Banks worldwide are trying to curb the soaring cost of AI agents as spending on the tokens that power them could easily reach 20% of total labor costs, according to McKinsey’s Global Banking Annual Review.
Responsibility for controlling AI-related spending has shifted to new teams that some banks call AI FinOps. Initially, banks sought to rein in costs with systems that directed employees to different AI models based on the complexity of their queries.
“But those models are ineffective when AI agents are working, because they communicate with each other and tokens quickly add up,” said Vik Sohoni, a senior partner at McKinsey and one of the report’s authors, in an interview with Reuters.
As a result, banks are becoming more selective about where they deploy agentic AI, focusing on areas with the strongest expected returns. Productivity-related savings are relatively easy to measure, but banks are increasingly looking at AI-driven revenue growth to determine where higher spending is justified, Sohoni said.
Customers are adopting AI much faster than previous technologies, according to the report. McKinsey expects AI agents to become widely used within three years, compared with the 5 to 10 years it took mobile internet to gain broad adoption.
A McKinsey survey of 46,000 retail banking customers worldwide found that most already use AI to compare banking services, find higher-yield savings accounts and refinance or consolidate debt. Among AI users, 72% said they would allow a banking agent to carry out one financial task, although they would cap autonomous approvals at $500. Meta’s Muse agent was downloaded 5 million times in the US during its first 22 days after launch.
While AI agents could help banks boost revenue and cut costs, they could also pressure profits by shifting deposits from non-interest-bearing accounts into higher-yielding accounts. So far, that impact has been limited because most banks block external AI agents from authorizing transactions within their systems.
(Reporting by Tatiana Bautzer. Editing by Mark Potter)
