Julia Krauwer remembers well standing on stage at her employer, ABN AMRO, ten years ago. “I was an innovation manager back then,” she tells Patrick Grasza. “I explained to my colleagues what AI was and what we might be able to do with it. Among other things, I said at the time that AI was absolutely incapable of replacing human creativity. I’ve come to think a bit differently about that now. Back then, I also never would have imagined that I’d be asking educational questions to something like ChatGPT.”

Krauwer, now a Sector Banker for Technology, Media & Telecom at the bank, simply wants to say that it’s difficult to look into the future and make bold predictions. “Especially since late 2022, so much has happened in the development of Large Language Models, but also in robotics, for example—more than I had anticipated at the time. You won’t hear me say so quickly anymore that AI can’t do something or would never be able to.”

High Expectations
The rapid development of AI has significantly raised expectations, notes fellow panelist Patrick Grasza. He is Head of Data Management at Intergamma, the organization behind the DIY chains GAMMA and Karwei. Adoption often stands in the way between dream and reality, he notes: “AI is currently viewed too much as a magic solution. Organizations are fixated on the perfect end result, but they forget to invest in the foundation. At Intergamma, we use AI precisely to strengthen these foundations by strictly focusing on data quality. True innovation requires strategic patience and a willingness to build iteratively. That level of maturity is often still lacking in external organizations.”

On top of that, companies come up with solutions that the average user simply doesn’t need, Grasza continues: “You often see a Ferrari being built, even though the user can’t go faster than sixty kilometers per hour.” He calls it “innovating for the sake of innovating.” “Not looking at what it solves, but at what it can do. Yes, you can build an internal LLM to edit your emails, but that’s obviously a very expensive solution.”

Intergamma’s AI chat client
“Technology is a means, not an end,” Grasza continues. “That’s why we always start with a specific business challenge and only then use a rigorous business case to assess whether AI is actually the most cost-effective solution.” He cites a specific example from within Intergamma, where he serves as a bridge between data strategy and practice: “At the time, we were testing whether we could use an AI chat client to guide our suppliers through the data submission process. The question we asked ourselves then was: did we really need to do this with an AI chatbot, or would it be better to just set up a call center?” Intergamma developed a business case to assess the costs and impact. The result? “The costs were very close. Expanding our capacity to call suppliers manually turned out to be just as expensive as implementing the new technology. In the end, we opted for the new technology because it’s more future-proof in terms of scalability. Still, it’s always good to ask yourself every time whether adopting a new technology is actually the right path.”

Kodak moment
“The business case as it stands today could, of course, look very different in five years,” Krauwer notes. She believes it’s “potentially dangerous” to assume the technical status quo will remain unchanged: “At Kodak, a major player in the film industry, there was Steven Sasson. He invented the first digital camera. It wasn’t very useful—the device produced blurry images and was incredibly expensive—so he was laughed at. They decided not to pursue it further and to protect their own business.” It didn’t end well for Kodak.

Krauwer continues: “You can’t use today’s technology as the starting point for your strategy for the next ten years. You have to look ahead a bit. And that’s difficult. I talk a lot with companies in the tech sector, but also outside of it. I see it as our role as a bank to spark conversations about the future—to ask thought-provoking questions, even without having all the answers, to make organizations more resilient.”

“Everything that comes out of LLMs sounds reasonable.”

Safe Sandbox
“No one can predict exactly what technological developments the coming years will bring,” Grasza continues, “but the underlying foundation is, to me, indisputable. What’s clear is that data will continue to form the absolute foundation of our business operations in the coming decade. After all, you see that suppliers have more and more data, that consumers want more and more data, and that legislation and compliance also require more and more data. In addition, I see AI primarily as the technological lever to achieve that scalability.”

As Head of Data Management at Intergamma, Grasza sees AI governance playing a key role in the coming years: “Across the market, we’re still in the early stages of technology adoption, a phase in which we’re experimenting a great deal. On the one hand, that’s not a bad thing, because you learn through trial and error, and it helps you keep improving. On the other hand, I think we need to think very carefully about how we experiment, what data we share, where we store that data, and what risks are involved. At Intergamma, we therefore offer employees the opportunity to experiment in a kind of safe sandbox. We’re building an environment where people at various levels within the organization can gain experience with AI.” In that environment, people are allowed—no, they must—make mistakes without the consequences being disastrous: “Making mistakes is essential, as long as you organize it within controlled frameworks with clear safety nets,” says Grasza, emphasizing the importance of governance. In this way, Intergamma fosters innovation without jeopardizing business continuity.

“I think a culture centered on curiosity is extremely important,” Krauwer agrees: “That’s sometimes at odds with a culture where everything has to deliver immediate results.”

Delicious dish
In addition to a drive to experiment, data quality is of the utmost importance, as Grasza knows from the market and from conversations with colleagues: “When organizations build their first AI application—such as an internal chatbot to make knowledge more accessible—they quickly run into the same wall. Before you know it, the output turns out to be far too erratic. That immediately hits the nail on the head: an AI application is only as powerful as its underlying data architecture. If documentation isn’t properly structured, you’ll see that such a model immediately starts to ‘hallucinate.’” That teaches an important lesson, he says: “You can only make a delicious dish with good ingredients. Data quality and availability are the essential prerequisites for the success of any AI initiative.”

“Making mistakes within defined boundaries is essential.”

The Risk of Convincing Output
Krauwer sees a risk emerging here: “Whereas before you simply couldn’t do much with a good AI model fed with poor-quality data, we now, of course, have these very powerful LLMs. Everything that comes out of them sounds very sensible. Because the output is so convincing, I think the issue of data quality gets somewhat overshadowed. A demo can produce quite impressive results without actually building on the right knowledge and data.”

“Exactly, and that’s the biggest pitfall right now,” Grasza responds.
“The model masks the rotten foundation. That’s why at Intergamma, we adhere to the firm principle that good data is always a prerequisite for the successful application of AI.”

Fundamental Questions
ABN AMRO conducts extensive research into the use of technology by Dutch companies. Krauwer notes that they often associate AI with efficiency. “Of course, it’s good to use AI for that purpose. But it’s also important to think about your service offerings in the long term. What happens if AI can do the same things you do, or if interactions with end customers largely take place via chatbots? Will your product remain unique enough, or will you end up in a race to the bottom, with low prices just to make it onto that AI shortlist at all?”

The focus on efficiency—often driven by a fear of falling behind—pushes those kinds of questions to the side, Krauwer observes: “The fundamental issues end up at the bottom of the priority list. Because everything has to be done right now; everything has to deliver immediate results. Companies are looking to see if they can do things faster, find more customers, and so on. They don’t have the time—or the mental bandwidth—for strategic questions .”

“Maybe AI is the lever we can use to create that time?” suggests Grasza. Krauwer: “That’s a very good point— I love it. Actually, that’s exactly how you should look at it.”