Nationale-Nederlanden was an early adopter of AI, Tjerrie Smit tells his tablemate. Smit is Chief Analytics Officer at the NN Group, the publicly traded parent company of the insurer and financial services provider. “In early 2023, we launched our first major GenAI use case, in collaboration with ChatGPT. It was a call-logging application that automatically summarized call center conversations between agents and customers. We’d already been in contact with OpenAI, and we also had experience with machine learning. That allowed us to deploy it quickly. Technically, it went really well, but implementation turned out to be mainly a human challenge: people have to use the technology; they have to learn to work with this kind of AI. Everyone has to go through that same learning curve.”

RandstadGPT
“Very relatable,” responds Martin Woodward, Director of Global Legal & Global Responsible AI Officer at Randstad. It reminds him of a learning experience of his own: “One of our first true generative AI applications was our own RandstadGPT. It was back when more organizations were creating sandbox environments for chatbots. Nothing actually went wrong, but what you quickly realize with your own sandbox is that you can’t offer the same feature parity.” In other words: keeping up with the frontier AI labs—as Woodward refers to companies like OpenAI and Anthropic—is impossible. Yet employees do expect the latest innovations. “Why can’t RandstadGPT do what ChatGPT can?” they wondered, for example. Woodward: “That’s when the decision was made, and we fully committed to Gemini. We said goodbye to RandstadGPT fairly quickly. It just goes to show that you can’t keep up in terms of features and functionality if you develop it yourself.”

Scalability and Repeatability
Technology is advancing rapidly, offering companies new possibilities with AI practically every week. “Of course, you have to test new technologies in pilots first—you can’t do without them,” says Smit. “You could keep experimenting indefinitely, but at NN Group, we emphasize scalability and repeatability. AI can do a lot, but it’s not a magic wand. It often just involves hard work, a lot of trial and error, and continuous improvement—over and over again. Scalability and repeatability are so important because you can reuse successful solutions across the organisation.”

Governance framework
Smit takes over: “At Nationale-Nederlanden, we primarily try to apply AI to our core processes. If we can make these faster, better, and easier for customers, that’s where we’ll see the greatest benefits.” Woodward points out that this approach also carries risks: “By focusing on core processes, you’re choosing a path that many companies haven’t yet dared to take.” This brings Smit to the topic of data governance: “We use a governance framework based on the seven principles of trustworthy AI, as formulated by a European Union expert group. These principles were first set out in a report by an EU expert group, which later served as one of the foundations for the AI Act. They include robustness, bias, non-discrimination, human oversight, and transparency. These are all core values that must be closely monitored in every AI use case. If you do that, you can safely deploy AI in your core processes as well.” Woodward: “Exactly what you’re saying. You can indeed see all those principles reflected in the AI Act.”


Trust-Based Approach According to the Randstad legal expert, governance is often wrongly viewed as a brake on innovation: “It’s quite trendy to rail against all kinds of laws and regulations. It’s practically a meme. I think that when it comes to AI, we can actually learn from what wasn’t done when social media first emerged, even though we now see that it has both advantages and disadvantages. I think it’s good that lawmakers are taking this seriously, but data governance is also simply part of sound business management. It provides the foundation for trust, one of the core values instilled in us by our founder, Frits Goldsmeding. In our staffing agencies, human interactions fostered that relationship of trust, and we’re deeply committed to extending that into an AI-enabled world.”

AI for governance
Smit adds: “The financial sector is subject to oversight, so we’ve traditionally been heavily regulated. We’re used to working with rules and governance—you simply need them to do things safely. In my experience, people who complain about governance often do so not because of the rules themselves, but because of their ineffective implementation. That’s a challenge we need to tackle together.”

Woodward: “This is another area where AI can help us make progress. We’re successfully experimenting with GenAI-driven workflows. For example, we can partially pre-fill certain assessments, just as the tax authority does with tax return forms. Actually, we’re very enthusiastic about using AI for governance. It provides us with more insights and saves time. A human in the loop then only needs to verify the outcome. Of course, many aspects of our work ultimately require human judgment—not to determine legal compliance, but primarily for ethical review. This is and will remain particularly important when it comes to recruitment, selection, and HR.”

A diverse group of people
The Randstad executive is also a strong advocate for multidisciplinary teams. “To organizations that are starting with AI tomorrow, I would advise them to bring together a diverse group of people. One of the things that makes AI truly unique in the business world is that you need so many different types of knowledge, insight, expertise, and skills from people to make AI projects a success. These include people with your own background, people with an IT background, people with an information security background, as well as compliance specialists—and, of course, the people from the business side who can identify problems and inefficiencies. If you want to have even the slightest success with AI, you’ll have to give everyone a seat at the table.”

Start with the problem
Smit has a “don’t” and a “do” for companies getting started with AI. “Don’t get started just because you think you have to do something with AI. You need to use AI to solve your business’s number one problem. If you don’t have a problem, don’t go looking for one.” The Chief Analytics Officer at NN Group speaks from personal experience: “In 2015, when I started working on machine learning solutions, I went around the company trying to sell my exciting new technology. ‘Who wants to use this!?’ It didn’t work at all. You have to flip it around and start with the problem.” Then the “do”: “Make sure you have a solid ethical framework in place. Determining in advance what constitutes good AI and what doesn’t is very important. That prevents you from losing sight of that line later on due to enthusiasm. Always ask yourself whether something is possible, whether it's allowed, and whether you actually want to do it.”

Shiny object syndrome
“AI FOMO” seems to exist, Woodward responds: “People rush headlong after the first salesperson who dangles a tempting carrot in front of them. They suffer from shiny object syndrome. They fall for the latest gadget, the newest tool, and the smoothest sales pitch. But for a successful approach, you first need to identify the problem you want to solve. Then you assess whether AI can help with that—and in what way.”