ING leverages AI for mortgage proposals, but regulations hold back

ING has recently demonstrated that its AI system is capable of preparing a mortgage proposal in 58 seconds. However, this development is not widely applicable across the entire Dutch banking sector. Other banks are opting for a cautious approach, not so much due to the new European AI legislation itself, but because of a complex interplay of privacy, cybersecurity regulations, and banking supervision. According to a bank that participated in a study by ACE + Company, they could have operated more quickly from a technical standpoint.

The study by ACE + Company, quoted by BNR, indicates that regulation acts as a brake. For instance, a single AI credit model must meet various supervisory standards, which brings about a significant administrative burden. Supervisory authorities such as De Nederlandsche Bank and the Autoriteit Persoonsgegevens independently scrutinize every aspect, which hinders the speed of project implementations. It often happens that AI models must already be adjusted even before the required checks are completed.

The biggest obstacle banks face is the burden of proof. They must demonstrate that there are no prohibited AI applications functioning within their systems, which adds extra work. While banks face little objection to the AI rules themselves, such as transparency and human oversight, proving the absence of risky applications is viewed as extremely time-consuming.

Despite these hurdles, there are already positive developments. ING takes a leading role in using AI for mortgage assessments, while other banks like ABN AMRO and Rabobank remain more cautious and primarily use AI for supporting tasks. The current regulation could lead to delays in processes, which can be frustrating for customers. ACE + Company advocates for a comprehensive regulatory approach, enabling banks to meet multiple laws with a single proof dossier. This could enhance efficiency while simultaneously reducing delays in the assessment of credit applications.

Read the full article from AI Insider.