AI use case

Automatic Document Classification and Routing

Automatically classify incoming documents and route them to the right processing workflows without manual intervention.

  • Executive
  • Finance
  • Human Resources
  • Legal
  • Operations
A shared inbox receives invoices, complaints, cancellation notices and scanned HR letters every day, all mixed together, each one needing to be recognised before anyone knows who should handle it. Fasfox designs solutions that read the content of these documents as they arrive, match them to the right category, and route them to the right team or system.

Post arrives before anyone knows what it is

A shared inbox receives invoices, complaints, cancellation notices and scanned HR letters, all mixed together. Someone opens each attachment to work out what it is before forwarding it by email to the right team, with no shared rule: what counts as “urgent” or “route to finance” depends on whoever is sorting that day, and the volume leaves no time to apply a consistent standard. The person doing the sorting is rarely the one who deals with the consequences of a misrouted document further down the line.

Reliable on the clean cases, shaky on the rest

An early sorting tool performs well on short, structured emails, tuned on a batch of clean examples picked for the demo. It struggles with long forwarded chains, attachments that belong to two categories at once, or scanned letters where the text recognition is poor. The system then produces a middling confidence score that nobody actually checks, because no one was assigned to own it.

A stable list of categories, as a starting condition

Automatic sorting assumes a finite, shared list of target categories, plus an exception queue with a named owner for ambiguous cases. Agreeing that list and naming that person is an organisational decision to make before the tool is built, not a technical setting adjusted afterwards. Skip this step and the exception queue simply fills up unread.

Where the system hands back to a person

We build systems that read the content and attachments, match them to a category, and extract the details needed downstream. Clear-cut cases route straight to the right tool or team. Ambiguous ones land in a visible triage queue, along with the score and the signals behind the uncertainty, so a person can decide with the full picture.