Making Data Simple: How data science is helping to improve aviation
30.40 Copyright YanniVEVO, all rights reserved to Yanni. Song Name: The Rain Must Fall. Listen here, http://bit.ly/2iU3uDU
Hungry for more? Check out our previous podcast episodes of Making Data Simple:
- Episode 1: Making Data Simple: The big data problem
- Episode 2: Making Data Simple: End of tech companies
- Episode 3: Making Data Simple: A new definition of client care
- Episode 4: Making Data Simple: Will machines take our jobs?
- Episode 5: Making Data Simple: Growth Hacking - Not just for start ups
- Episode 6: Making Data Simple: From 2D to 3D -- Augmented reality data visualization
- Episode 7: Making Data Simple: The 5 areas businesses MUST get right
- Episode 9: Making Data Simple: Making data fun & easy with Caleb Curry
- Episode 10: Making Data Simple: Data movement at size and scale
- Episode 11: Making Data Simple: Cloud computing, part 1
Al Martin: Hey folks. Welcome back to the series making data simple. This is Al Martin here. Today’s topic is data science in aviation. We could maybe spend all day because I fly a lot. So, I may have a lot of questions for Ioannis Gamvros. Hi IOANNIS how are you?
Yianni Gamvros: Hi, good.
Al Martin: That’s Yianni, like the composer. Did you know that our producer (Kate Nichols) — she’s a huge Yanni fan and I am told that she went to a Yanni concert? I —it’s the only one that I know of. But, she must be — I presume you’re a different Yanni.
Yianni Gamvros: I am a different Yianni, yes. Absolutely.
Al Martin: Very good. I had to embarrass her so as — while she’s listening on here. So, I’ve got a lot of friends that are pilots, so this is going to be an interesting conversation as far as I’m concerned. And I ride in a lot of planes. I think I’ve been traveling every week. I got (sic) to tell you, the last flight I was on — so, we took off from La Guardia. And — I mean, it was the worst turbulence that I’ve ever had. I mean I was going up and down and it felt like, as I’m coming up — it felt like we were hitting something or something. Because it would just like shake — like, you know, just a repetitively shake, and then stop, and then do it again. And, it was terrible.
Anyway — so then, finally I got up, I — you know, about 10 minutes later it calmed down and I went to the washroom. And when I went — you know, they hadn’t turned off the seat belt sign yet, and I said to the flight attendant, I said, “I hope you don’t mind. I’ve got to go to the washroom after that.” And she goes — she goes, “Well, you know what that was, don’t you?” And I said, “Turbulence?” She goes, “No, no, no. We were hitting jet wash from the last plane.”
Yianni Gamvros: Okay, yes.
Al Martin: And I was like, “I wish you wouldn’t have told me that.” She goes, “Well, didn’t you feel us turn sideways very quickly?” I said, “Yes, I felt that. But I didn’t know we were hitting jet wash.” She said, “Yes, we had them backed up too close together and there was a big plane in front of us.”
So, hopefully, with data, you’re going to tell me how to solve that. Now, I don’t know if you can do that. But look, I do fly a lot, so I’ve got a lot of questions. So, speaking of airlines, introduce yourself if you would and tell us what experiences that you have with the airlines and what you’ve learned. And then we’ll jump into data science.
Yianni Gamvros: Sure, sure. So, currently, my role in IBM is the Global Enablement Leader for data science and machine learning. So, I teach our sales representatives how to talk about data science. How to present the value proposition, and how to present the key differentiating points for the IBM portfolio.
My experiences in the airline, just like you, I used to be a road warrior. Not anymore. I currently have an eight-month-old at home, so for the last year or so I haven’t been traveling as much. But I do have a lot of experience from the time when I was a consultant — a data science consultant. And I talked to a lot of different airlines about the different challenges they face, and we looked at several issues — operational most of the time. And we looked at the back end and how they’re dealing with those issues and how we can help — and how data science can help them.
Al Martin: Okay, very good. So, look, when I’m flying I’m thinking about all kinds of things. And usually, I don’t know I’m — usually, it’s not always a great experience. I’m just trying to get from point A to point B. But I got to (sic) believe that there is data that we should or could be used to better that experience. Can you talk about, you know, the importance of data in the aviation industry? And then we can go from there.
Yianni Gamvros: Sure, absolutely. So, I mean there are a lot of things that the airlines are tracking, right. And most of the time they actually do a pretty good job. And most of the time you don’t even know that they have all these models and all this data that’s being accumulated in the back and the way they’re using them, you don’t even know that that’s happening behind the scenes, right.
So, the one thing that you might notice — or, since you’re a business traveler actually, you might not be as price sensitive. But, for most people out there, when they go out, their first touch point with the airline is the ticket price, right. And in many situations today you can find a very, very cheap price and actually, most people are really happy about that. And that’s actually — there’s a lot of data science behind the scenes to actually get you that very cheap fare, right. And then there’s everything else that happens to make sure that your flight leaves on time, the airplane is ready, the equipment is ready, all the maintenance checks have been done, the plane passes all the FAA checks, and the pilot and crew are certified to fly that plane, and the ground crew does all their checks and all their, kind of, final tasks and prep for the plane on time, right.
So, all that happens behind the scenes, but actually there’s a lot of operational processing. There’s a lot of operational data behind the scenes that drive all of these decisions even before you actually take off, right. And then after you take off, there’s obviously a lot of things that the dispatchers and the pilots are tracking. Specifically, with the weather to make sure that the routing is such that you don’t hit all turbulence (sic) spots, right. And then — that you get there one time through — and in the safest — and, you know, fastest way possible.
Al Martin: Well, what I’ve heard — and you tell me if this is correct — but I know IoT is still in its infancy. For the most part of the industry, as far as I’m concerned. I mean, connected car has a long way to go. Being in a smart building - a truly smart building, top to bottom - still a long way to go. But from what I understand — maybe you can confirm this or not — but, the engines on the jets that we’re taking are very well connected and they go back to a base where people are monitoring, you know, everything top to bottom within these engines itself. Is that accurate? And could you (unintelligible).
Yianni Gamvros: That is correct. Actually, the engines are one part of the plane that’s actually much better connected than anything else on the aircraft. Actually, in many situations today, and for many aircraft types, the people in the back that are riding through their wifi, they have better information than the pilots up top. So, unfortunately, that still is an issue. But the way that the airlines counterbalance that is they have people on the ground that — the dispatchers that are basically looking at weather systems, they’re looking at routing data, and then they’re informing the pilots, right.
So, yes, it’s not — we’re not fully there in terms of IoT connectivity, but we’re getting there.
Al Martin: So, are you a pilot yourself?
Yianni Gamvros: Oh no, no, no. I’m not. I’m not. Just an enthusiast.
Al Martin: Alright, well fair enough. And are you working some of the data science within these airlines?
Yianni Gamvros: Yes, so some of the projects that I’ve been personally involved with have to do with, for example, matching pilots and crews with the airline schedule. I mean, that’s a very, very complex problem to solve. You need to make sure, obviously, that you have the right pilot that’s certified for the right equipment type. And that you leave enough time for the pilot to go from one flight to the next, and then you give them enough time overnight to rest. And hopefully, you get them — at the end of the day, hopefully, you get them back to their home base, right, so that they can sleep at home. Obviously, that’s not always possible, but that’s the ideal situation, right.
And then there are other issues with ground control, right. So, making sure you assign the right gate to the right flight, and the gate can support the equipment, that the gate can support the flight and you don’t have this very, very large jet sitting right next to each other, and kind of there might be some safety concerns there.
So, that’s another project that we did. For another project that we’re doing actually, in collaboration with Weather.com is to recommend which flights get diverted once a big weather system an airport and is going to impact the capacity of the airport to accept new flights, right. And so, you have to slow down the landing grade for that airport, and then you have to figure out, basically which flights get diverted and where do they go.
Al Martin: So — but tell me — here’s what I don’t understand, what’s the difference between traditional analytics that is being performed on the data that you just described or what is really the magic of data science coming in and making these assessments. So, I mean, more like machine learning models that you’re using.
Yianni Gamvros: Right.
Al Martin: Which is which?
Yianni Gamvros: Right, no. Very good point, very good point. So yes — so there’s actually a ton of traditional analytics — if you want to call it that — and BI, business intelligence — that the dispatchers on the ground and the operations people the maintenance people, everybody that is looking at the back-end operations is looking at. Which is (sic) just basically gives them just a current state, right. A view of what’s currently happening.
Where data science and machine learning and some of the other methods that airlines are using — like, for example, mathematical programming, constraint programming, which are other data science methods not as popular as machine learning right now is — they’re looking to get recommendations from those models and algorithms in to the hands of the dispatchers and the operators so that they can more quickly react to very dynamic situations. And they can have these recommendations in front of them, and then they can make the judgment call as to what — how I’m going to deal with that special situation that has just come up, right.
So, those are — that’s the difference between the more advanced data science models and algorithms. Providing recommendations and providing instruction to the back-end operators
Al Martin: I mean, so — are these data science algorithms currently in use today? Are they future state kind of things, is that where we’re going? I mean, kind of give me a feel for where we’re at and where we’re going, relative to data science typically.
Yianni Gamvros: Right. So, in most cases, there’s only partial adoption, right. So, in some cases they’re being used today, but there are many, many other processes and operations where they’re being built right now.
Al Martin: Alright, so — but — let's — okay, you had mentioned something and it’s kind of a silly question in some sense. You said, there are folks on the ground — you referred to that. Where are these folks? And the reason I ask where these folks are, because when I think of data science I think if — you know, I think of models, I think of machine learning, I think of projecting and anticipating, predicting. And I — you know, I could be proven wrong, but — you know, to do that right you have to have the data science seeing these models working and then being able to better the models as they go while the software kind of betters itself if that makes sense. So, you correct me — you know, my thought process there, if you will.
Yianni Gamvros: Yes. So, the operators are sitting at the network operation centers for all the airlines. All the airlines have at least one or two of them and they have an army of people there, right, that’s doing all the dispatching and it’s tracking all the planes, tracking weather, tracking all the pilots and crews and where are they going to stay overnight, and, you know, what’s their next flight, and so on and so forth, right.
But you’re right. So, at the same time they are collecting all this data, right. And the data scientist within the airlines are kind of sitting to the side. They’re seeing how existing processes are at work today and they try to inject, basically, technology and try to inject these models, the predictive models, the operations models, in the hand of the dispatchers in order to help them out.
Al Martin: What is the, like, the most profound change because of data science, if any, that goes in so far today, already that we should be feeling?
Yianni Gamvros: Sure. So, that’s — actually, it started quite some time ago, right. So, it has to do with pricing. So, back in the ‘80s I think it was American Airlines basically, that pioneered the very dynamic pricing. They call it yield management, where basically they keep track of their seat inventory, they keep track of the prices of competitors, they keep track of demand, and they also forecast demand and they forecast how sensitive people are to prices. And they’ve kind of figured very quickly that people are very sensitive to prices and that’s why they need to constantly update the prices for the seats they have on the airplane.
Al Martin: What about the customer experience? Just the — I mean, so now we’re done with pricing, we purchased the ticket — the overall customer experience. Here’s what I can’t understand, I’ll tell you. Why is it that there’s so many overbook situations that still exist today? Just on — or, I was just getting on a flight the other day and, look, they went up to over 1,000 bucks if somebody would give up that seat. Obviously, they’ve learned from recent events that you’ve got to go to the point where somebody finally bites on one of that - on that offer. But, why does it even have to get to that point if we’re heading in to big data science land?
Yianni Gamvros: Right, right. Very good point. And here’s another point of view. If you ever get on a flight and the flight is half-empty, right, you find empty seats, then the data scientists at that airline are not doing a good job, right. So, overbooking is the result of data scientists actually doing a very good job and figuring out the fact that in many situations, people just miss their flight, right. So, they might — they cancel last minute, or their plans might change, or they might be stuck in traffic on the way to the airport and they just might miss their flight, right.
So, when that happens, we want to avoid that seat flying empty to its destination. And the reason they do that - the way they do that is by actually booking additional passengers. So, in most cases — I want to say in close to 100% of the cases, all flights at some point are overbooked. But, then when — as you get closer to the takeoff time, then you have these last-minute cancellations and then most of the time it works out that you have just the right number of passengers and the right number of seats.
Now yes, sometimes it doesn’t work out and we hear about it through the speakers when they’re trying to find a volunteer. And that’s when you have those incentives come in.
Al Martin: So, are you actually making a case to me that because there are so few overbook situations that we see, that’s actually a testament to how well they’re doing in terms of pushing the envelope as close as they can get to an absolute - or a plane at capacity?
Yianni Gamvros: Exactly, exactly. That is absolutely right. Remember anytime you get on a flight and there is even one empty seat, then someone hasn’t don’t their job correctly in the data science department of that airline.
Al Martin: Hmm, I’m going to have to think about that. All right, sorry. So, what can we expect in the future though? Alright, so, we’re putting these things in place, it sounds like, you know, we’re just on the cusp — at least as far as I can tell, unless you can convince me otherwise. What should we start seeing here in the new future that should excite us?
Yianni Gamvros: Right. So, it will be transparent to the end user, right. So, the end user, in the end will just go online, book their ticket and their flight will leave on time and it will land on time, and they will be on their way, right. But behind the scenes what airlines are doing is basically making sure, again, that they are taking in to consideration all the operation constraints, all the safety constraints, the maintenance is done right. Many of the problems that airlines currently have has to do with all the maintenance checks that the FAA imposes on them and they have to ensure that they get that plane off of the operational schedule and get it to a maintenance facility to — for people to work on it, right.
So, all these processes right now, or a big percentage of those processes is happening manually. And what I expect to see in the future is that more and more of these processes will actually be driven by more advanced data science models and algorithms and they will track the complexity of all these moving variables. They will look ahead, and they will provide recommendations basically to the operators and the maintenance team, the operations teams, the dispatchers, all those guys on the ground.
Al Martin: So, I correlate that to, you know, more on time flights, no more sitting for 60 minutes on a tarmac waiting for all the other flights to leave because of whatever reason, and continued pricing optimization.
Yianni Gamvros: Yes. So, actually I think pricing optimization is one of the reasons why some have these travel horror stories, right. So, actually a friend of mine very recently flew from Chicago to San Francisco. He was flying with his wife, two very young kids, and he was telling me that he was having this, kind of, horrible flight right. Where they basically got to the airplane, they weren’t sitting all together, there was no entertainment so his three-year-old was kind of, you know, nagging throughout the flight, and had just this - one of these horrible experiences.
And when he told me that I asked him — my first question was, “Okay, so how did you book your flight?” And basically, he said I went to a search engine and then I picked the cheapest flight that was at — that came back, right. And, you know, we as consumers have been doing that, right. And we’ve been doing that for such an extent (sic) that even maybe a $5 fare difference is actually a making a difference in the ticket that we’re going to purchase, right. And so, basically, I told him, “Well, there you have it, right. You basically had the experience that you paid for.”
So, actually it’s more of the mismatch of — it’s more of a communications mismatch, right. So, airlines are kind of pushing away out of the list price - out of the basic price for each seat, they’re pushing out some of the additional amenities and some of the additional services that they offer, right. And most consumers are actually not aware of that, right. So, they par for just the basic seat and then everything else is essentially extra.
Al Martin: Yes. Are we getting to the point where we’re going to be paying for everything extra? It’s getting — it’s heading in that direction, is it not?
Yianni Gamvros: Exactly. And I think we as consumers have officially, you know, done it to ourselves, right. I mean, since we’re — have these very constant and very specific behavior whenever we book our tickets, I think it’s going to go that way, for sure.
Al Martin: Seems like there’s still a lot of opportunity to be had here though. Because, you know, I think this is tough business for many reasons. To your point, we want it all for the cheapest price possible. But, like, you know just the other day I was rebooked when I was in the air. The only problem is, I was rebooked, like, the following day because my, you know, my next lag was — there was something wrong with it or whatever. But so, then I had to get off the plane and then I had to find the desk and I found a better option.
But, hopefully, the machine learning eventually would take over such that, you know, it’ll learn my preferences, what I need, what I’m willing to accept, what I’m not willing to accept. So, you know, it’ll present me with the options I’m most inclined to take. Either — whether that’s the — just the travel option, and/or, you know, what seat I want in terms of — you know, just the travel itself, I guess is what I’m trying to say.
Yianni Gamvros: Oh absolutely, absolutely. And I know for a fact that actually there are some solutions like that in play today currently with some of the airlines. But, actually most of the time they just look at rebooking everyone automatically once the flight gets canceled, and kind of, they try to do the best out of that. And obviously, they give preference to passengers that have higher status with an airline and, you know, if you’re a casual traveler obviously you’re not going to get preferential treatment.
But it sounds like you travel quite a bit, so I’m surprised that - you know, maybe your airline didn’t really have that in place.
Al Martin: Well, I mean in fairness — I mean — I think it — I think this year I’ve been on-time more than any other year. But, that probably is still 40% not on-time if that makes sense. So, 60% on-time. And it’s improved, but, you know, there’s a lot of weather implications, there’s a lot of different things. As long as it’s safe, you know, I’m pretty happy, I guess.
Yianni Gamvros: Actually, yes, Al, that was going to be my other point then. You know, you — we can write all the data science models and we can kind of throw all the math we want at the problem, but then, you know you have a storm over kind of a major hub, right. Or a storm system moving throughout the country that you know, the flight paths have to cross and then it — all of that goes out the window, right. I mean, you’re not going to avoid the delay when your airport capacity goes down by 50% because of the storm, or more.
Al Martin: Well that — that was actually — I would expect that some of the best opportunity to really get innovative would be to match up the data science — and you mentioned a little bit of this earlier, I thought — with, you know, the weather data and the weather patterns in making, you know, somewhat on-demand decisions based on, you know, that information. Models, you know — so you’ve got the weather model, you’ve also got, you know, everything else — you put the — you know, where the flight’s going, you know, who else they’re going to be waiting for on the tarmac, whatever the case may be. But I think putting all that together I think would bear huge fruit.
Yianni Gamvros: Oh, absolutely. Absolutely, yes.
Al Martin: So, look, you know, when I typically look at a maturity model for data, it goes from operations where (sic) back office type item or issues, then it goes to warehousing, then it goes to more data science, and then new models beyond that. Are almost all the airlines right now — are they pushing in to the data science area, do you believe? Or still some of them stuck in back office, warehousing? Where are they really at from a maturity curve perspective.
Yianni Gamvros: I think they’re close to pushing at the new model envelope. I think, especially the major airlines for sure have a big data science department with a lot of people working there, very knowledgeable people that have studies, you know, hours and hours at school on how to build the models and you know, what are the latest approaches and the latest algorithms.
Some of the other airlines that might be low cost airlines maybe outsource some of that to independent software vendors and other packages that are out there that have some vertical solutions that might do some data science, kind of — you know, within — off the self kind of data science, if I can say that.
Al Martin: All right, fair enough. Hey, I know that there’s some videos out there on Watson analytics around aviation, is there anyone that you’d recommend? We could put it in the notes of this podcast. But I didn’t know if you wanted to make any mention to it.
Yianni Gamvros: Sure, yes. So, there’s a great video actually where — you know, it tells the story basically of a passenger kind of waiting for their flight. And they’re not an expert, they’re kind of a very novice, you know, data scientist and they want to analyze some data and — analyze some flight delay data and Twitter sending them data and Watson Analytics makes that very, very easy for them.
So, you. So, if we can put the link in the podcast that would be great for people to see how that works out.
Al Martin: Fantastic. So, look, I’ve got a problem that I need you to work on, on aviation. So, in addition to the last flight I was on - so, I go turn my rental car in and the rental car center now because these airports are getting bigger and bigger, is, like, three miles it seems like, from the actual terminal. So, I take a train, and I hit like three, four, five stops. I get to the terminal, I walk probably another half a mile to get to the TSA pre-check. I get through the TSA pre-check, I look in my jacket, and I see the car keys.
Yianni Gamvros: Oh no.
Al Martin: So, I — and I had no place to drop these things off. How can we predict that, and somehow, you know, I think — bin you could throw these things in. It was very, very painful because I had to re-track everything I just talked about. Barely made my flight. So, think about that. That needs to be solved, because with these key-less cars, that is going to be a major issue. I’m here to tell you. I guess everybody’s not an idiot like me.
Hey look, a — so, well, before I begin, I want to — three more questions, lightening round. But, anything else you’d leave the listeners with in terms of data and aviation?
Yianni Gamvros: Yes, I think what can make your flight experience much better is to really know what you’re paying for in that kind of basic price. And if you are someone that’s traveling or planning to travel more than four times a year, then you should really pick, you know, you own favorite airline and kind of stick with it. Because there are tremendous advantages if you stick with an airline and you (unintelligible) and kind of get in to that first layer, or level of status with them, then you can get some benefits out of that. So, I think that can be very beneficial. And just get what you know what you’re getting for the price that you’re paying. I think that’s very, very basic.
Al Martin: Alright, terrific. Thank you. Hey, so I want to do a little bit of a lightening round. It gets a little bit more personal. Just a couple of questions, it’ll be all good.
One question is — the first question is, what’s the most exciting thing you’re working on right now?
Yianni Gamvros: So, for the aviation industry specifically we did a project — we were designing this project really that had to do with which flight to assign to what gate, right. So, I mentioned that a little bit earlier in the podcast. But, I thought that was pretty interesting because we don’t give much thought to it, right. And it seems like a, you know, pretty basic problem. I mean, you have plane landing, you have a gate that’s open and you can say, “Okay. Well I’m just going to take that plane over to that gate.”
But it turns out there’s a lot of moving parts and a lot of complexity. You might have another plane right next to it that might need some additional space, so you might not be able to, kind of, put those two planes right next to each other. You might have to do some re-shuffling of the gates if you have some delays.
That’s when it gets really interesting, right. Which of your gates do you change? Which of your flights do you change? And how do you ensure that the new and the old gate for the flight that you are going to change is actually pretty close to each other, right. You don’t want to make these announcements over the speaker where you’re forcing the people to, kind of, you know, walk half a mile or a mile across the terminal to get to their new gate.
So, I thought it was very interesting. Very practical problem and kind of, huge implications on the day-to-day life of different travelers.
Al Martin: Is it going to get to the point — this made me think, you know, like in stores they put notes by diapers and that kind of stuff to get people to purchase more. Are they going to do that now with gate — when they switch a gate? You’re going to go by this pub that everybody’s going to stop and spend a ton of money, you know. Are they going to get that smart on us?
Yianni Gamvros: Ideally, no. Ideally you want to change to a gate that’s very, very close by so most people will be in the same seating area, right. They won’t even have to get up and they will — you just change, basically to that gate.
Al Martin: So, gate changes — that I certainly take for granted, that’s for sure. I’m sure there’s a lot more in to it then, again I think about. Hey, any book you’re reading right now? I always like to see whose reading good books that I can put it on my list and then I read it myself.
Yianni Gamvros: Sure. So, catching up on some to-read material that I had kind of noted away. So, I’m reading The Negotiation Genius by Deepak Malhotra and Max Bazerman. And I’m also reading The Goal which is all about the fury of constraints and how do you provide recommendations in very constrained environments and it actually applies — that specific book is actually written for manufacturing, but it applies a lot to airlines, and traveling, and just logistics in general.
Al Martin: All right, good. Good, we’ll put those in the show notes. I got them on my list. From a data science perspective — anything those that are data scientists out there that you’d recommend in terms of learning or where to go? I mean, something that you follow, whether it’s a podcast or otherwise, that you recommend?
Yianni Gamvros: Right. So, actually, we just came out with the Watson Data Platform blog on Medium. So, I think everybody should really follow that. There’s a lot of good material there on some generic topics on machine learning and data science. And there’s some more specific topics on features of the Watson data platform and the IBM data science experience.
And another Medium blog is Inside Machine Learning. Which, again, deals with different use cases and how you can potentially solve them using the IBM data science experience.
Al Martin: Fantastic. Where can listeners reach you? Do you have a Twitter handle? LinkedIn? We’ll put them in the show notes, but what’s your preferred form on engagement?
Al Martin: All right. Anything that I left unsaid?
Yianni Gamvros: No, I think we covered quite a bit. Thanks, Al.
Al Martin: Hey, you don’t listen to Yanni, do you?
Yianni Gamvros: I don’t, no.
Al Martin: You know, I did mention that (Kate) one of our producers listens to Yanni, didn’t I? She actually went to that concert. I mentioned that at the beginning, didn’t I?
Yianni Gamvros: You did, yes.
Al Martin: All right then, all good. I have to embarrass her a little bit because I know she’s listening. All right — and she’ll probably cut it out. If you cut this out, I will be very upset.
Hey, thank you so much for joining us today. It was informational. I’m going to go research the — some of the links and the blog that you’ll give me and I’m also putting those books on my list. So, thank you. I appreciate it. And for the listeners out there, I’ll talk to you next time. Thanks, Yianni.
Yianni Gamvros: Thanks, Al. Great talking to you.