The data: Medium-sized and large banks are dramatically accelerating the rollout of genAI tools, according to a recent survey: 47% have “already rolled out generative AI either internally or in a client-facing application.” Revenue gains and costs savings from genAI have also come into relief:
- Revenue gains are mixed thus far: 38% of respondents experienced an increase of less than 5% as a result of genAI, and 39% said 6% to 10%.
- Respondents expect higher increases in revenue from investment in genAI within the next two years: 36% expect a 6% to 10% increase, and 43% expect 11% to 20%.
- Over the past one to two years, the most respondents have seen the greatest cost sayings in technology (13%) and customer service (12%). In the next two years, they expect to see the greatest cost savings in customer service (16%), operations (14%), and technology (14%).
Digging into the data: Banks are experimenting with genAI, but progress isn’t even:
- GenAI solutions were in use by 20% of respondents and agentic AI by 8%., But banks have struggled with genAI investments despite reported or expected gains.
- Among the 16% of respondents who had genAI use cases in production, 40% said those use cases failed to reach the desired result and/or were discontinued. And among the 29% who had use cases in development, 33% said they failed in development.
Zoom out: As banks’ internal use cases for genAI have grown, they’ve publicly embraced the technology for such cases, and the largest have invested billions in AI overall. This year, Goldman Sachs launched an internal AI assistant across the bank, and JPMorgan Chase rolled out its LLM Suite to most of the company.
Our take: Banks’ genAI adoption is complicated by the depth and breadth of use cases across institutions and the degree to which each is investing in proprietary software and fundamental research.
GenAI is a crucial enabling technology for internal and external banking use cases. But three years after the public release of ChatGPT, institutions’ AI maturity differs dramatically. Experiments with genAI tools are necessary and should yield cost sayings and gains in productivity. But results won’t be equal. The race is on.