A Solution for Finding Solutions
For customer service agents, Natarajan says the solution is to be able to find the right answer in the moment, and fast. “What's the current problem in finding the right answer? Well, when I search in the knowledge base, I have to click through so many answers and then decide which is the right one. But if the first one I click gets me the right one, I'm done. Better yet, if the system can synthesize the answer from multiple sources when needed, that’s a big win.” In practice, this increased efficiency improves the experience for both agent and customer.
“Our agents are the voice of our business to our customers, working tirelessly on their behalf in the moments that they need it most. We are excited to apply the most innovative technology to ease the cognitive load on our agents and enable them to provide faster solutions to the issues our customers are calling about,” shares Tamara Sigler, head of servicing strategy at Capital One. “We are focused on delivering effortless resolutions for every customer in every situation. As the needs of our customers evolve, so must our intelligent solutions.”
To date, the agent servicing tool has been used more than 10,000 times by thousands of agents across the business, achieving relevant and accurate data retrievals for more than 85 percent of results. That’s an impressive percentage, considering that since the company is using generative AI models, the technology is trained to constantly apply its skills and data to more and more complex tasks.
We often don’t think about customer service from the customer service agent’s point of view. For the customer, the issue began when someone potentially stole your credit card, when your Wi-Fi stopped working, or when your order arrived but didn’t work. You’ve already invested all this time (that you already didn’t have) to get the problem solved. But from that customer service agent’s perspective, this is ground zero. They are, in some sense, already three steps behind.
If a customer service agent can’t quickly find the information needed to solve the problem and address a need—no matter how kind or professional they are—the anxiety and stress for both them and the customer intensifies. That’s why companies like Capital One are finding new ways to integrate generative AI tools that actually speed up the process of problem-solving.
“One of the noble aims of AI has long been transferring the cognitive burden from a human to a system,” explains Prem Natarajan, EVP, chief scientist and head of enterprise AI at Capital One. The way he sees it, for most complex tasks, “improving the customer experience necessitates improving the agent's experience.”
To enhance the experience for both agents and customers, Capital One has created a new proprietary generative AI agent servicing tool. It leverages state-of-the-art advances in semantic search and retrieval augmented generation to reinvent how agents are able to accurately search our knowledge base and get access to information to resolve customer questions more quickly and efficiently than ever.
The tool essentially creates a flywheel of insights and feedback that help to continually improve the search and summarization results of the generative AI models, helping agents and—ultimately—customers.
Traditionally, an agent would have to take a customer’s human-language question and translate it into a keyword search in their database at work to find an answer. “With generative AI, we are able to say, these two examples are the same question,” says Natarajan. “I can index the answer in a way that the retrieval system knows, ‘Oh, that is the same answer to bring back.’ The technology in this gen AI agent servicing tool lays the foundations for that.”
And because Capital One has thousands of agents using their agent servicing tool every day in thousands of use cases while serving customers, the tool is getting tested at scale and can keep improving. “It's a fantastic opportunity for us to improve the technology and also optimize the solution with their contributions,” says Natarajan.
This kind of continuous learning and application is what makes the underlying technology within Capital One’s agent servicing tool capable of scaling across many different tasks, use cases, and applications.
The development of Capital One’s first homegrown generative AI tool is emblematic of its broader approach to AI—focusing on responsible, human-in-the-loop solutions that are based on solving real human needs. As the company advances its AI initiatives to support consumers and broader financial inclusion, it continues to bring diverse perspectives and equities to the table—from collaborations with leading academic institutes and multi-sector partnerships to philanthropic and community-centered investments.
For many businesses and professionals, there are certain tasks that don’t require critical thinking as much as the ability and patience to wade through a deluge of information to complete a task. That can apply to educators or scientists looking for updates in research and subject matter to update their curriculum or studies. It can apply to risk professionals trying to learn what has changed from the previous version of policy or guidance to the new one. It can apply to lawyers redlining contracts for clients, or health professionals learning new methods and treatments, and myriad others.
“Similar to industries like healthcare, at Capital One we know banking can play a critical role in enabling consumers to live their lives. We take that responsibility very seriously and design our customer service to meet the needs of our customers using the best tools,” says Sigler.
So far, the Capital One team has seen success through testing models as well as through anecdotal experiences of agents using the tool. And it’s the latter that, to Natarajan, means that the technology is a success.
“Anything that lights up the impact for the end users is incredibly inspiring and energizing for the technical teams,” he says. “You might have this notion that the technical teams are there because they enjoy the technical work, but more often than not, the sunshine in their days is when that impact has been delivered and acknowledged by the users.”
To him, that means that both the agent and the customer experiences are being reimagined for the better. It means that customer-facing agents can have better access to the information that helps them solve problems faster. It means that the customer may experience less stress—and the customer service agent will, too.. “I think at its best AI unleashes a virtuous cycle of feedback and improvement that drives increasing satisfaction through every interaction down the chain,” says Natarajan.
“We are sloping up through these more and more complex use cases—it’s like building the ‘stairway to heaven,’” explains Natarajan. “You start with one step at a time, and you make sure you've learned from it.”
Generative AI models have the unique ability to synthesize information from multiple sources to retrieve a single response, as well as the ability to translate natural language into a keyword search. For instance, you may ask a friend “Hey, is there a place nearby where I can take cash out?” Most database search engines wouldn’t do well in answering that exact question, and would instead need a sentence that includes keywords for that search, like “Where is the nearest ATM?”
Designed With Intention
