AI in the Fight Against Insurance Fraud

“AI can enhance human judgment, but it cannot replace it,” says Professor Ana Isabel Barros, keynote speaker at the “De Boef de Baas” event on March 26. She shares her insights on the democratization of AI. On the risks: input, trust in output, and loss of skills. And on the question of how insurance specialists can effectively use AI to tackle fraud and (cyber)crime.

This article was published earlier by Verbond van Verzekeraars. Text: Ellen Jonges | Photography: Ivar Pel

About Barros

Ana Isabel Barros is a researcher and expert in the fields of intelligence, data science, and AI. She is a endowed professor at the Jheronimus Academy of Data Science (JADS), a lecturer at the Police Academy, and has been a principal scientist at TNO since 1997. She works on intelligence, complex system models, and operational analyses for defense and security.

The Red Card

She begins her session in the Verbond auditorium with the green and red card exercise. Who uses AI on a daily basis? It’s one of the questions Barros asks the attendees of the sold-out event. The room turns green. Or not quite. A single red card appears. “On days with a lot of meetings, I don’t use it,” replies the person holding the red card. But is that really the case? Because according to Barros, AI is more than ChatGPT and Copilot. “It’s the scan you use to unlock your phone. The algorithm that determines what you see on social media. The suggestions for similar shows on streaming services. It’s the robot vacuum that keeps the floor in your home clean. And… AI is also the parking assist in your car.”

The Democratization of AI

“This means AI is present everywhere in our daily lives,” Barros emphasizes. Often unconsciously, through the technical features in phones, household appliances, and cars. And also consciously, in the form of AI tools that people are using more and more frequently, both in their personal lives and at work. “AI is thus becoming increasingly democratized, which means it is no longer just the domain of AI specialists, but belongs to everyone.”
50% of property and casualty insurers use AI

This democratization is also evident in the fact that more and more companies are adopting AI. “Its adoption is rising rapidly,” says Barros. In her presentation, she refers to a CBS study that examined, among other things, how many companies use at least one of seven selected AI technologies. In 2021, that figure was 13.1%. By 2024, that percentage rose to 22.7%. The same study shows which sectors are leading the way: Information and Communication (58%), Professional Business Services (39.8%), and Financial Services (37.4%).

It also highlights figures from the European regulator EIOPA. The Report on the digitalization of the European insurance sector shows that 50% of property and casualty insurers use AI. Among life insurers, the percentage is slightly lower: 24%. Furthermore, an additional 30% to 39% of insurers expect to start using AI within three years.

Opportunities of Agentic AI

Companies are increasingly experimenting with and utilizing Generative AI, a form of AI that responds to a question or instruction—known as a prompt. And with the emergence of Agentic AI, a system that not only provides answers but also performs tasks on its own, new possibilities are opening up across a wide range of fields. Barros: “AI can assist with triage in straight-through processing, where incoming documents and transactions are processed automatically. But it also offers opportunities for efficiently onboarding and training employees. At JADS, we are currently conducting joint studies with banks and the technology industry to explore the potential of Agentic AI for knowledge retention.”

The main focus of the De Boef de Baas event is on the potential of AI to combat insurance fraud. Barros: “AI can support investigations into fraudulent claims by identifying patterns and inconsistencies faster than claims adjusters and fraud specialists. Not just when it comes to a single claim, but especially across multiple claims or even across organizations. Especially as collaboration within the insurance sector continues to grow.”

Identifying fraud patterns more quickly

Barros returns to the topic of collaboration among supply chain partners several times. “It’s good that the Association advocates for this in its new ‘Resilient and Vigilant’ vision, because one thing is certain,” she warns, “criminals are very smart, highly adaptive, and have the same access to AI as professionals. It’s also a huge source of inspiration for them, which means they can quickly scale up and adapt their knowledge and methods. Moreover, they aren’t bound by the AI Act and the Ethical Framework for Data-Driven Applications. They just want to make a lot of money.”

She continues: “Criminals also know, by the way, that companies don’t share much data yet and operate in silos. They take advantage of this by submitting a particular claim not to just one, but to a whole bunch of companies at once, which still causes a lot of damage. Through information exchange between supply chain partners in the sector, fraud patterns become visible much faster. And that’s where AI can help.”

Risks of AI: Input, Trust, and Skills

When implementing AI systems such as Agentic AI, insurers must be mindful of 1) the input, 2) trust in the output, and 3) the loss of skills.

“Machine learning-based systems are trained using data. That data may be incomplete and contain biases,” Barros emphasizes. “Additionally, AI systems are created by people, who have human assumptions and biases. The system easily adopts that information.”

Another point she wants to convey to her audience is that people tend to favor suggestions from automated decision-making systems. “We tend to think unconsciously: it was created by smart people, so it must be good. That’s dangerous, because AI can hallucinate and make things up on the spot. In the legal system, this has already led to significant problems. So be critical of the output and check the results carefully,” is her advice.

The third risk Barros identifies is the impact of AI on people’s skills. She cites a quote from Confucius: “I hear it, I forget it. I see it, I remember it. I do it, I understand it.” “So if AI does something for you, you don’t learn it. Take the calculator as an example. How well can you still do mental math? Scientific research shows that certain parts of your brain are no longer activated when using AI. Put simply, it actually means that it makes people lazy.”

Expertise in insurance fraud is and remains crucial

To conclude his keynote, Barros addresses the question of how AI can be effectively deployed in the fight against insurance fraud, despite the risks. “From all my years in law enforcement and the military, I know that experts truly have a knack for sensing when something is off. There’s always a grain of truth in that, so we need to keep leveraging it. This applies to human knowledge and experience regarding insurance fraud as well.”

Due to the risks associated with AI systems, Barros believes AI is particularly valuable when in the hands of experts. They possess the knowledge, context, and critical judgment needed to properly assess AI’s outputs and refine them where necessary. Those lacking this expertise run the risk of reinforcing ignorance and biases. “AI can enhance human judgment, but it cannot replace it,” is therefore her position.

Yet in practice, she observes that some companies are trying to bring in as many AI specialists as possible. And that they want all employees to take a prompt engineering course as quickly as possible. Barros concludes: “That’s not how you’ll succeed. It’s precisely the combination of different disciplines that makes the difference. Professionals with subject matter expertise, an eye for legislation and ethics, and that sharp ‘something’s not right’ instinct—you can’t just automate that.”