Capital One Builds AI Platform Around Customized Open-Weight Models

Capital One is building its artificial intelligence solutions around deeply customized open-weight models rather than relying solely on off-the-shelf foundation models. The company has established a scalable, multi-agent AI architecture that significantly utilizes its proprietary data.

The foundation for this development was laid years ago through early investments in data transformation and cloud adoption, which enabled a rapid response to the current wave of AI. Capital One has constructed a centralized, enterprise-wide AI platform incorporating built-in governance. The company fine-tunes open models with its own data, providing a distinct competitive edge.

"We view our data as a huge advantage that nobody else has, something that general frontier models cannot provide. So we are taking that data and deeply customizing these models," stated Kel Vanee, MVP of machine learning engineering at Capital One. He also emphasized the critical role of real-time data in bringing fresh context to customer and associate interactions.

An example of this approach is a customer service workflow for bank fraud, which employs a multi-agent system named MACAW. This system routes interactions to specialized agents equipped with built-in governance and guardrails. This enhances both customer service efficiency and support for employees handling complex fraud cases.

Looking ahead, Capital One anticipates a growing significance in model routing and proactive, event-driven AI. The company believes that combining various models can yield improved accuracy compared to individual models. Proactive AI can automatically respond to identified situations, thereby boosting monitoring and resource utilization.