Banks Move to Rein In AI Agent Costs as Token Spending Climbs, McKinsey Says
McKinsey's annual banking review says token bills could reach 20% of labour costs, pushing lenders to set up AI FinOps teams and deploy agents more selectively.
By Nakoda News Desk
·2 min read
Prefer Nakoda News on Google
Banks around the world are trying to control the cost of running AI agents, as spending on the tokens that power them could reach as much as 20% of their total labour costs, according to McKinsey's Global Banking Annual Review, released on Tuesday.
The finding cuts against the main case banks have made for agentic AI, which is that it will lower operating costs. Some lenders now give responsibility for AI spending to dedicated teams that they call "AI FinOps", Reuters reported, citing the review, and are concentrating agent deployments on the areas where they expect the strongest returns.
The first round of cost controls sent employee questions to different AI models depending on how complex they were. Vik Sohoni, a McKinsey senior partner and one of the report's authors, told Reuters that this approach stops working once AI agents are involved, because agents talk to one another and run up token usage quickly.
Banks are also changing how they justify the spending. Productivity savings remain the easier gain to measure, but lenders are increasingly looking to revenue growth driven by AI to support larger budgets, according to the Reuters account of the report.
The consultancy expects AI agents to reach wide use among customers within three years, compared with the five to 10 years it took the mobile internet to reach broad adoption. In a McKinsey survey of 46,000 retail banking customers worldwide, most of those who already use AI said they rely on it to compare services, find higher-yielding savings accounts and refinance or consolidate debt. Among AI users, 72% said they would let a banking agent carry out one financial task on their behalf, though they would cap autonomous approvals at $500.
That shift carries a risk for bank profits. Agents acting for customers could move money out of non-interest-bearing accounts into higher-yielding ones, eroding a cheap source of funding. The report says the effect has been limited so far because most banks do not allow outside AI agents to authorise transactions inside their systems.
The question for bank boards is whether those blocks hold as consumer agents spread. Reuters noted that Meta's Muse agent was downloaded 5 million times in the United States in its first 22 days.
Sources
- Reuters · Banks seek to rein in soaring AI agent costs, McKinsey says
- Global Banking & Finance Review (Reuters) · Banks Take Action to Control Soaring AI Agent Costs, Says McKinsey
- AOL (Reuters) · Banks seek to rein in soaring AI agent costs, McKinsey says
Produced with AI assistance from the sources above. How we work
By
Nakoda News Desk
Nakoda News covers AI and business. Every story links the sources it draws on; how we work is set out in our editorial standards.


