, call; a kind of low denominator effect. So because COVID is such disruptive force in the airline industry that it made it very difficult for kind of traditional... The traditional revenue management forecaster to forecast people's travel behavior. So the traditional forecast is not doing very well, so you'll essentially lower this denominator, dramatically, and then... Yeah. So basically that the improvement that you see, a large part of it was actually due to the fact that the traditional way of forecasting is not performing as well during this disruption. So as a result, we kind of said, "Well, can we actually look before COVID?" Fortunately, we have some very nice data partners. They're able to provide us some shopping data even before COVID time. And we were able to see before COVID, this improvement was not as significant as 40%, but it was still like around 20%. So this is still very, very dramatic, So... [laughter],

Surain Adyanthaya: Yeah.

Michael Wu: So definitely there's a lot of signal in shopping data. So...

Surain Adyanthaya: Yeah, I think that's huge. I'd go as far as to say, it's sort of a breakthrough in demand forecasting, in revenue management, what you and the team have done. So I'm very excited to see it in production in the airline world. Very excited. I guess, Michael, we've spoken about shopping data, but what about other forms of data, like events data etcetera? Can they help as well?

Michael Wu: Yeah. They can help. We actually did explore that a little bit, and using events data to kind of predict, essentially, people's shopping behavior. You know you may think that, "Yeah, great. Shopping data is great." So you... Once you have the shopping data, you could kind of forecast future demand, but for you to get shopping data, people have to shop, right? [laughter] So, yeah. So if people don't shop, you're not gonna have shopping data. So the problem is that people don't shop... Typically, okay, don't shop for a flight until probably a few months on only the older a few months before, before the flight departure. If they wanna leave or stay in for Christmas or something like that, they'll probably look for a flight in September, October or something like that. Typically, that's what most people do. So you're not gonna be able to kind of forecast far in events like you used to be, like for up to a year in events.

Michael Wu: So that's the problem. If you... To use shopping data to help you improve forecast, you need shopping data, but people have to shop. So if people haven't actually start started shopping, you're not gonna be have those data to help you forecast the far future. So this is where events data can come in, because events are typically announced or kind of planned much, much earlier. So they can actually... We can actually use those events data to kind of estimate what kind of shopping behavior they were able to drive. Okay? Some big events like a... The football game or the World Cup, or whatever. They would drive a lot of, I would say, shopping behavior for these destinations. So if we can... If there... If we know this events way early, we can actually use this events data to kind of estimate what is the impact on shopping would be, and then we use that as a early kind of a indicator of how that could impact future demand in travel.

Surain Adyanthaya: Right.

Michael Wu: So yeah.

Surain Adyanthaya: That totally makes sense, and I'm... I am sure that as we go down that road, you'll find that there's tremendous value in using that data as well. This is so exciting. As someone who's been in the revenue management field for a while, breakthroughs don't come through every day. So this is exciting stuff, for sure. So while we're on this topic, can you give us some more use cases where AI can help an airline with it's operations and efficiency?

Michael Wu: Well, let me ask you a question back then like, since AI is very good at automating I would say, these mundane and repetitive tasks that's typically done by a human, let me ask you, are there some... What are some of the most mundane and repetitive tasks that airline must do today that are... Because they're still very important?

Surain Adyanthaya: Well, there are things such as irregular operations, flight disruptions that require re-booking of hundreds or thousands of passengers as storms roll through across the US or other parts of the world. That's... I feel like that's a very important repetitive task, and also there are tasks like just the communication channels between the airline and the consumer along the entire traveler journey, informing about delays, about boarding procedures, etcetera. It's extremely important and helpful.

Michael Wu: Yes.

Surain Adyanthaya: I think that's another area that's pretty repetitive, but could be very intelligent, potentially with AI. Just a couple ideas.

Michael Wu: Yeah, I think... So I would say that those are the precise needed area that AI can help, because these are... Because they need... Obviously, they're important, because airlines are still doing that today and... But they're rather repetitive, so if you actually collect data on how AI... How a human have handled this situation in the past, and used those data to train the AI to mimic what these humans do in the past, then those are... There's your case... Your use cases. So yeah.

Surain Adyanthaya: I see what you did there. You turned the tables on me. [laughter]

Michael Wu: Yep. [laughter] You can't ask all the questions. You know?

Surain Adyanthaya: So... I'll remember that. So airline industry is really, I think, fortunate because it's extremely data-rich, airlines have used data from for many decades to... For important decisions, support, things like revenue management and other things. So how do you see the airlines going from where they are today to sort of a more AI augmented big data strategy in the future?

Michael Wu: So moving from this kind of tradition of business operation to AI, there's actually a fairly standard kind of a, I would say, maturity trajectory. So typically a company will go from no data to have some data [chuckle] right? And when they have some data, the first thing they do is that they do what they call descriptive analytics. They just summarize the data, so that they present this data to some decision makers so they can make a decision from those data, and then they move into... Once they have more data, they can actually start to use this data to build predictive model to estimate some quantity that they don't have before, and then again, but they will use this kind of information about their prediction or of their forecast, present those to the decision makers, so they will help them make better decisions, like having an eye into the future. And finally they go from, I would say, descriptive to predictive to prescriptive. The... Now these systems are able to kind of tell you what to do, "In this situation, you should do this because that optimize a certain business outcome." So... But at the end of the day, I would say, this type of going from descriptive to predictive to prescriptive, humans still make the final decision. AI could tell you what to do, right? But at the end of the day humans say... I can always say, "I don't want to follow what the AI recommend. I'll do what I think is correct. Or what I think is the best way."

Michael Wu: So... And then... So the thing about making this transition to this AI future is that we... Since AI essentially automate this human decision and action, and they're able to learn, so you need to kind of essentially to make... Give AI the opportunity to make those decisions and actually learn from it to get the data. So I would say that airlines are actually, yeah, definitely much more mature than lots of other industry, in a the sense that they've been doing predictive analytic and prescriptive analytic way before, I would say, the consumer world even catches on. So I would say, and then to move to this next step of being AI augmented kind of future, they just have to essentially let AI make some decision. Like give them some... I would say be brave and be courageous enough to let this AI make decisions so that they can actually learn and be better at making this decision in the future.

Surain Adyanthaya: Right. Got it. Well, this has been a fascinating discussion, Michael. Thank you so much for joining. As always, really insightful and eye-opening thoughts and ideas. So I hope we can talk more in the future on some of these topics as well.

Michael Wu: Sure.

Surain Adyanthaya: Thank you so much.

ABM Resources Test Stream for Your {demandbase.industry} Industry