ING Tests AI on Its Most Complex Mortgages

ING has deployed an artificial intelligence agent for mortgage applications in the Netherlands after a pilot announced in June. The system reviews applications, explains possible outcomes and recommends next steps. A bank employee then assesses the application and decides whether to approve or reject the loan.

About 70% of ING’s mortgages are classified as complex. They include applications from freelancers who do not have fixed incomes. Marco Li Mandri, ING’s change lead for scalable agentic products, described mortgages as a difficult area for AI because applications require specialist knowledge and vary from case to case.

“You would probably think that at face value, mortgages is the worst domain to apply AI,” Li Mandri said at a press event in London. “In reality, our view is when it comes to agentic, this is the perfect domain because agentic works very well in domains that are not very standardised.”

The system has reduced the time required to process a mortgage from seven days to five. ING is expanding it to other countries, including Germany, where some simpler mortgage applications can be approved within 30 minutes. Human employees continue to make the final decisions, and the AI agent must explain the confidence behind its recommendations.

ING has no planned job cuts linked to the mortgage system, according to Li Mandri. He indicated that staffing changes could occur in other areas of the bank.

ING uses Google Gemini Enterprise technology to develop its AI products. Bahadir Yilmaz, the bank’s chief analytics officer, said ING chose to continue working with Google after employees became familiar with its tools and the bank reviewed its existing technology.

The bank is testing a conversational AI chatbot in the Netherlands and plans to launch it in Belgium, Germany, the Netherlands and Italy in 2026. The chatbot provides personalised information, helps resolve problems and can take certain actions. It will operate alongside ING’s mobile app.

Yilmaz described conversational banking as a service customers are likely to seek in the next five years. The chatbot’s initial functions include providing information, solving problems and taking actions. Future functions could include sending and receiving money, applying for products and providing advice within regulatory limits.

ING is also testing AI for wholesale and investment banking. Yilmaz said the technology could help deal teams prepare for client meetings by reviewing documents and analysing transactions. Investment banking deals can last six to 18 months and involve teams of 10 to 20 people.

ING has about 3,000 data and analytics employees across its business lines. The bank keeps its AI specialists in a central data analytics organisation while assigning them to projects across the business. Yilmaz said the structure helps ING recruit and retain data scientists, engineers and product managers.