Peyto Lake seen from above in Banff National Park, Alberta

Workflow Automation · Claude Fable 5.1

Claude Fable 5.1 removed 1,900 documents a month from the back office

A property management group

All case studies

Results

Measured against the case agreed at the start.

1,900

documents handled per month

98.1%

straight-through processing

-83%

reconciliation exceptions

Models used

Built on the current generation, chosen on evidence.

Claude Fable 5.1

Every model is benchmarked against the client's own historical work before it is trusted with any of it, and it runs on their data inside their controls.

The automation

How the flow actually runs.

Each block is a step in the live automation, from the event that starts it to the record it writes at the end.

End-to-end automation · Document in, ledger entry out Trigger Action Decision Human Output
  1. TriggerDocument arrives

    Invoice, lease or contractor report, in any of nine formats.

    Inbox / scan
  2. ActionRead and structure it

    Fable 5.1 extracts fields with a confidence score on each one.

    Claude Fable 5.1
  3. ActionMatch to the record

    Three-way match against purchase order, property and approved work.

    Ledger
  4. DecisionConfidence above threshold?

    Anything uncertain or unmatched routes to a person instead of posting.

    Rules
  5. HumanException review

    Roughly two documents in a hundred reach a human, with the source page beside them.

    Human
  6. OutputPosted and filed

    Ledger entry written, document filed, audit trail complete.

    Finance system

Every step is visible and editable by your own team - nothing in the chain is a black box.

The challenge

What the operation looked like before.

Four back-office staff processed roughly two thousand documents a month across nine formats. Errors surfaced weeks later at reconciliation, and no one could say with confidence what had been approved by whom.

What we did

The work, in order.

  • Tested Fable 5.1 against the incumbent process on a month of real documents before switching anything on.
  • Built extraction, matching and approval as one end-to-end flow with every step visible.
  • Set explicit confidence thresholds so uncertain documents route to a human rather than guessing.
  • Wrote a full audit record of every extraction, match and approval.
“The team went from typing documents to checking the handful that genuinely need judgement.”
Finance Manager, a property management group

Book a working session.

Ninety minutes on your operation with people who have done this work before. You leave with the first three things worth fixing, whether or not you hire us.

Start a conversation