Aviation’s digital revolution 2.0 - What is the state of the art for airlines?

Transcript

Johnny Thorsen:So in the interest of operational efficiency, Marco and I decided to drop the introduction from him. So we have kicked 1 minute out and I'm trying to get back on track here. My name is Johnny Thorsen. I'm with a technology company called Serko. And today I have a really strong panel of tech people from the supplier and the services side who will talk with us about What are they doing in the new digital revolution 2.0, AI revolution? I started the morning counting how many times AI had been mentioned. I stopped at 50. So clearly the idea is to try and take a peek into the future and see where will we be, not 5 years from now, but 6 months. This is happening so fast. So let's get the panel introduced. Tamur, starting with you. Quick background, role, responsibility?

Tamur Goudarzi Pour:Yeah, Tamur Goudarzi Pour. I'm 26 years, I think, now with the company, Lufthansa Group, and I'm head of strategy.

Sara Walter de Freitas:Sara? I'm Sara. I'm director for digital and e-commerce at TAP Air Portugal. And by the way, welcome to Lisbon. And I've been— well, I'm still a newcomer. I've joined this industry, this amazing industry, only 4 years ago. And well, looking forward for this panel. Let's have some fun.

Johnny Thorsen:And you're relatively new to the industry compared to at least Tamur and I.

Sara Walter de Freitas:I try to be. The more I get unbiased by an industry, and I've been working in highly regulated industries and in digital for over 25 years now, and the more I stay like a newcomer, Vivian, over to you.

Vivian Dsouza:So my name is Vivian Dsouza, and I work with a company called Accommodations Plus International as the VP of Business Development for the APAC region. So I've spent the last decade primarily focusing on digitization of flight operations, but with API, my focus has been to digitize one specific part of the flight operations, which is the crew accommodation. and the disrupted passenger recovery from an accommodation management standpoint.

Johnny Thorsen:And it's pretty interesting, the company is literally called API, but it's 41 years old?

Vivian Dsouza:41 years old, yeah.

Johnny Thorsen:So before the term API didn't even mean anything. So you have a very hot name now, but you probably have a lot of explaining to do. So let's kick off with a kind of quick look at your own experience with AI technology. Give us an example of something you have tried with AI, either professionally or privately, where you got surprised and impressed.

Vivian Dsouza:Yeah, so, you know, I'm going to take a step back. Okay, there's been a lot of conversation of AI, and well deserved. You know, AI is a really powerful tool. There's so much that can be done. But I'm going to take a step back and just talk about digitizing of those functions which are still done in the traditional manual way or a semi-automated way. And I can give you dozens of examples of airlines that do this, but, you know, one of the hot names today, we started the day off with Air India, so I'm going to take one specific example of Air India among, you know, 100 airlines that we support. We took a process which was 100% people-driven. People would use emails, use fax, use telephone, use Excel sheets and PDFs to run the entire process. We took that, put in technology, obviously backed with automation, backed with solvers, backed with machine learning, and now powered by AI also in measures. And we took that process and digitized it up to 98%. Now, to give you a scale, that basically meant if there were 150 people focusing on this in the past, this has now been dialed back to a handful of people at a supervisory role only. Now, that's for For me, that's amazing. That's the Aviation 2.0 digital revolution that I would talk about.

Johnny Thorsen:Yeah. Like you said, you did not need AI to achieve that. It can help you turbocharge and deliver more.

Vivian Dsouza:Absolutely. Now, that being said, there is still AI in the background. There is still AI behind the scenes doing a lot of stuff which helps us get to that point in time. Again, it's still taking the traditional and making it next-gen.

Tamur Goudarzi Pour:Cool.

Johnny Thorsen:Sara, over to you.

Sara Walter de Freitas:If it's about the surprise, first of all, to manage the expectations, I'm not an AI expert. But the biggest surprise of all was when this whole hype around GenAI started. For me working in digital space for over 20 years, it was a surprise that suddenly everyone was speaking about something that in fact, has been used for decades. GenAI, yes, it's new, but AI has been there for many years. In the digital space, as you were saying, digitizing is the thing. We've been working with that for many different objectives. With this revolution, it's not only about having a futuristic solution, it's really having a strategic enabler. In this industry, it's really about how fast and responsively we will be able to transform the industry and enhance the experience, not only for the customers, but obviously for managing the disruption and for being more cost-efficient. Still, the surprise is the speed. Everything is fast in our world today. AI doesn't— it's not different from that. I remember that in TAP, we started building an agent as an experiment. We're still quite responsible and we do doubt about the scale and the speed in which we can deploy these solutions. 2 years ago when we started building an agent as an experiment, we had to bring in engineers and developers and it was all about hard code. 2 years after, you have just— you go on Google platform or Microsoft or whatever, and you have an agent. Easily, anyone can build what we were building 2 years ago with a lot of effort. It's really about the speed and the responsibility of delivering solutions at the speed responsibly.

Johnny Thorsen:I asked you before the session, TAP, you got hit hard today by the strike. Was AI involved in deciding which flight to cancel, in managing the whole process?

Sara Walter de Freitas:We wish we could say yes, but no. It's not because we don't want. Yes, we do have to be completely transparent. Of course, we do have an LLM that has been supporting and evolving within our operational teams to support this kind of situations. But is it still— is it already at the level that we need? No, for sure not. But we are on that journey. I think it's also not about where are we already, but how are we getting there. We're not an early adopter for sure, but we understood very rapidly that we needed to be on that track as fast as we could. Again, thinking very carefully on what should be the priority, how can we do it responsibly. The fact that the policies are still being— it's not even built, it's conceived. We're still trying to understand how all this can really be used. So we are being careful, but at the same time, we know that we have the responsibility of doing it as fast and well as we can.

Johnny Thorsen:Tamur, obviously you're a big group. You're changing strategy over the years multiple times. Where are you right now?

Tamur Goudarzi Pour:I think the colleagues already mentioned a few items, and I think we can probably now agree beyond the hype factor This is the fastest adoption of a technology in the history of humankind. Yeah, so that is something just by the speed of how it progresses now over the planet. This is unique.

Luis:Yeah.

Tamur Goudarzi Pour:So of course we have to be careful what's possible, what's not. But take on the— why is that so? It's so because simply the pressure comes mainly from the B2C side. So we have seen other technologies in the last year like blockchain and others, they have not penetrated as much. Simply also because the use case was not the wide adoption that we see now on the B2C side. Like everybody of us is using it. And I learned that now the first children age group, Generation Alpha, trusting the AI more than their parents.

Commentator:Yeah.

Tamur Goudarzi Pour:So it gives us food for thought. So, but from the commercial side, there's still a lot of limits of the applications we see though. A trend and a certain parallelity, but a certain trend of this. First it was digital first, then it was mobile first, then it became social first. Not so much social about anymore, it's more one-way street of consumption now. And now it becomes the kind of generative AI or agentic AI first in a new trend commercially. But when you see application It's very much early funnel of that travel experience where things are moving fastest because you have no need for deterministic and rather stochastic but creative processes. And that's why it works for us from that side. When you look at the applications further down the funnel of our value chain, then it gets more tricky also because we need more deterministic processes. And so far, and you asked for an example, I've not seen many, and the recent, I think, McKinsey study showed a lot of people dealing with it, very few have scalable applications.

Commentator:Yeah.

Tamur Goudarzi Pour:One, and I can mention one now, our first one that convinced me so far, is one on cybersecurity where we actively now actually manage to filter out attacks and manage attacks with AI, with the new generative AI. So that set of millions of attacks that we get per month, per per month, we can filter them down to 54,000 per month, of which 50,000 are handled by the generative AI now. That means only 4,000 now need to be tackled per month by our people. This is the first time I've seen really traction. And thankfully, we are, you know, in the end, free from any major mishap in the last time. Knock on wood. Hopefully, It stays like this, so this helps. So for me, that's the first case I've seen really scalable adoption in routine business. For the others, it's still in the emerging trends, and we can talk about what are the trends there, but that's the first example.

Vivian Dsouza:Yeah.

Johnny Thorsen:So just a few numbers to give people an idea of why this is happening so fast. As Tamur said, it's already the fastest evolving new technology. Last year, there was about $200 billion invested in the AI infrastructure. So that's the data centers, that's the framework where all the processing is happening. This year, we are on track to go above $400 billion. And when the year started, expectations were that it would be about $300 billion. So within one year, it's already gone up by a significant number. And nobody knows where we will be next year, but everyone agrees it'll be more. Just yesterday, Amazon and Microsoft jointly announced $50 billion of investment in India for new AI infrastructure. And when you start thinking about that and this amount of processing power that's coming online, it just tells us we have only scratched the surface. ChatGPT have 2.5 billion prompts per day right now, but they have also shared that 1,000 prompts cost $3 in processing cost. So you can kind of do the math. This is expensive tech and it comes with some problems on the power side, sustainability issues. So for an airline, and I'll jump to you, Tamur, this explosion probably coming in look-to-book, what, how will that impact your models? Are you ready for that or are you going to find new ways of controlling that?

Tamur Goudarzi Pour:Yeah, I mean, look-to-book is a very specific commercial topic. Now, often you get, of course, looks that you have to pay for instead of people booking then, and you pay for the looks. So to explain also for the people not familiar with that, but first of all, I think there are 2 major issues there before we come to look-to-book. It's the question of, first of all, trust, how we secure that and make it something from the governance side, a thing that we can make safe and secure. And the second topic we need to tackle is autonomy. So when the more the thing become autonomous, the more we have to make sure, 'cause we don't know what's in the black box.

Johnny Thorsen:Yeah.

Tamur Goudarzi Pour:Or stochastic black box at the moment. We have to handle that. These 2 topics will be key for finding guardrails that we will need to identify also as an industry. So we need, and you know, maybe pharma industry is much more regulated on— that's the consumer side— while we are very regulated on many topics but not on this topic, that more on the commercial side. We need to find a way as an industry to have minimum rules of the game being established. And this, I think, will also be a question of us. Now I'm talking about the airlines in the IATA context, and we have to address it fairly soon because the rules will be otherwise made by other people. And we cannot be the substitute for hyperscalers. They will do what they do, but we need to find a way to organize ourselves.

Johnny Thorsen:Yeah.

Tamur Goudarzi Pour:So that is the— after NDC and offers and orders, the 3rd big topic we need to tackle now. And this then— sorry, coming to your look-to-book ratio question— also refers to finding guardrails for this topic. And you have different options now. The minimum option you have is that you say, I identify an industry repository that will allow us to find minimum rules how that an agent identifies itself, how you find rogue agents, etc. That's the minimum, let's say, guardrails rules that we need to find. You can go further on, let's say, on the offer side. And then it's the question, do you find a way to manage the look-to-book ratio question. You can also go on the product side, certain product features. And ultimately you could, and that goes further, also go on the question of will there be a, let's say, protocol element that we would run together, for example, MCP server or agent-to-agent, whatever. So these are the levers you can play, and we need to decide how far we want to go.

Sara Walter de Freitas:Yeah.

Tamur Goudarzi Pour:Is it Wild West? Or do we do something about it? And my plea is let's do minimum the first stage. On the second one, either there will be something that will emerge evolutionary, or we find also a way to govern the look-to-book question, which is of course the question— always the 3 questions. It's the question of customer experience, the question of cost, and the question of being in control. So 3 C's. We need to manage this. I think there will be another initiative coming for that, so we cannot just sit still. As I said, this is very important, very urgent.

Johnny Thorsen:Sara, jumping to you and TAP.

Sara Walter de Freitas:The topic of the loop-to-book is quite a straightforward topic for this industry. Again, as still a newcomer, I find quite strange that, for example, we call modern retailing to something that was modern 30 years ago, and that we pay for our own inventory, and that we pay for the searches on that inventory. I would really like to see with this change around AI a deeper change in the industry. In the model. I know that I'm— well, I'm probably challenging something that it's impossible to change, but I'm a believer and I think that nothing is impossible until you do it. Someone already said this. I completely agree with what you said, but still, I think that that discussion needs— should, not needs. I don't want to be presumptuous in stating the rules, but we would need as an industry to join forces and really—

Johnny Thorsen:Yes.

Sara Walter de Freitas:Play this game differently. We don't need to own the game, to establish the rules by ourselves, of course, because then we would change the game around. But the model needs to change because otherwise AI will be just another layer of a huge cost for the industry that at the end of the day will be on the customer side. to pay for. That's my perspective.

Johnny Thorsen:Vivian, you are operating on the hospitality side of the industry and you're actually critical for the airlines to keep the operations going. Are you seeing any pressure from airlines for modernizing or changing the service stack?

Vivian Dsouza:Absolutely. To give you context, we don't book flights. We don't help booking flight tickets, neither do we do that. But what we do is on any given day, probably a little bit lesser in Lisbon today given the situation around, but on any given day, we are booking about 65,000 rooms on behalf of 100 airlines in over 5,000 cities. Now logistically, that's a nightmare if you look at it from a traditional lens because you've got delays, cancellations, swaps, the name changes of the crew at the last minute. And we do this entirely with less than 500 people who typically deal with last-minute things like emergency situations where you just can't rely on technology. Now, that being said, the airlines are strongly pushing for increased automation because of the other problem, which is lack of manpower. Now, If an airline can use that limited manpower in different places, they are more than happy to support automation. And that's where we come in. And that's exactly where our drive for pushing AI comes in. Now, that being said, and to Tamur's point, 2 of the things that we identified early on was one, security, and the other one was putting guardrails. And the reason for this is because if we screw up, The next flight isn't going out. Your passengers aren't reaching their destination. So it's operations critical, and we treat it with that level of severity where we need to be extremely measured in how much and how fast we adopt AI, but not to the point that we hold ourselves back from really exploring what the potential of that is.

Johnny Thorsen:So switching gear a little, AI is new technology. That means new skills. You probably don't have the expertise inside the house. As an airline, are you able to get the talent? Are you finding partners externally? Are you trying to build the kind of expertise internally?

Sara Walter de Freitas:Yeah. So in TAP, we have, as I said, we're on that journey. We are partnering already for over a year with the main drivers from a technology perspective worldwide to really start, first of all, understanding what this transformation really means. How can we drive this transformation inside out? Because it is about technology, but it's first and foremost about data quality, about culture, about sustainability, about how responsibly we can work with this. We need to start by changing the way that people interact with technology because the adoption with AI can be much faster. We need to start evangelizing for the adoption and then going out with it. To find the talent, first of all, we are partnering with with major companies for these. We are planning to have our own area supporting this AI transformation. In fact, we already created it, and that's critical. But there's still a long, long way ahead. It's not like the upskilling for the AI adoption in a responsible way is as simple as hiring for a commercial area or even for a traditional IT area because everything is still being discovered. How can we cope up with this transformation and at the same time find the right talent and in parallel upskill everyone? Because otherwise, it will be a huge internal disruption between These people that already know, that are prepared, that are setting the tone, and everyone else that will be left behind. We cannot do that.

Johnny Thorsen:Tamur, how are you managing that in the Lufthansa Group?

Tamur Goudarzi Pour:It's a tricky one. I think what helps us in our industry are we are technology-driven companies normally, so that helps as a starting point. Of course, there's always the question, where you take existing talent and where you bring somebody in either as a partner or you try to bring talent in. And we're traditionally not very good at that. So that's something where I would say the social biotope of certain companies makes it not so easy to bring somebody from outside in. Also, when we work together with some of the startups, our track record is not great on that one because we often have this closed legacy systems that's not easy to penetrate. That's why offers and others should break that. So that's a bit of self-criticism that we hopefully can change also. But it's very important also that we now come to the next stage of the operating model of AI. So, so far it was very much about experimentation and it's still important, but we need now to get the next stage of get a more concerted organization of those initiatives if we want to really scale them. As I mentioned, one example, others not yet there. That means for us a stronger collaboration of those business owners with AI competencies together with the digital integrators, together with the AI experts that come for each topic now, and the digital engine. This needs to be concerted. And we should— and that's the danger— not have too many AI platforms now. So you either work directly with one of the hyperscalers, But often that is not enough. You probably need, for certain topics, you need also support also from outside companies, and I have to decide who can bring me that additional value that fits into my future picture. And there, be careful of those who just want to close the holes in the road rather than to build the road with you, uh, in the direction of the road. Yeah, so often we see deficiencies and we just fill the holes, and that's often for the startups, unfortunately, B2C-driven, often the case. So, but if you find the right partner there, I think that can help. And what we don't want, because we are a little bit burnt as an industry with monopolists, another intermediary layer that you're completely dependent on. And we made our mistakes there too. So I think that is now the question of how much own resources, how much you organize your operating model, how much you work directly with the hyperscalers, and where you bring special knowledge in. as partners or even in integrated joint venture of other companies. The last sentence I want to say on that is one key rule should be that everything you invest now, every decision you take, do this AI-first check according to the rules you've set, hopefully in your company. But do the check. Ask everybody before you make the decision, make a check. If that is the mindset that you have, I think that's so urgent in these days. Then it's already a big step forward if everybody does that, because often people are not really that aware while they're in their routine. So this AI-first check, I think, seems simple, but it's often forgotten because also people don't know they need support for that. So that would be my take for that. But it's still a journey and we see where we're going.

Johnny Thorsen:And of course, it's not like airlines are having a lot of unused IT resources, right? We are, we're 15 years into the NDC journey. We are in the early stage of offers and orders. Is this going to create a conflict on the resource side? Are you going to add new or are you going to stop things from happening and redirect resources?

Tamur Goudarzi Pour:Well, offers and orders, as I said, the second initiative after NDC, better organized in the industry than the one before, not perfectly. I think that we are on a better way now for having a modular architecture. Very important is for the pure order, it's a deterministic process. The AI only can help you to a certain degree on that side, maybe for the bridges between the old and new world, maybe for orchestration of the modules. But as long as we don't have yet a deterministic way of using the generative AI for it, and there's some like neurosymbolism and other attempts now to bring linear programming with generative AI together, we'll see where that leads us. But you need to have that reform that we all have to do for the next 5 to 10 years in any case. In any case, and they will be supported by AI. But then you have, of course, on the office side, much more potential.

Johnny Thorsen:Yeah.

Tamur Goudarzi Pour:Everything that's driven from revenue management, network planning, all those disciplines, they will be now AI enhanced for sure, fairly quickly, I think. We're not talking about 5 years, we're rather talking about the next 2 to 3 years. That's happening quickly on any side. Operational, you mentioned it, there we don't have yet the holy grail for the overall orchestration. We'll see. But the offers and orders still needs that reform of the fundamental backbone from the '60s. That has to continue irrespective of the other trends coming. But you just have to check where you can apply new technologies. And the more they become deterministic, the more you can integrate it deeper into the funnel, the backbone.

Johnny Thorsen:Yeah.

Tamur Goudarzi Pour:So that's a bit the thinking about it.

Johnny Thorsen:Sara, on the, on the TAP front, same view or?

Sara Walter de Freitas:Well, we can't afford to stop. So even though we would love to have the ability to now focus on these big or bigger transformations, the fact is just— is that we have just— we are now close to ending our process of full adoption of MDC, which is a major achievement for us. We did it in a quite accelerated mode. We are going through this pathway of digitally transform the company in many different areas. We are already on the journey of offer and order. We are like 25% of that long, painful journey.

Johnny Thorsen:Sure.

Sara Walter de Freitas:And at the same time, we have to cope up with this. So we are still trying to understand how we can manage these different streams of acceleration to modernize the company. We have been quite successful in these, so we also don't want to risk jeopardizing what we have already achieved because now we have this AI ahead of us and we need to adopt it so that we also don't lose the train. It's really a matter of getting the right balance Without going crazy, if I may say. We're being cautious. Again, we don't want to be— we are not for sure an early adopter, but we don't want to be the late adopters that when we get there, everything already changed so much that we are again on the legacy mode. It's about the balance. When it comes to resources, We also need to understand how we can use not only the partnerships and to the point that you were mentioning before, the hyperscalers, just a side note on this. The hyperscalers of AI are also at a very different level of maturity. If you look into the solutions with Microsoft, Microsoft or Google, you have different pros and cons and different levels of maturity. It's not an easy decision. If we could give a recommendation based on what we already learned as a company in what we've been discovering with these big partners, it's that you need to find in each one Where do you fit? What really can make a difference for you? Not only for now, as you also were mentioning, it's not about you have a hole in the road and you just want to fix it now. No, we really need to build the road.

Johnny Thorsen:This is foundational.

Sara Walter de Freitas:Yes, it's really foundational. We cannot do the same mistakes that were done in the past with other technological adoptions. It's about the balance. It's also about the size of the company. It's not comparable, the strategy that a Lufthansa Group can take on this versus a TAP strategy. You also need to consider that. What really makes a difference to your operation, to your business, to your customers, and what is really feasible, because the shiny thing, yes, it's there and it's shiny the same way for all of us. But then what really is feasible and can really support you on the long term, it's different.

Johnny Thorsen:One of the challenges with the new AI solutions is the so-called hallucination problem. where you will literally get something that's so obviously wrong. And how are you going to manage that in your development of services? Are you going to have hallucination check experts, or are you going to have other logical processes built into the flow? Have you looked at that, Vivian, in your environment? Yeah.

Vivian Dsouza:So as a technology company, it's an imperative for us to stay ahead of the curve. And again, as one that has gone from manual to digital to mobile first and now to AI first, one of the first realizations we had is you can't treat this transition to AI the same way every other transition happened. A couple of weeks ago, we did probably the biggest step towards realizing this, which is we brought in a Chief Transformation Officer who now treats AI as a separate piece of the puzzle without mixing it with what was done. Now, that's how we would compartmentalize, put guardrails on it, but also make logical decisions without getting emotionally pulled in with what was already built. I think that's going to be for, at least from the perspective that we— I'm coming from, that's going to be a big driver on bringing a fresh perspective to to, again, a traditionally set way of doing things, but at the same time not push it too far ahead that, again, hallucination. That is probably one of the biggest fears that we have because when we started toying around with AI and then as it got sophisticated, we realized that AI started to go crazy. You don't want on an operational critical system a crazy person running your airline, right? So absolutely, and that's why we decided let's compartmentalize this division and run it under the overall digital umbrella.

Johnny Thorsen:Any thoughts there on— Tamur?

Tamur Goudarzi Pour:Yeah, maybe to add on this. Well, first of all, there are some processes that need deterministic answers, and there you cannot apply yet. As I mentioned, there might be new technologies coming by merging those technologies, but when you look at the Destination you mentioned. How was it when we moved from the landline to the mobile phone?

Sara Walter de Freitas:Yeah.

Tamur Goudarzi Pour:We accepted 98% functioning mobile phone reachability. 2% was fine that we couldn't find, you know, I didn't get my reception. I was still fine. I still have big trust in mobile phones. Still today we don't have 100%. So we are probably in that technology of the gentrification and Generative AI, probably somewhere in the 80s now.

Johnny Thorsen:Yeah.

Tamur Goudarzi Pour:So we need to go higher even for those more creative or more stochastic kind of solutions that we're looking for. But that will come because the technology will improve by the day and the ratio will go up. And as soon as we're somewhere in the 90s, the trust level is high enough that it can be applied. And this will come in any case, I think. Because the technology is actually developing in this direction. So for that, I have no fears. It's rather that we have to make sure that whatever the autonomous decision is, then that it takes, as I said in the beginning, is guardrailed. That's much more important in that sense. So it's not so much the hallucination, but rather the autonomy that we have to make sure that we don't go into the wrong direction. So in that sense, I'm fairly confident that we go in the right direction as long as we guardrail Sara?

Sara Walter de Freitas:Well, in this highly disrupted world, many occasions hallucinating, I think that we have— and I even heard our CEO mentioning this a couple of days ago— we do have to continue building on our customers' trust. We're very cautious with scaling up or even— well, one thing is experimentation, and we're at that stage. When we think about scaling and deploying solutions that have still that high risk, and as I agree with that analogy of it's like in the '80s with many other technologies, or probably it's now at the '90s level. We want to be very cautious in that. We prefer to take that approach rather than risking more. Maybe we should, but we feel that we cannot afford to do that because we have a lot to build on our customers' trust. Still, one thing that is critical is to bring on board those same customers working closer to the airline and co-developing— not co-developing, but co-creating, co-thinking about these solutions, really helping us also in understanding not only what are their needs, but mainly their fears. One of these days, I attended a talk about AI with some of the gurus out there. in AI, and one of them was saying, and he proved mathematically, that 40% of what Copilot or any AI agent in our days delivers to us, to any end user, is only 40% accurate. 40% accurate.

Johnny Thorsen:Yes.

Sara Walter de Freitas:Guess what? We all use it. It's at the same time scary, To see that even though it's still so inaccurate, the adoption is on the run. But as a company, we want to have a more cautious approach on that. To your specific question about how will we tackle, will we test until we get it to the hallucination stage and then fix it, or we will have a lot of agents working on the hallucination of the hallucinator? We still don't know. We're finding out.

Johnny Thorsen:No.

Sara Walter de Freitas:But it's a topic. Just to add a side note on this, I love the fact that you mentioned cybersecurity, and I want to know more about that because it's, for me at least, it's even more a topic when we see what is happening, what has happened, recently with some of the bigger tech companies in the world. It's really about one single connection and everything drops. We all stop. The world stops. Cybersecurity together with AI, it can be so, so powerful in a good way or in a very, very bad way. I think that industry really needs to look into that.

Johnny Thorsen:And I think that leads to kind of— we're almost out of time here, but running an airline used to be about understanding the hardware, the physical business operation. But now the dependency on the technology layer has become so much bigger. Tamur, you shared how many systems are involved in, in the Lufthansa Group.

Tamur Goudarzi Pour:Well, we have the discussion about offers and orders, and I think we know it's a 5 to 10 year exercise simply because there are more than 200 systems attached to it.

Johnny Thorsen:Yep.

Tamur Goudarzi Pour:And we need to make sure that step by step we exchange that with AI or without AI. So it's a fairly complex endeavor. So we have to start carefully. We have to separate, we have to segment, we have to modularize and limit the risk of what we do, of course. But there's no way around it. We will actually tackle every of these modules and see what has to be changed there. Also with the support of the technology of AI, generative AI, and the assistance and the orchestration that will come up very soon of agents, where agents speak to agents, and probably more than we just give information. For example, when somebody reaches our server, we will have an own agent that will negotiate with the other agent according to a certain protocol. That will come more and more, but we need also internally orchestrate it. So that's a fairly complex topic, and we have to treat carefully, but it's going to come in any case, and it gives us great chances now going forward. It feels like, you know, we are departing to the West in those days in the US when they treaded for a new country, for new land. I think that's totally exciting. So I'm fairly positive despite also the caveats I think we all have to take care of.

Johnny Thorsen:And I think ending, how will you manage data, right? Because if data goes into these models, you lose control of where they go unless you have a a private model running? Is that something that's been decided?

Vivian Dsouza:So I'll give you an example of a data breach and, you know, what happened. So on the 1st of November, we actually went live at one of the largest airlines in the world. A week before this, and they were moving from a legacy tech to a new, to our tech. A week before this, there was a data breach and the legacy tech went out. It was sheer luck of the timing where because we already had ours in the final stages of cutover, The operation went smooth. Now think about the alternative scenario. Talking about 45,000 crew members who could have been stranded, blackout.

Johnny Thorsen:Mm-hmm.

Vivian Dsouza:That's the criticality of having that security in there. And talking about guardrails, just to give you a small example of, you know, one of the things that we realized early on, we built a tool that would be able to predict what accommodation utilization would look like in a disruption scenario based on past utilization and global trends. Very early on, we saw hints of the hallucination of AI. We deliberately decided to guardrail it to only cap the predictable accuracy to 80%, fully knowing that it was not trustworthy yet. Probably in a couple of months' time or years' time, the technology could improve and we could get to 100% or at least a 99.9% accuracy.

Johnny Thorsen:On that note, thank you very much for sharing your thoughts and the kind of the plans that are in place right now. Thank you very much to the panel.

Sara Walter de Freitas:That was great.

Commentator:Okay, don't go, don't go, don't go just yet because we've got the responsibility to bridge you over next 5 minutes. Guess what? I actually recorded— don't you guys go either because there's some interesting stuff coming up. So I recorded the whole speech and I asked AI what AI thought. And I just want to tell you what some of the things that AI said. I said, so on the speaker who proudly digitized 98% of a process, 150 people reduced to a handful thanks to technology proved that airlines do believe in miracles, just not necessarily on on-time performance. Wow. Okay. On the speaker surprised everyone suddenly discovered AI, He or she reminded us that the goal is enhancing customer experience, not populistic solutions, a noble sentiment in an industry whose apps still forget your password. Okay, all right, okay, this person, this AI is quite intelligent. So on the engineer who said things were done manually 3 years ago, but now we can do it in Google or Microsoft, my favorite line, AI's favorite line, is they can handle strikes with AI. Honestly, neither can management. That's really insulting. On the cybersecurity example, filtering millions of attacks to 50,000. It's good to know hackers have the same punctuality as a storm in Heathrow. Okay. All right. On the speaker quoting investment numbers in the hundreds of billions, the speaker said AI is expensive. Yeah, sure. But compared to the cost of a single aircraft, it's basically pocket lint. It goes on and on. Just so that you know that AI can also— has a sense of humor. And it could go on. Having said that, there's been some really interesting main messages that came out of this session. Go for it.

Luis:Cool. Thank you, Alex. So I'm Luis from Amazon. We've been using AI for over 25 years, and I have a couple of insights for you. One of them is, you know, what limits AI adoption is not AI itself or tech. Actually, you're using it so much more than we all think. It's about data. It's about culture, it's about org. Okay. And just to give you examples that we're already using it, you just don't realize it so much, is that you already see— and I love the title of The Quiet Revolution for this, for this, for this panel, because if you think about it, in customer experience, airlines are already using virtual call centers, deflecting 30% of the calls, automating, voice activating, a lot of self-service. They are using 360 view of the customer to detect customer events and inject them into their customer profiles across the entire journey. They are doing what-if scenarios on planning and scheduling, not on aircraft utilization, but straight on P&L. That's fantastic. Okay. And if you think about operations, airlines are already using really, really well AI. Some of us, some work with us, that are here in the room to actually do IROPS in a recovery that is so much faster, to actually unlock the data that is in the systems that are actually decades old, to be able to optimise, to go total revenue management, to actually influence directly the biggest assets, which is aircraft turnaround, and actually to wrap up, touch the biggest and single most important metric that was mentioned at the beginning of today. Which is ROIC, return on capital invested, which is still below the cost of capital. But not for long. AI will help us all solve that, and we are using it so much more than we think already. Thank you.

Commentator:Fantastic. We close with 3 thoughts. One, us airlines, we're stuck between ambition and capability. We've been speaking about that pretty much all day. Second, the real bottleneck isn't AI, it's data quality, culture. Organizational design. I think we heard that loud and clear. And finally, trust, trust, internal, operational, regulatory, and customer-facing trust will define the pace of adoption. Thank you.

Copyright policy: All transcripts on this site are the copyright of CAPA - Centre for Aviation. Our reproduction policy is as follows: you may quote up to 400 words of any transcript on the condition that you attribute the transcript to CAPA - Centre for Aviation and link to the original video page. All other use is prohibited. While we aim for 100% accuracy in the transcript, there may be some minor transcribing errors.