Case Study: Generative AI at ING

In September 2023, ING, a leading global bank, collaborated with McKinsey’s AI arm, QuantumBlack, to build and launch a customer-facing generative AI chatbot. This project aimed to address longstanding customer service challenges by leveraging cutting-edge generative AI technology. The chatbot represents a major milestone in ING’s journey to improve customer experience and operational efficiency across its global network.

Key Takeaways

• The generative AI chatbot was developed in just seven weeks, significantly faster than traditional chatbot development timelines.
• The AI tool assisted 20% more customers within the first seven weeks, reducing the need for live agent interaction and decreasing wait times.
• Once deployed across ten markets, the chatbot is projected to impact 37 million customers.
• ING prioritized security and customer protection by engaging risk stakeholders early in the development process and implementing robust guardrails.

Approach

ING and McKinsey adopted a collaborative and iterative approach to address the identified gaps in customer service. The teams began by defining the business opportunity: in the Netherlands alone, 40-45% of customer queries were resolved by the existing chatbot, leaving 16,500 queries per week requiring live agent intervention. Daily regression testing on 500 real customer chats enabled the teams to refine the chatbot’s performance in real time.

Risk management was a critical component of the process, with risk stakeholders engaged from the outset to co-develop guardrails tailored to ING’s needs. Additionally, more than 50 support functions were trained to ensure the seamless integration of generative AI technology, setting the stage for scalable deployment across ING’s global operations.

Implementation

The chatbot development unfolded over seven weeks, beginning with an in-depth analysis of the limitations of ING’s classic chatbot. A multi-step pipeline was created to retrieve knowledge from internal data stores, rank potential answers by relevance, and provide customers with multiple options for ambiguous queries. Guardrails were implemented to ensure compliance with ING’s regulatory requirements, such as avoiding advice on mortgages and investments. These safeguards were built using out-of-the-box tools from QuantumBlack Labs.

The chatbot was initially piloted with 10% of customers using the support chat function in the Netherlands mobile app. During this testing phase, the chatbot demonstrated its ability to deliver tailored, faster responses, significantly improving the customer experience.

Results

The generative AI chatbot quickly demonstrated its value by improving resolution rates by 20% compared to the classic chatbot, reducing wait times, and delivering a more satisfying customer experience. Customers benefited from instant assistance, minimizing the need for live agents to handle non-urgent queries. Operational efficiency also improved, as fewer calls were routed to ING’s call centers. The success of the pilot created a clear roadmap for scaling the technology to ten markets, potentially impacting 37 million customers worldwide. The technical foundation established during the project has also opened opportunities for ING.

Challenges and Barriers

The project faced several challenges, including the complexity of managing risks in a highly regulated industry. Ensuring the chatbot’s compliance required extensive collaboration with risk stakeholders and the development of custom guardrails. Building customer trust was another priority, necessitating rigorous testing to ensure the chatbot delivered accurate and reliable responses. The iterative development process, which involved daily performance reviews and A/B testing, demanded agility and close coordination between teams to refine the tool and address any issues promptly.

Future Outlook

ING’s success with the generative AI chatbot sets the stage for broader adoption of AI technologies across its operations. In the short term, ING aims to double the chatbot’s performance within the next six months, further enhancing its resolution capabilities and efficiency. Over the long term, the chatbot will be scaled to ten additional markets, benefiting 37 million customers globally.

To get the latest AI transformation case studies straight to your inbox, subscribe to AI in Action by AIX — your weekly newsletter dedicated to the exploration of AI adoption in business.

Elevate your understanding of AI transformation. Browse AI adoption case studies searchable by company, industry, use case, and technology.

Sources:
Banking on innovation: How ING uses generative AI to put people first

---

Let’s talk

Whether you’re looking for expert guidance on AI transformation or want to share your AI knowledge with others, our network is the place for you. Let’s work together to build a brighter future powered by AI.

---