Technology news and discussion on TalkFuse.
Banks seek to rein in soaring AI agent costs, McKinsey says
Banks are trying to control the rising cost of AI agents, as token spending could reach 20 per cent of total labour costs, according to McKinsey. The expense is prompting banks to create specialist “AI FinOps” teams and to focus deployments on uses expected to deliver the strongest returns.
Earlier cost controls that routed staff queries to different AI models do not work as well for agents, which communicate with one another and quickly use more tokens. McKinsey says agents could become widely used within three years; in its survey of 46,000 retail banking customers, 72 per cent of AI users said they would let a banking agent complete one financial task, with autonomous approvals capped at $500. Agents may raise revenue and cut costs, but could also shift deposits into higher-yield accounts and put pressure on profits.
In short
- AI agent token costs could reach 20 per cent of banks’ labour costs.
- Banks are targeting deployments where returns justify the expense.
- 72 per cent of surveyed AI users would delegate one financial task.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Banks have a responsibility to manage costs and protect profitability, especially when AI agents can consume far more tokens than earlier tools and may affect deposits and lending margins. Specialist AI FinOps teams and carefully chosen deployments could help ensure agents are used where they deliver clear customer or business value, while limits on autonomous decisions contain risks.
The case against
Tight cost controls and narrow deployment criteria could slow adoption of tools that customers appear willing to trust with routine financial tasks and that may improve service or reduce operating costs. Since agents can handle more complex work than simple query tools, banks may need to invest and learn at scale now to realise those gains, rather than judge them chiefly by near-term token spending.
Explained simply
Banks have long used software to process payments, assess applications and help customers manage their money. Newer AI agents can carry out tasks with less step-by-step direction, sometimes by asking other AI systems for help.
Each exchange can use computing resources measured in “tokens”, so costs may rise as agents take on more work. Banks are weighing those costs against possible gains, such as faster service or lower staffing needs, while deciding which uses are worthwhile.
The shift matters to customers and banks alike: agents could handle routine financial tasks, while banks would need to manage how much authority they give them and how they affect customer accounts and profits.