Capital One shifts AI strategy to prioritise customer needs over technology capabilities

Capital One has institutionalised a new approach to artificial intelligence known as "customer-back engineering". This strategy reverses the traditional model by prioritising real-world customer needs and challenges before determining the technological capabilities required to solve them. Managing vice president Ashish Agrawal states that this method allows engineers to address practical issues more rapidly than when they start with technology first.

The shift addresses a broader industry concern highlighted by McKinsey research, which indicates that organisations often fail to capture more than one-third of the value from their digital investments. This shortfall is frequently attributed to companies beginning with technological solutions rather than user problems, resulting in fragmented experiences. By flipping the script, Capital One aims to create cohesive transformations that directly impact customer lives.

To ensure engineers remain grounded in these realities, the bank has set a specific goal for every engineer to establish multiple direct touchpoints with customers throughout the year. Agrawal notes that the biggest hurdle for technologists in large corporations is typically a lack of direct access to the people they serve. By overcoming this barrier, teams can identify problems from a unique dimension, leading to what Agrawal describes as a multiplier effect in innovation.

This philosophy has been demonstrated through the launch of Chat Concierge, a multi-agent AI framework designed to assist car buyers and dealers. The system performs tasks such as comparing vehicles and scheduling test drives within a single conversation. By embedding AI directly into workflows, the bank shows how such tools can transform customer experiences in financial services when guided by genuine user requirements.

The implementation relies heavily on high-quality, well-governed data as a non-negotiable foundation. Agrawal explains that a clean data layer is essential for orchestrating agentic loops, enabling systems to perceive, reason, and execute tasks effectively. When combined with agentic AI tools, this rich data ecosystem allows the bank to move from incremental fixes to high-velocity transformation.

Recent survey data from MIT Technology Review Insights supports the momentum behind these developments. The findings show that 70 per cent of leaders report using agentic AI to some degree, with banking executives expressing strong confidence in its ability to improve fraud detection and security. Looking ahead, most executives expect to continue leveraging these technologies to enhance both security and the customer experience.