Bank of America, JPMorganChase execs report returns on AI 

• Key insight: Large banks are starting to see measurable returns on AI.
  
• Supporting data: BCG estimates AI could unlock over $370 billion annual banking profit by 2030.
  
• Forward look: ROI may need to be redefined as generative AI scales workforce and workflow leverage.

Source: Bullets generated by AI with editorial review

Research the Boston Consulting Group published this week found that AI could unlock more than $370 billion in annual profit for the global banking industry by 2030, yet most institutions remain far from ready to capture it, the consultants said.

"Actual examples of real value attained from deploying AI — in the form of cost savings, revenue gains and [earnings before interest and taxes] increases — are hard to find," the BCG consultants wrote in their report, which came out Wednesday. "Most retail banks are piloting AI in some functions or operations. A few have embarked on more ambitious end-to-end transformation of functions or workflows. But results so far have been limited."

In the wild, however, some bank technology leaders say they are starting to see quantifiable returns on their AI investments.

JPMorganChase is "absolutely seeing value from AI," Teresa Heitsenrether, the bank's chief data and analytics officer, said at American Banker's recent Most Powerful Women in Banking conference. "It's very quantifiable, but it's mostly from traditional machine learning right now with areas like fraud and credit card marketing."

For instance, the bank has seen measurable improvements in fraud avoidance, fraud detection, and improved ability to determine which transactions truly are risky and should be denied versus those that seem unusual but are legit, she said.

"These are all the things that AI does every single day that create value, not only for the country, but for our customers," Heitsenrether said. "It has been for years, and continues to evolve and get better."

At Bank of America , which this week said it plans to invest $13 billion in technology next year, Hari Gopalkrishnan said the bank is seeing cost savings through AI-based self-service tools like Erica, which over the course of its life has handled three billion customer interactions. On a recent day, it had two million customer conversations.

"We know that, for example, those two million conversations that happened yesterday are dropping call center volume and helping customers serve themselves on their own time, so that's a clear drop to the bottom line type of savings," Gopalkrishnan, chief technology and information officer at the bank, said at Evident's recent AI Symposium. His bank has also seen reduced fraud losses after upgrading to AI models.

Bank of America runs 270 open source and proprietary gen AI models throughout the organization, Gopalkrishnan said at the bank's investor day on Wednesday. The 3,000 employees in its Global Markets unit use large language models for search and summarization of market research. Payment staff get access to policies and procedures through a bot called askGPS. Relationship managers use gen AI to create first drafts of client meeting prep documents, which saves time, Gopalkrishnan said.

"The hours of time they save can then be used to deepen prospects … freeing up time to then go after the high-value opportunities," he said.

Chatbots in trading operations help simplify the trade lifecycle and diagnose issues up front, Gopalkrishnan said. AI models used in fraud detection have led to a 55% reduction in fraud losses. And 18,000 software developers use coding agents to optimize the development process, which has led to 20% productivity improvements.

Some benefits are hard to measure

It's difficult to calculate ROI on gen AI, Sumeet Chabria, CEO of Thoughtlinks, said at the Most Powerful Women in Banking conference.

"There's a one-time infrastructure investment needed at the enterprise level, a financial commitment that has to be made, and that overhead cannot be passed to the first few use cases," he said. "A lot of times, the first few use cases are all about experimentation anyway."

Heitsenrether pointed out that though large language models are saving employees' time, it's hard to attach dollar amounts to that.

"We've rolled it out to a lot of people, and I know for a fact that they're gaining hours of productivity a week, but that doesn't show up anywhere in an income statement," she said. "It's very hard to measure it. It's just a ubiquitous way that everyone's going to work, and if your employees don't have it, by definition, you're that much less efficient."

Rethinking return

Gopalkrishnan said Bank of America is reinvesting some of the gains it's getting from AI.

The 20% productivity lift in the coding life cycle, for instance, is being reinvested in new initiatives, "which means I go from $4 billion to $4.1 billion without spending $4.1 billion," he said. "And that's the virtuous cycle we want to create."

ROI may have to be rethought, Heitsenrether said.

"What would you do if I could give you 2,000 more people in your business at zero cost tomorrow?" she said. "It's that type of thinking that I think just hasn't absolutely permeated everybody's consciousness yet of just how much scale and parallel capacity is going to be possible."