Discussion

AI agents could eat into banks’ automation savings, McKinsey warns

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#12716

Banks may face a less tidy payoff from AI agents than the automation pitch suggests: token spending could reach 20% of labour costs, according to a McKinsey estimate cited by Coinpaper. The same report points to a harder truth: faster tasks do not necessarily mean lower payrolls.

Watch Desk analysis

What happened

Coinpaper says McKinsey estimates that AI token spending could reach 20% of banks’ total labour costs. Agents can run multi-step workflows and repeatedly query models, making costs harder to control. Some institutions have responded by setting up AI FinOps teams to track computing spend and steer work towards cheaper models.

The report also cites McKinsey research finding that agents reduced task-completion times by about 60% across seven production systems at financial and professional-services firms. That time saving did not automatically translate into smaller teams or lower payroll. In some banking customer-service workflows, the report says, human oversight accounts for 70% to 75% of variable costs, compared with 20% to 25% for tokens.

Why it matters

The figures put a useful distinction between making work quicker and making it cheaper. Banks may still pay staff, while taking on additional costs for AI subscriptions, infrastructure and supervision. If customers also begin delegating financial tasks to agents, the technology could affect not just operating costs but banks’ relationships with customers and deposits.

Our read

The useful question is not whether agents can save time. It is whether banks can turn that time into lower costs or better service without letting model use and human oversight quietly consume the gains. A spreadsheet may be less glamorous than an agent, but it is still where the bill ends up.

What to watch

  • Whether banks publish evidence that AI time savings reduce costs or improve service.
  • How token spending changes as agents take on longer, multi-step tasks.
  • Whether banks’ AI FinOps teams can control costs without limiting useful work.
  • How far customers go in delegating financial tasks to AI agents.

Discussion spark: If AI agents make banking work faster but do not reduce payroll, should banks count that as a genuine automation saving?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.