A joint ING and McKinsey team worked closely together for seven weeks to build a generative AI chatbot that offered customers immediate tailored help while maintaining clear guardrails to mitigate risk. The team started with an in-depth analysis of the existing chatbot to identify specific challenges. The final solution consisted of a multi-step pipeline to generate the best answer for the customer including knowledge retrieval from available data stores and a ranking of potential answers by helpfulness. In the case when several helpful answers arose, the system would offer multiple options to the customer – a process called disambiguation.
Lastly, before any answer was sent to the customer, a series of guardrails was applied. In order to do this, ING risk stakeholders were brought into this process from the very start and ING-specific guardrails (to avoid giving advice on mortgages and investment products, for example) were built using out-of-the-box tools from QuantumBlack Labs, our AI and machine learning innovation hub.
The gen AI chatbot was released on testing days to 10 percent of real customers in the Netherlands who were using the support chat function on the mobile app. It offered a customer experience that was demonstrably superior to the classic chatbot, providing customers with much more detailed and tailored responses, and helping them resolve their queries faster. Since launching in September 2023, thousands of customers have interacted with the new gen AI chatbot, making it the first-of-its-kind real-life customer-facing pilot conducted in Europe.
“By prioritizing real customer needs, security, and ease of use, ING was able to develop a bespoke customer support tool that gives users the best possible experience,” says ING chief operating officer Marnix van Stiphout.