At the recent VB Transform conference, Milind Naphade, Head of AI Infrastructure Technology at Capital One, shared how they are driving growth in the automotive business by building an intelligent agent platform. Naphade noted that the company began designing these intelligent agents 15 months ago, aiming to make them resemble human agents capable of effective communication and problem-solving with customers.

Image source note: The image is AI-generated, provided by the AI image generation service Midjourney

To achieve this goal, Capital One drew on its own operational model when designing its intelligent agents. Naphade said that the financial services industry needs to effectively manage risk, so they created an agent to evaluate the performance of other agents. This evaluation agent was trained based on Capital One's policies and regulations. When issues were detected, it would promptly roll back the process. This structure allowed each intelligent agent to function like an expert team, each with different areas of expertise, working together to solve problems.

Capital One's automotive agents mainly support its dealer customers, helping their customers find suitable cars and loans. Through these intelligent agents, consumers can view car inventories available for test drives. Naphade revealed that key metrics for dealer customers, such as customer engagement and sales leads, increased by 55%. This more natural form of conversation allows customers to receive service around the clock, even getting help through chat late at night if they encounter car trouble.

In terms of data and models, Capital One has a wealth of data resources but needed to explore the best ways to introduce context and model architecture. Naphade and his team used methods such as model distillation to optimize the architecture of the intelligent agents. "Understanding agents" was a focus of their project, as it requires complex information interpretation. Additionally, they conducted multiple experiments and evaluations before launching the AI application to ensure the system's effectiveness and safety.

Naphade also mentioned that, since there was no precedent to follow, they faced many challenges during development. Despite this, Capital One is eager to apply intelligent agents to the travel industry, especially in customer interaction. With the new lounge opened at JFK Airport in New York and its popular travel rewards credit card, Capital One hopes to further promote this innovation.

Key points:  

🚗 Capital One improves car sales through intelligent agents, with key metrics for dealer customers increasing by 55%.  

🤖 Designing intelligent agents draws on the company's own operational model, ensuring efficient communication and problem-solving.  

📊 Data-driven decision-making, Capital One explores the optimal model and context introduction methods.