If one were making a list of the companies that have most exploited the potential of data, analytics, and artificial intelligence in the 21st century, Capital One would probably be at or near the top. Spun out of the credit card business of Signet Bank in 1994, Capital One was founded on an “information-based strategy.” It was a pioneer in the use of data and analytics (and one of Tom’s poster children for his 2006 article about competing on analytics) for its targeted marketing, marketing experiments, variable credit card interest rates, and analytical risk management. As of September 2026, it is the eighth-largest bank in the U.S. by assets and focuses not only on credit cards but also consumer and commercial banking.
It’s not surprising that Capital One is very active in AI as well. It is ranked second among global banks in AI capabilities, in Evident AI Index of 2025, following only the much larger JPMorgan Chase. Both of us (Tom here and Randy here) have written about AI at Capital One in the past, but, given how quickly AI is changing, we thought its activities were worth updating for our last AI in Action column for MIT Sloan Management Review.
To learn about the bank’s current focus, we spoke with Prem Natarajan, executive vice president, chief enterprise AI officer, and chief scientist; and Amy Lenander, executive vice president and chief data officer. Natarajan has a strong technical background, having worked in a prominent AI role at Amazon and as a professor and senior vice dean of the School of Engineering at the University of Southern California. Lenander has a 23-year history in different business roles at Capital One, including heading the bank’s U.K. business.
Tom was in the midst of doing some research on the economic value of AI when he interviewed Natarajan, so that was one of the first issues he explored. Natarajan, a long-term AI enthusiast and expert, had an interesting take on the issue: “You can’t achieve transformation if you’re focused on quarterly ROI.” He believes that with a powerful, fast-changing technology like AI, it’s both too early and short-sighted to measure short-term ROI. Capital One’s leaders believe that AI could reshape the banking and finance industry, and the company is dedicated to thriving through the changes to come. “The ability to manage and get access to financial services is crucial to having a satisfying life, and AI will be a massive enabler of that,” Natarajan said. “The ultimate ROI is a happy customer.”
Core AI Use Cases at Capital One
Natarajan mentioned several areas in which AI is already having a substantial impact at Capital One, including software development, fraud reduction, and customer service. Providing high-quality customer service has long been a focus of the bank, and AI is being used to further improve it. Generative AI is being used to surface and deliver the right answers to human agents faster. That provides stress relief for the human agent, and the caller experiences a faster transaction with more accurate answers to their questions.
Another AI use case in customer service is call summarization, which employs agentic AI to recall the most important elements of conversations with customers and use reasoning capabilities to interpret complex customer issues, particularly on fraud-related calls. Agentic AI allows a complex summarization task to be broken into subtasks for each agent to perform. Natarajan said that only when the bank added a reasoning agent to the summarization did the performance of the process become satisfactory. Several hundred agents are now using call summarization and have realized substantial productivity gains, with the summaries added to the customer record. The early adopters have seen a “massive improvement in customer experience,” Natarajan said.
Any well-managed bank also needs to excel at fraud prevention and risk management, and Capital One is using AI in multiple ways to improve its processes. One of them is in auditing risk and compliance; the bank has built an intelligent assistant for risk associates, using open-weight models, advanced analytics, and analytical AI. Risk assessments can be done more quickly with better quality control, script review assistance, root cause analysis, and data-driven assurance at scale. Use of the intelligent assistant in audit risk and compliance has reduced quality errors by 85%, Natarajan said. It has also reduced time spent on manual audit tasks, saving an estimated 5,000 hours of audit time across 30 auditors in only a few months. One audit task, for example, has been reduced from five hours to three minutes to perform.
Natarajan emphasized that the bulk of Capital One’s AI initiatives are oriented to enterprise and line-of-business capabilities rather than individual productivity. “Whatever we introduce,” he noted, “we put our backs into it. It’s not ‘Here’s a tool — go use it.’” The bank invests heavily in training for any new use case and in making sure people have the capabilities needed to use new tools competently. As Natarajan, who worked closely with the intelligence community in a previous job with Raytheon BB&N, put it, “Every revolution in technology has only happened when tradecraft has evolved alongside it.”
The Data That Makes It Possible
Natarajan and Lenander both asserted that data is critical to competitive advantage with AI. Natarajan commented that “proprietary data brings competitive advantage with AI,” and Lenander noted that “building a unified data ecosystem creates a competitive moat.”
Lenander views the current focus on AI as an extension of previous capabilities. “We have a long history of automated decisioning for credit, marketing, and other important domains,” she said. “Early on, the method was regression, then machine learning, then more complex machine learning, and now generative AI. The data we used were initially structured, and now it’s a mix of structured and unstructured.”
For example, she said, the data used to train customer service assistant models is from call transcripts, among other sources. Those transcripts are also analyzed by generative AI to determine what themes the bank is seeing in customer interactions, such as how they’re responding to the rollout of new products. With the combination of that data and AI, the bank can see trends in near real time. The bank also analyzes clickstream data from the public-facing website, and from the mobile app and website customers use to make transactions. With that data, analysts can learn how customers spend their time and attention and can optimize the company’s marketing funnel, sharing the most relevant information with customers to resolve their banking issues.
Lenander pointed out that even though AI can take very unstructured data and make use of it, its outputs are better when the data is organized. Just as Capital One organized its data to train machine learning models, it is now investing resources in curating its unstructured data. One key tool that has been helpful is knowledge graphs, which characterize and map the relationships between data entities such as people, concepts, and objects, to identify, for example, which customers use which banking products. Curating more data underneath the knowledge graph makes it more valuable.
The bank gives responsibility for the curation of unstructured data to the people who are closest to it. In fact, it uses the same framework with unstructured data as structured data by managing it as “data products” that are packaged and ready for data consumers to use, in whatever format those data consumers need. With data products, ownership stays close to the data itself: The customer service team, for example, owns the data product for customer pathways through the call center system. This approach to data products is similar to Capital One’s general approach to data management, with federated data ownership and central data capabilities and standards. Lenander oversees a central team that ensures that all of the data products fit together, and she helps federated groups build their data products and apply a consistent, high bar for quality and usability.
Capital One has also taken a step into the market for data management software. Databolt was developed by the bank’s internal software team to tokenize data for security purposes. Databolt enables organizations to secure sensitive data across large language models and AI workflows without exposing personally identifiable information and is now marketed to external customers through Capital One Software. It is the second product for that business, following Slingshot, a tool that helps companies predict and manage cloud spending.
Our discussions with Natarajan and Lenander reassured us that this company that we’ve studied for decades is still leading the field of data, analytics, and AI-driven enterprises. It’s a trait that at this point is deeply embedded in Capital One’s DNA.