AI Insights: Generative AI across airline operations

Lufthansa Group established its Digital Hangar to focus on the customer experience just over a year ago. It's no surprise that the unit is looking at generative artificial intelligence technology as part of the brief.

As part of our ongoing series on how travel brands are using generative AI solutions from companies including OpenAI, Google and more, Digital Hangar's chief technology officer Christian Spannbauer shared his thoughts on the technology and its impact on the travel industry.

We began working with generative AI … at the beginning of this year with the first announcement and availability of OpenAI services. There’s a big difference to the experience in other areas of technology, it’s available right away so everyone can just open a session. No big tech setup is necessary. What we have to ensure is compliance with internal regulations, local laws and whatever rules there are ethically, culturally and so on. It typically takes effort and time to convince internal stakeholders because they want to ensure it’s used in the right way, that the data stays with us and that it’s not unintentionally getting out. It’s less a technology challenge than a cultural or a skills challenge.

Our current work with generative AI is focused on … multiple areas. Those areas are not surprising. I would expect most companies to be looking into them. It’s about content creation, it’s about any kind of conversation servicing aspect and automation when it comes to text-related complexities or challenges. We’re throwing in some documentation on our internal regulations, rules, handbooks, guidelines and standard operating procedures to train the model with and give our internal employees access to in a very straightforward way. There are a lot of ideas, if you start one, other functional departments get inspired, it’s easily democratized, which is good. That’s the difference and power compared to what we see with other innovation. 

The biggest challenge for us related to generative AI … is not a technology challenge. We’re collaborating closely with Microsoft and OpenAI so there is support and understanding, and we’re integrating our platforms very quickly. I don’t see difficulties, on a low level, with open data available in the large models coming from the internet. But when it comes to internal data, it needs a certain quality so that the model is delivering the right results. Hallucination is still an issue in the public models and even more challenging for the internal ones. You have to understand your data, its quality and whether it’s based on up-to-date data in order to make a decision. Is this the right or wrong answer and then train the model to bring back the right things. That’s the biggest challenge. Also, data is distributed in multiple systems, with not all of them directly accessible so you have to pull it together in a way that the model can handle or that we can train to handle.

For the travel industry overall, we see the most potential for generative AI to … it’s inspiration. For example, people want to go somewhere for golf, they find a place and it must be reachable by Lufthansa. That’s an obvious use case but not a differentiating one because it’s already available today. There are a lot of little and larger companies providing services like this so the hype for these more generic use cases will go down. Then it comes to topics which are really differentiating and based on our internal capabilities, for instance, the entire experience with Lufthansa. If I look at our app today it requires that I know what I want, but if I remove the app and use a generative AI based Copilot, it’s a different interaction model. We already have data about our customers, we know their preferences. We track events with our customers so we can make the interaction model we’re having today on our app into a personalized conversation like a concierge service. It can make life easier for our customers, it’s constantly learning and giving the right advice.

One year from now we expect to be using generative AI for … all of our functions to a certain extent. We have widely democratized the use of it, we have brought understanding of what can be done, and we see productivity in our daily business. There are efficiency gains for our developers. Already today I think 40% of Github source code is AI generated. Developers can supercharge their efficiency with these little tools. There’s also our administrative work. If I look at our meetings, for example, there is somebody inviting, someone doing the minutes, somebody following up. I think in the near future we will see this done mainly by Copilot. The technology will evolve so we will see even more conversion between speech, text, picture and video for use cases we are not thinking of today. It’s really hard to predict the next steps.