AI lessons from the boardroom
KLM CIO Aart Slagt on why an AI model only has value if users understand and trust it
By Rutger Huizenga
AI is now on almost every agenda. Not as a gadget, but as a strategic question: what should you use it for, how do you scale it, and how do you prevent experiments from getting stuck in the pilot phase? In the interview series "AI lessons from the boardroom," we speak with executives and board members who have to make those choices in practice. In this edition: Aart Slagt, Executive Vice President Information Services & CIO at KLM Royal Dutch Airlines. “The time for experimentation alone is over. We are now focused on truly implementing AI at the core of our business.”
Aart Slagt has already had a full day by the time he sits down to talk about artificial intelligence. Meetings, coordination, decisions, operational questions: as CIO of KLM, he constantly moves between technology and practice. That also fits his profile. Slagt is not an executive of fashionable buzzwords, but someone who comes from operations and still looks at things through that lens. After ten years in operational roles and now five years as CIO, he knows that plans fall apart if they do not resonate with colleagues on the operational side.
“We, too, have gone through a phase of experimentation here.” It’s been a useful stage, says Slagt, “because it helped the business and IT understand each other better.” But as far as he is concerned, that phase is now over. “We can no longer let a thousand flowers bloom. We want to create focus: where does KLM want to deploy AI to the fullest, which processes does the company want to redesign from an AI-first perspective, and where do we deliberately choose not to apply it?”
Not every part of KLM lends itself to far-reaching automation, Slagt knows. Some things you simply want people to do. “Take cabin service on our aircraft. That cannot exist without human contact.” But in other areas, AI can make a real difference, and that is exactly where you want to accelerate it: in planning, optimization, administration, and support. A good example is the model KLM built together with KickstartAI to better predict how much catering is needed on board, based on the actual number of passengers who end up taking the flight. It sounds small, but the impact is significant. “On the one hand, you can serve passengers better with the catering that is really needed,” says Slagt. “On the other hand, you prevent a huge amount of food waste, because far less has to be thrown away. So it is a double win.”
Trust and explainability
Another example Slagt is proud of concerns a joint project on contingency fuel: the additional fuel an aircraft carries for emergencies and unexpected circumstances. It naturally requires a delicate balance. Safety always comes first, but carrying too much fuel is also unsustainable and costly. “If you need it, it has to be on board. But you also do not want to take too much, because it adds unnecessary weight.”
Together with KickstartAI, KLM built a model that takes a higher number of variables into account than the traditional approach. Think of load, passenger numbers, wind, weather systems, and conditions at the destination. It is a unique model, but “ultimately, the captain is responsible and decides how much fuel is taken on board. So it is important that you can explain how such a model works, which variables go into it, and why the advice comes out the way it does.”
Knowledge
And that, according to Slagt, is exactly where things often go wrong. “A model may look excellent on paper, but if people do not understand it, nothing changes on the work floor.” That is precisely why it is important for end users (in this case pilots as well as the pilots’ union,) to be involved early in the development of such an application. And it is equally important for executives to understand both the capabilities and the limitations of AI. Slagt says: “If you have to make decisions about investments, risks, and applications, you cannot follow this from a distance. You really have to immerse yourself in what AI can and cannot do.”
At KLM, that meant the leadership team did not get yet another presentation, but instead had to sit down behind a laptop themselves, in this case at Google’s office. Slagt: “The assignment was simple: build your own AI assistant for an advertising campaign. Or create an agent that scans aviation news every week and puts it in your inbox. Only when you do it yourself do you see what AI can deliver.”
KLM noticed the same thing when generative AI started advancing rapidly and employees wanted to start working with it. That is why KLM built Talia: a secure environment in which employees can safely work with models such as Gemini, Mistral, and ChatGPT. “It is important to keep building the bridge between the theory behind AI and its practical application. The colleagues that will work with it need to understand it, embrace it, and gain trust in it.”
The Dutch debate
But according to Slagt, the learning doesn't stop there. Anyone who really wants to make serious progress with AI also has to look outside their company.For that reason, KLM is active in KickstartAI, an initiative it co-founded with NS, Ahold, and ING. For Slagt, the value lies not only in the technology, but also in the way of collaborating: companies, sectors, and specialists coming together to solve real problems and put AI on the map in the Netherlands.
He calls KickstartAI, not without reason, “the coalition of the doing.” It is not a network of endless meetings and discussion groups, but a foundation that simply gets to work and delivers strong results.
And according to Slagt, that is exactly where the gain lies. “By joining forces with other companies, you bring in AI knowledge as well as a fresh perspective on your own processes. It is precisely that combination of data and AI specialists with people who sometimes come from completely different sectors that produces the best solutions. I would like to see more companies in the Netherlands benefit from that by joining KickstartAI.”
It is a pragmatic and open perspective that is still far from self-evident in the Netherlands, Slagt observes. “ We sometimes focus too much on the harm AI could cause,” he says. “And then we surround the technology with fences, rules, and restrictions.”
A shame, Slagt believes. Because AI can make companies more productive, organize processes more intelligently, and strengthen the Netherlands as a knowledge economy. At the same time, international developments are not standing still. Other countries are investing heavily in AI and rapidly building up knowledge and applications. “If we do not do it, someone else will,” he concludes. “Then we will be left behind. It would be a great shame if we let that happen.”
Three lessons from Aart Slagt for other organizations
1. Stop open-ended experimentation in time
Pilots and proof-of-concepts are necessary to explore possibilities and bring business and IT closer together. But at a certain point, an organization has to make choices. Where do you really want to deploy AI, what do you want to scale up, and where do you deliberately choose not to apply the technology?
2. Do not build a model for the technology, build it for the user
An application only creates value if the people who have to work with it understand it and trust it. Especially in regulated environments, AI must be explainable, align with professional responsibility, and be developed together with end users. That is why, for example, KLM involved pilots at a very early stage in the development process of its contingency fuel application.
3. Make AI a board-level issue, not a specialist silo
Invest in the knowledge level of the company’s leadership. Anyone who wants to steer investments, risks, and implementation must personally understand what the technology can and cannot do. That requires not only vision, but also hands-on experience.
