As fraud evolves, so does the investigation

Fraud investigations are becoming more digital, faster, and more information-intensive. AI can reveal connections that humans easily overlook. But a signal does not necessarily mean a conclusion. That is precisely why the investigator’s expertise is also changing.

More information, more connections

A permit application, subsidy case or fraud investigation is no longer just a single form and a handful of supporting documents. Investigators combine information from documents, transactions, public registers, corporate structures, digital traces and networks. At the same time, criminals are using digital technology and AI to operate faster, conceal structures and scale up their activities.

Technology is making more and more information accessible. This makes the investigator’s judgment all the more important: what should you look for, which signals can you trust, and what do they mean in context?

An uneven playing field

The playing field is fundamentally uneven, says Prof. Ana Isabel Barros, Professor of Data Science and AI at Tilburg University and JADS. Criminal networks can adopt technology quickly, without having to operate within legal or ethical constraints. Governments and law enforcement agencies, by contrast, need to work in ways that are transparent, explainable and open to scrutiny.

There is another challenge: AI models learn from historical, observable data, while criminals deliberately keep their structures hidden. As Barros puts it: “AI models learn from patterns in what is visible, not from what is actually happening.”

The answer, she argues, is therefore not more AI, but smarter use of it – combining technology with domain expertise, critical judgment and collaboration.

What the documents reveal

Consider the fictional case of Café De Zilveren Linde. Its owners apply for a permit to open a second location. Sanne Verhoef is listed as a shareholder and financier. Her husband Dennis, who has previously been convicted in connection with a cannabis-growing operation, remains out of sight. His interest in the business is held through an intermediary holding company.

Viewed separately, the documents in the case appear plausible. Analyzed together, however, inconsistencies begin to emerge. Sanne appears in the shareholder register but not in the Chamber of Commerce records; for Dennis, the reverse is true. Two documents each name a different person as the “sole shareholder” of the same holding company. Bank statements also show irregularities, while revenue increases sharply even though staffing costs remain unchanged.

AI can identify and organize these inconsistencies, anomalies and relationships within minutes, while linking each finding back to its original source. What would normally take days of manual work can be turned into a single, verifiable overview. But it is still the investigator who determines what those signals mean and what further investigation is needed.

The human role is changing

AI can process “incredible amounts of data in fractions of a second” and act as a critical colleague, prompting questions such as: have you considered this? But users need to understand when the technology is appropriate and when an outcome does not make sense, says Dr Daniëlle Sent, lecturer and researcher at JADS.

That requires more than technical skills. Investigators need to ask the right questions, verify sources and intermediate results, recognize bias, and understand the impact of missing or unreliable data. Procedures and responsibilities also need to be clearly defined in advance.