Bank technology executives honored on a list of
Krish Swamy, chief data and analytics officer at Citizens Financial Group, expects the bank's investments in AI in 2025 and 2026 will be higher than in 2024.
"We will do fewer things and do them really, really well versus trying to do 20 different things and expecting them to take flight," he said in an interview.
"My sense is that there is still an ongoing, sharp increase in AI investment in the broader industry, which I would expect to continue for some time," said Dan Jermyn, chief AI officer at Commonwealth Bank of Australia. "Unlike most other technology advancements, the implications of AI are profound for every aspect of a modern financial services organization, and so this is as much about investing in organizational transformation as it is in being able to deploy the latest models."
Swamy and Jermyn are two of the thought leaders on the AI 100 list, which was curated by data scientists at H2O.ai (an AI software company) and Evident (an AI research firm). The list includes AI luminaries such as Fei Fei Li, co-director of the Stanford Human-Centered AI Institute and author of the book "The Worlds I See," and Dr. Yann LeCun, chief AI scientist at Meta Platforms, as well as executives adopting AI in their companies.
Other bankers on the list include Chintan Mehta, executive vice president and CIO at Wells Fargo; David Tyrie, chief digital officer and chief marketing officer at Bank of America; Matthew Burris, head of fraud network strategies and data science at Bank of America; Jamie Dimon, chairman and CEO of JPMorgan Chase; Teresa Heitsenrether, chief data and analytics officer at JPMorgan Chase; and Dr. Prem Natarajan, chief scientist and head of enterprise data and AI at Capital One.
"We are highlighting some remarkable folks who've been bringing AI into their organizations and using it to transform product but more importantly culture, and obtain a return on investment thereby for their institutions," said Sri Ambati, founder and CEO of H2O.ai. "To do AI, you need to be a great thinker, but also a great doer and be able to bring people together."
Tyrie was honored for his team's work in scaling up Bank of America's chatbot, Erica. According to the bank, Erica has responded to 800 million inquiries from more than 42 million clients and provided personalized insights and guidance more than 1.2 billion times.
Dimon was chosen for his overall advocacy of AI. In his second-quarter letter to shareholders, Dimon likened AI to the "printing press, the steam engine, electricity, computing and the internet." Dimon has said AI will be embedded in every one of the bank's processes, including trading, research, equity hedging and customer service, often as a sort of co-pilot.
It was recently reported that JPMorgan Chase has
In interviews, other honorees shared some of the top AI projects they're working on.
Blocking fraud
Several bankers on the AI 100 list said preventing fraud is a top use case. AI has been used in the financial industry
Burris at Bank of America leads a team of more than 30 data scientists, data engineers and reporting analysts who develop AI models that can decide, in real time, which transactions are highly suspicious and therefore should be blocked. Burris estimates this protects the bank and its customers from about $100 million worth of fraud every year.
Burris also created a process that uses graph technology to detect and investigate organized crime rings. He does this work with his data scientists and 70 operations analysts.
At Commonwealth Bank of Australia, Jermyn also leads a team that uses AI to protect the bank and its customers from fraud and scams.
"One use case that particularly stands out is our innovative approach to tackling technology-facilitated abuse, in which a small but significant proportion of people use online payment transaction descriptions as a means to send abusive or threatening messages to the recipient," Jermyn said. "The team developed a novel combination of AI techniques to detect and reduce this behavior."
Since 2020, the model has blocked about 1 million transactions that contained abusive, threatening or offensive words in descriptions, he said. The most high-risk examples are passed to specialists.
The bank has published the research underpinning this model and made the underlying code available to other organizations, "so that every organization around the world can benefit, and tackle this insidious form of abuse," Jermyn said.
Commonwealth has also used AI to provide immediate support to customers in natural disaster zones, and helped customers obtain more than $1.2 billion of government benefits and rebates to which they were entitled, through its benefits-finder tool.
Developer, employee productivity
Swamy at Citizens has been building operating models the bank can use to build and scale AI models. (He has been at Citizens for a year; prior to that, he was doing similar work at Wells Fargo.)
These operating models give data scientists in areas like marketing and fraud a consistent way of thinking about problems, doing the work and running machine learning models.
"The big benefit of all of that is that you get a very high degree of quality standardization and repeatability in the work," Swamy said.
Almost 20 years ago, when Swamy started getting into data science, "a lot of that work was very bespoke, very artisanal," he said.
"It required large teams of people to come together. And when that happened, you were still figuring out what you needed to do – how to assemble the data, where to place it, how to get organized with the modeling effort, et cetera. By the time you were done with building a model, you had to then worry about having to protect the model." The end-to-end process could take 18 months, Swamy said.
Today, the demands for machine learning are through the roof, he said, "and we have only so many data scientists to go around. We've got to get more organized, more efficient, with better quality – the quality bar is moving up. As regulators get more familiar with some of these tools, I think their expectations are starting to rise as well. So you've got a confluence of limited supply, increasing demand, increasing expectations from regulators. What you need to have is a methodology or a standardized way in which this gets done."
One benefit of this operating model approach is that if the bank starts a project with a few data scientists and then discovers that the work is bigger than initially thought and more people are added, "they'll know exactly what's happened before and what's coming next," Swamy said. "So their integration into that team becomes a lot more seamless, a lot quicker."
He's also working on automating some of this work, so that data scientists can spend more time getting deep into the business problems, into the nuances of the data and into the solutions the clients are looking for.
The bank has also been piloting generative AI, learning from those pilots and getting ready to put them into production. Swamy sees an overall shift in the industry from generative AI being cool, interesting and fun to being something that needs to have practical applicability.
A possible use case for generative AI at Citizens, he said, is categorization of customer complaints – it could help match a customer's angry comment to a relevant regulation under which the message should be filed.
"These models can do this type of categorization really well, because you are simply asking it to understand the English language and say, did it fit into this category or does it fit into this other category?" Swamy said. Generative AI might also be used to categorize documents submitted in a loan application process, like documents verifying income, identity or residence.
