Rewind: Sept. '09 Interview With ITA Software's Jeremy Wertheimer At The Beat Live

In light of Thursday's news, we thought readers might benefit from reading an interview with ITA Software co-founder Jeremy Wertheimer that took place in September 2009 at The Beat Live and was published that month in The Beat. The interview occurred a few weeks after Air Canada said it suspended work on a new reservations system constructed by ITA. Wertheimer took questions from Jay Campbell and audience members. He discussed the Air Canada development as well as other topics related to airline technology, faring and merchandizing, the complexities of making things simple, tailored selling by airlines and the point at which consumers get perturbed by too much information...

Campbell: ITA Software and Air Canada were working for many years with a lot of effort and a lot of money to build a host reservations system and you didn't even get to the scariest part, which is actually converting. How far did you get?

Wertheimer: The product is built. The analogy we like to use is, "I built this artificial heart, now it's just a simple matter of doing the heart transplant. Who wants to be first?" In a nice way, we spent a number of years learning exactly what they wanted. They were wonderful teachers, both in terms of what an airline needs and doing that schizophrenic dance of being very new and very forward-thinking in terms of the new things they wanted to do--unbundling and rebundling and having travelers take care of themselves instead of needing the servicing of employees--and also supporting the full range of legacy systems. Plus [they are a] big flag carrier, a big international carrier and the icing on the cake, a member of the Star Alliance, which has its own additional rules--in case you don't have enough. So they were a nice partner and that partnership is still there. We are still working with them and they are still supplying resources to help us. Essentially, the code for the inventory system and res system is done. The departure control system, which is the airport system, we're still working on. We'll be finishing those over the next year. We'll be rolling it out at a certain rate with Air Canada and they'll be making announcements over time. And now we're talking to a number of other airlines about rolling out the whole system or pieces of it.

A traditional inventory system is flexible the way Henry Ford was flexible at the beginning: You can have a car any color as long as it's black. You can have any product you want as long as it has a fare basis code that identifies it, and as long as it is settled a certain way. Our inventory system, when it wakes up has no idea what anything is. You tell it. It is all in data. It is not in software or burned in. So you can say, "This is the kind of product; we only have some number of seats and this is how we sell them. Here is another product called lounge access, keep selling those. Here is a product which is how many skis we can take on the plane and maybe that depends on the plane. Here is a product that is someone else's product, maybe hotel rooms, and here is how you handle that."


Campbell: Obviously unbundling and merchandizing has been a big piece of Air Canada's strategy over the last few years. Was the system intended in part to make all the features and attributes available to the industry at large, or are there industry-level systems that still won't allow that?

Wertheimer: One of the pillars of the system was absolute flexibility in how our product is put together. We should be able to change the architecture of our product second by second or minute by minute. One of the pillars was that all product details live in data and not in software. If they live in data, they can be changed instantaneously, or as quickly as you like. There will be a tremendous amount more of this work to come. We are just at the beginning of it.

From an airline perspective, they have to work enormously hard--someone used the example of Allegiant Air [charging a $70 base fare and $34 in extras]. Look at what they have to do for that first $70. They have to defy gravity and take this big piece of metal, fill it with fuel, launch it in the air ... a lot of work. And for that they get a tiny, little profit margin. They can make again that money by just selling you a pillow, or by selling you someone else's stuff. For all that people make fun of the bag fees, if you look at the numbers being announced, you're not going back. Matching supply and demand is a good idea. At the end of the day, I really will pay $50 for a little more legroom. Somebody else won't. I really don't need to check bags. I stopped trusting that a long time ago. So it doesn't really matter what you charge me for a checked bag. But for someone else it makes a big difference. It really is efficient to sell me what I'll pay for and not sell me what I don't value. That was a fundamental part of the inventory system--it had to be completely flexible.

It's easy to say, but very hard to do, because then when you make a change, that has to change on the Web site, on the mobile devices and in the legacy communication. And then you go talk to the legacy systems and oops, they can't handle that communication. Right now you have miscellaneous charge orders, etc. that are not quite there yet: "Our legacy provider can't handle this but we need to provide it to the customer. How do we bridge that right now?" You get situations like: "I paid for the better seat, but then I got the upgrade, but then I didn't get the upgrade because we changed equipment, so what happens to what I paid for the better seat?" You want to have that whole chain of interactions so you can do good customer service recovery, so the customer does not get mad at you. To handle that properly, you want to build the whole thing beginning to end to handle it. Right now, those back-end systems don't handle it. Now you have the opposite of seamless--an obvious seam or even a tear. If we weren't in the industry, we would be either amused or outraged by the failures. "Everyone can board who is in my Zirconium class, or if you are Precious class on somebody else." But they don't really have that communication worked out very well. "Are you really a Lufthansa Senator? Well, I don't really know because the protocol doesn't really support asking that question in an easy way."

So you drop the ball a little bit at the airport when you are ground handled. Or if you try to do through check-in, the domestic U.S. airline system says you have a reservation, but the foreign airline says, "That is nice for you, but my computer does not think you have a reservation. I am going to make you check in again." Airline check-in failure rates are in the 20 percent to 30 percent range. It is complexity that is just not handled properly. There are challenges.

Campbell: What is it about code sharing that is such a mess?

Wertheimer: To be Shakespearean about it, "Oh, what a tangled web we weave." You are pretending that it is your airline flight 9999 when really it's some other airline's flight 1234. You can maintain that fiction for a little bit as long as people don't dig too deep. The problem is that you, the marketing airline, have really solid protocols about what happens if it's raining, if the plane is late, etc. But what if it is not really your metal? Well, you send a message. Will the operating carrier do exactly what you would? There is very little chance of that. On the one hand, it is a very necessary, efficient, smart thing because it doesn't make sense for one company to fly airplanes all around the world; that doesn't have the right scaling properties. And it is nice for someone to think that the same logo is taking them on their entire trip. So those partnerships are good things. But it is a lot of complexity. From a pure science side, you have millions of lines of code in your organization that work exactly through this business process in this complicated case, and the other organization has something different that they do. You connect the two together and you pretend that it is one. It works often, but the seams show up. Interlining, too, has its challenges. It is a fundamental part of the industry and on the whole it is something we just live with, but it adds lots of complexity.

Dave Hilfman, Continental Airlines: You talked about airport systems. Are you also doing flight ops and payroll and all those types of things for Air Canada?

Wertheimer: We have a drawn a line and said we are only doing the passenger service part. It is talking about passengers and not about cargo and its talking about above the wing--all the commercial stuff--and we don't really go below the wing and talk about how you put fuel into that plane. We are starting to make little forays into some of those areas, but our focus is on the passenger-service side.

That said, there are some functions that traditionally live in those operational systems that want to be more on the commercial side. A standard example is, you walk out in the morning and it's raining, or you have a piece of equipment that isn't going to be available. Which flight gets affected? Part of the challenge is that the operational guys know the plane from the bottom. They know what it is, its tail number, the number of engines, but they don't know who is sitting in it or how much they're getting paid. I remember one person who was on the commercial side at a big airline in the U.S. would have to go out and yell at the operational people. They would see this plane leaving in the morning and say, "That plane isn't really full. It's going to our hub. There will be another one in a couple of hours, so that is the one that we can scratch." But [the commercial person would say], "You don't understand. Those are our most expensive, most loyal customers taking that morning flight to go connect to various places and we're selling them that seamless experience. You are now creating a huge tear in that seamless experience. That's not the flight you scratch. You scratch another flight." There are some functions that because airlines are very siloed probably need information across different silos, and need to be re-architected. One of the fun aspects for us in building a new system is you get to move these pieces around and give them visibility to each other.

Campbell: Regarding the components of the passenger service system that you have built, does it make any sense at all to think about a merchandizing solution that airlines can buy even if they use a different host system?

Wertheimer: There is a lot of modularity. But each becomes even more complicated because now instead of having a system you control all of, it's a solution with pieces here and pieces there. Pragmatically, that is where you end up. People say, "I don't have the budget to replace all of that now but I really have a problem over here and I have a budget to address that problem. Whether it is merchandizing, reaccommodation or other different modules, there certainly is a driver from the customer side. Sometimes we might say, "It really makes sense to put these three things together. It will cost you three times as much if we have to split them." Often we figure out quicker ways to put those three things together.

There also is the issue of customer communication. In the old days it was about who owned your PNR. You went into ABC Travel Agency and you bought on AB Carrier and you used the ABCD Underlying GDS, and you knew who owned your record. But now it is not clear who owns the system and who owns the record. Now, maybe my BlackBerry owns it. Or maybe my iPhone owns it. That means everybody else has to give me information about changes. Now that you have all these computers that are hooked together on all these networks, you have so much more computing power, memory and bandwidth than the core, official, Stalinesque system. Over time it is inevitable that some of those functions just migrate into that cloud of communication and computation that surrounds us.

Campbell: As far as the passenger service system, what happens to it if you don't find a launch client?

Wertheimer: We'll find one. We're persistent.

Campbell: There's only so many. Is it a North American product?

Wertheimer: No. We're talking to international airlines. The system is there now rather than saying to somebody, "We'd like you to be the first heart recipient of this new mechanical heart and why don't you hang around for five years while we build it." Now it exists. We just need you to lie down on the table and we'll go to work. There are things you can do to mitigate the process and come up with a smooth transition--which doesn't mean smooth. It just means that instead of one big bump there will be three or four smaller bumps--but some path where you can roll out pieces of it, and that is starting to happen. You can roll out pieces of a system and actually roll out a little more system than you might think you need, but you get the chance to wire up those new systems. So when we roll out an availability system, we get to hook up a whole bunch of the arteries and veins that you need for a reservations system. Same for the shopping system. There is another type of system that we are looking at rolling out, re-accommodation systems that are very clever about saying that if anything goes bump, can you quickly reaccommodate everybody and do the shopping very quickly for them in the new context that they are in? So when we get to those systems a little sooner, we can hook up the arteries and veins, and when you get that last piece in, there is less cutting you have to do.

Campbell: Some people who are new to the industry are amazed at how all the airline systems don't talk to each other. "How do they not know who I am?" How can the industry get past that, at a time when there isn't tons of money out there to reinvent all this stuff?

Wertheimer: That really does require something of a rewrite. The problem is not that they don't know who you are. It's that many different systems think they know who you are but those views are at different levels. When you use your device or go on a Web site, the CRS knows who you are, or the underlying mainframe airline res system knows who you are, and that representation is your name all in upper case letters in the Roman alphabet. You might have a first initial, or you might have a Mr. in front of it and basically that is your name. The system that was built 20 years later--the frequent flyer system--got smart and decided to give you a number. So now I can disambiguate a little more because I have a number. It's a really hard problem to figure out whether you are the same as that other person [with the same or a similar name]. There is a lot that goes into that on the computer science side. Giving you a unique number is OK. Your social security number in the old days was a good system until people starting worrying about security. The nice thing is, now you are carrying a device. In the world of electrical engineering, the world is much better-ordered than in the world of people. Every single device on the Internet when it is manufactured gets a unique code that identifies it. Luckily, now if we can't tell who you are, we know what that is. And now with biometrics coming out, when eyes or fingerprints are scanned, then we start to know who you are.

I predict that once systems can reliably tell who you are, a whole lot of typing will go away. You won't have to keep saying things like "I am Jay Campbell and here is my address. Let me tell you again. I am Jay Campbell and here is my address, blah blah blah." Once systems know who you are, then a whole lot of the stuff that people manage might be able go away. Because it won't be: "Now let's go get the virtual Brinks truck to drive over to ARC to get the ticket and drive over somewhere else to get the money." It'll be: "I know who you are. I know your bank account. OK, fine, we're done. We can do a payment." Or: "I know who you are and I know who you work for. Fine, I trust them, we can work a transaction." Or: "You got delayed but now I am reprocessing, and since I know who you are, no problem."

For a long time, it's been the case that if you carry one of those [mobile devices] around, it's the same as having a tracking device on your ear if you are a sheep and a biologist wants to study sheep migration. Essentially, we know exactly where you are, where you have been, and now I can take of you.

Campbell: In terms of who is going to pay for all of this, are you trying to say that the technology is less expensive, so that question of funding should be minimized over time?

Wertheimer: No. Technology is horribly expensive. There are two parts to technology. There is the stuff you manufacture. You think about it once and then you know how to stamp them out. That gets cheaper over time. And there are lines of code, and that is just horribly expensive and gets worse and worse. Software gets more expensive. These systems are complicated. To make them look simple is even more complicated. In terms of the paying for it, it is a tough question. For me, I just went and held up a bunch of investors and said, "Give me money," and they were happy to, and we have one product [shopping systems] that sort of pays for the investment in the other product [airline reservations systems]. But these are big investments, nine-figure investments to try to create new systems. It is just rough. Maybe in 20 or 30 years they'll be made in China, like mechanical things are now. I am not sure that will come very quickly because it's kind of hard, but it will come eventually, I suspect. Or you can get crazy high school students for free. That works for some things--if they work at all you are happy because you got them for free--but for something you rely on all the time, that still is a tough problem that still costs money.

Pamela Keenan Fritz, Egencia: Talking about the passenger service system that you have built, it sounds fascinating because you are talking about this personal experience. At the time of booking, you've worked out a whole load of personal data about me the traveler, and then you give me the right price and proposal on the day. What is the future evolution of that? Does it evolve into an auction over time, where you could continually be receiving communication with the carrier, where they could make you different offers?

Wertheimer: I think you have bits of that happening now. Air Canada had a concept that said, "There are certain things that you might buy at the airport. I will give you a discount if you buy them right now," and lock in the revenue. So a bag that might cost $50 at the airport maybe costs $40 if you buy it now. The idea about, by the way I know who you are, I know what you've bought and I have a fluid market going on ... for 10 years I have wanted someone to do that and to say, "Why only evaluate the most efficient market clearing point? At one point, I want to do it continually." Right now it happens in a sort of a painful way: "I am not going to give you a seat because I am still deciding who pays me more money. I am deciding if I can juggle them around and give them better seats and now I will give you a seat." I think you will have a more fluid way of doing that.

One of the things that is required to have a more fluid way of doing that is for people to get in touch with you. Because if the only interaction I have with you is at the time of purchase and when you check in--when I give you a piece of paper that says 4A--I don't have the option of easily putting you in 2A because I have to get hold of you. Once you know that you actually have this gadget with you and that I can reliably get in touch with you and I know where you are--or I know you are in the gate area, I know that you are in the airport at least, or I know that you are not in the airport--then once you have that connection and I know the real state of things, then the possibilities are tremendous. I would suspect the evolution is--at least based on my experience--you should be trying to sell me stuff all the time. To me, it's never the wrong time to try and give me an exit row or give me a little bit more legroom.

The only requirement for that is being able to reliably get in touch with you over time and also capturing the information that says that is a good thing to do. It is amazing how far the industry has gotten with the systems that underlie it because the systems that underlie it know so little about things and they have so few ways of getting at the data. There are basically two tables: by flight and by last name and PNR, and that's it.

Our system, which is perhaps another extreme, has 400 tables and things like, "Let's find all the other people that were booked by the same travel agent, who talks to the same admin, who controls that group of people in that same department, who are traveling together. Maybe they all would like to do something together like get upgraded."

So you really have to capture that richer information upfront because it is very hard to drive it later. You need to have the accurate information, you need to have a history to know what's an appealing offer versus spam--"stop bothering me, I don't want to buy that"--and you have to be able to get in touch with me. Once you put those together, you have the opportunity to be flexible enough to do a lot more business.

Holly Hegeman, PlaneBusiness Banter: This exuberance about how the airlines are going to be able to take this information and know where I am and what I am doing ... I don't want them to know where I am and what I am doing. Don't you think we are going to hit a point where it just gets to be too much and people start saying, "I don't want to do this?"

Wertheimer: It's all about selectivity. When you have two billion people being boarded each year, they are all individuals; they all have different preferences. Frankly, when the United kiosk says, "For $50 I will put you in a seat with six more inches of legroom," I want to kiss the kiosk. I am so happy. Different strokes for different folks--something that somebody likes versus something that somebody doesn't like. To me, what would be ideal is a system that basically says, "Every time I offered that guy the extra legroom, he bought it. I am going to find all the extra legroom that I can and try to sell it to him. Somebody else that has never taken me up on that ever, they are 5'4"; they are very happy. I am going to stop offering them the legroom." Whereas there are certain things that are offered to me like a pillow. I don't want a pillow. If I wanted a pillow I would buy a pillow. Thank you very much, I don't need your pillow. So, again the challenge is if you treat those two billion people getting on an airplane as if they want the same thing, that is just unlikely given our species. We have different things that we want. There is a great opportunity, especially when you know who the person is that you are talking to and you can learn very quickly, probably from one instance.

Ideally, you should be able to give feedback to these systems. When somebody gives you a flight alert ... the thing that you often have in most systems is, "Was this information helpful?" which is kind of annoying, but after awhile you realize it is really good because you can say, "No, I never want to hear from you again." You can learn that. Or if someone says, "Yes, you just saved my trip, do that all the time from now on." One of the challenges that we have is to develop enough information about people to actually treat them the way that they want to be treated and not treat them all the same. If you treat them all the same, then yes, you are going to annoy some fraction of the people."

Hegeman: You are talking about an expensive proposition for an industry that has no money. What I am concerned about is that the attempts that are ham-handed in certain respects, instead of being the perfect situation like you described. But I don't think we are going to see that on a grand scale for a long time.

Wertheimer: Perfection is a high goal. United is not perfect, but I do appreciate being offered that exit row. There is a learning process and it will be a number of years as we transition to knowing more and more about who we are, who our customers are and being more and more personalized about them. But I certainly think it is worth the effort because to me, that $50 that they got for upselling that seat at the last minute, that is kind of found money and that is money that I am very happy to part with. To me, that is a match made in heaven. But I just want to be clear: Communication that is not wanted is sort of like harassment. You really don't want to be harassed all the time. I absolutely agree. On the other hand, there are things that I do want to know about and so I think the challenge and the opportunity is to parse between those two.

Campbell: We have been talking about the consumer-facing apps and interfaces, which if they are effective they are very simple. Behind the scenes, there is an immense amount of complexity. Airlines over the years have been criticized and questioned for their yield management practices. Do you see any evolution toward simplicity behind the scenes, where it all begins?

Wertheimer: If you see simplicity, it's really a lie. The day Delta filed its SimpliFares was the largest ATPCo filing we had ever seen. The way this industry does simplification is that if you have a thousand fares in the market, and you want to simplify, publish another five. The reason is that all those were serving some function. One was the one for somebody's brother-in-law who did that deal and has a fare that just sits there. With legacy systems, you have legacy information. You have piles of fares there that no one knows what they are actually for, but they are afraid to get rid of them because maybe something will break. When they want to make a change, they just add more and more to it.

Simplification can be a good thing, but there can be simple and there can be so simple that it's not going to make any money. You really have to find that middle ground where it is not so complicated that it confuses customers but you still have the opportunity to try to make a profit. We'll be skirting that surface for a while.

What is interesting about all these new systems and tools is that now you essentially have two information economies. You've got the old, standard, officially sanctioned information--I know that sounds Stalinesque--where you go ask the Edifact message of some big mainframe and you get an answer. But then you have this newer economy where you go ask a friend or another system for the answer. You have the two of these co-existing and there is a lot of information. At the end of the day when you want to do a purchase or do something where you really care about high quality, you go to the source and you go through the rules. But when you are just surfing around looking for information, you'll go all over the place because you can't get all the information from the old Stalinesque systems. They don't have it and they are not fast enough. Yet the new systems are not blessed yet as: "You may rely on this and it will be good for you." You will see over time a shift that after a while the money will go to where the people are. After a while, you'll find ways to bless those transactions and those information exchanges on the lighter-weight system--with the appropriate protections--and you won't have to go back as often to the old Stalinesque system.