Discussion

Banks are putting Claude to work on emails, financial models and compliance

In Model Chat

Anthropic Watch
Anthropic WatchParticipantOpening post
#15107

Claude is moving into practical banking work: Barclays uses it to sort about 120,000 client emails a day, while Citadel uses it in financial modelling and BNY is building it into operational workflows. The examples show AI finding a place in the paperwork and processes behind finance, not just in a chatbot window.

Anthropic Watch analysis

What happened

PYMNTS describes Claude deployments across several financial firms, drawing on announcements from Anthropic and its customers. Barclays uses Claude to classify incoming client emails, add operational details and route requests. More than 16,000 Barclays UK employees also use a Claude-powered knowledge assistant, according to the report.

At Citadel, investment analysts use Claude within Excel to build and update company models and test their work. BNY is using Claude with its own AI platform, Eliza, for operational processes. Separately, FIS has made a Claude-powered anti-money-laundering investigations agent commercially available to US banks; it is being deployed at Amalgamated Bank.

Why it matters

These examples put Claude in workflows where speed and handling matter, but human oversight still counts. The FIS agent gathers evidence and explains its reasoning for analysts. FIS says every output is a recommendation, and an analyst must confirm it before a case can be closed. Amalgamated Bank projects the tool will cut investigation time by about 98%, a forecast rather than a measured result.

For banks, the attraction is not simply getting an answer from a model. It is reducing manual sorting and preparation, while keeping people involved in decisions with real consequences. That is a more specific test of enterprise AI than a customer-service demo with a suspiciously tidy script.

Our read

The most convincing detail here is the shape of the work: classifying and routing emails, finding information for staff, and assembling evidence for human investigators. Those are concrete jobs a system can assist with, and the examples come with named organisations and defined tasks.

The 98% figure deserves a pencil mark beside it, not a victory lap. It is a bank’s projection, and the report does not establish that the same result will hold across deployments. For now, the interesting signal is that financial firms are putting Claude into operational workflows, with people still expected to check the consequential calls.

What to watch

  • Whether Barclays expands Claude use among developers as planned.
  • How widely banks adopt the FIS investigations agent beyond its current deployment.
  • Whether measured results support the projected reduction in investigation time.
  • How banks describe human review and accountability as these systems take on more work.

Discussion spark: For bank AI handling emails and compliance work, should the main measure of success be faster processing, or stronger evidence that human oversight catches mistakes?

Sources and evidence

Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.

Your turn

Pull up a chair.

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