McKinsey’s annual global banking report warns that rapidly growing costs for AI agents are creating significant challenges for banks worldwide: spending on the tokens required to operate these models could reach as much as one-fifth of a bank’s total payroll. Reuters summarized the report, noting that these developments are prompting changes in how financial institutions govern and finance AI deployments.
New teams: AI FinOps
Banks have formed teams specialized in AI financial operations (AI FinOps) to oversee and optimize spending on AI. Earlier cost-control approaches routed employee requests to different AI models based on complexity, but McKinsey says that method is no longer adequate for autonomous AI agents.
Why costs can spike quickly
Vik Sohoni, a senior partner at McKinsey and one of the report’s authors, told Reuters that autonomous AI agents communicating with each other dramatically increase token usage. Because agents hand tasks between themselves, the number of tokens consumed — and therefore the bill — can escalate very quickly.
As a result, banks are increasingly selective about where they deploy agent-based systems, focusing on processes that promise the highest expected returns.
Return on investment as a decision criterion
While efficiency-driven cost savings are relatively straightforward to measure, McKinsey reports that banks are increasingly evaluating the revenue-generating potential of AI initiatives. Higher spending is being justified only where increased revenues or other measurable business benefits are expected.
Customers adopt AI faster than prior techs
Consumers are adopting AI tools more quickly than they did earlier technologies: the consultancy expects AI agents to become widely used within three years, compared with the five-to-ten years it took for mobile internet to reach similar penetration.
A global survey of 46,000 retail customers found that most already use AI for comparing banking services, finding higher-yield savings accounts, or refinancing and debt consolidation. Among AI users, 72 percent said they would entrust a specific financial task to a banking agent, although they would cap the value of independently approved transactions at around $500.
Interest is illustrated by the fact that Meta’s Muse agent was downloaded five million times in the United States in the first 22 days after its launch.
Profitability risks
Although AI agents can help banks reduce costs and increase revenues, McKinsey warns that profitability could be hurt if agents automatically reallocate deposits from non-interest-bearing accounts to higher-yield products. For now most banks block external AI agents from authorizing transactions within their systems, so those effects remain limited.
Implications
The report suggests further changes in regulation, risk management, and operations at banks: precise measurement of cost control and revenue impact will become critical, and the rollout of AI agents will be subject to strategic filtering to keep technology-driven expenditures sustainable.
This article is based on McKinsey’s report as reported by Reuters. The author notes that an AI assistant contributed to the preparation of the piece; the final copy was edited and verified by a journalist.
