AI & data

AI scanned post sorting for accountancy practices

Scanned post sorting reads each letter your practice scans, works out which client it relates to and files it in that client’s folder. It then creates a task with a short summary and any deadline it found, which a person confirms before anyone relies on it.

An example of how the studio would build it. No live services are connected.

From a scanned letter to the right client folder and a task.

  1. Step 01

    Letter scanned

    Incoming post is scanned to a shared folder in Microsoft 365, one letter per file.

  2. Step 02

    Letter read

    Claude identifies the sender, the client, any reference numbers, the type of letter and any dates it mentions.

  3. Step 03

    Client matched

    The client is matched by a reference such as a UTR or company number, and anything without a confident match goes to a review queue.

  4. Step 04

    Letter filed

    The scan is renamed consistently and moved to the client’s folder in SharePoint or OneDrive.

  5. Step 05

    Task created

    A task with a short summary and any deadline is posted to the client’s team in Microsoft Teams, for a person to confirm and act on.

The result

No letter about a client sits unread in a pile.

Deadlines are suggested, never assumed: a person checks each one against the letter. A letter that can’t be matched confidently is never filed in a guessed folder.

Before we connect it.

Where this is useful

Useful for accountancy and bookkeeping practices, and other professional firms, that receive letters from HMRC, Companies House and others on behalf of clients. A sensible first version sorts and summarises the post into a review queue, with a person confirming every filing until the matching has proved reliable.

What we need to connect

  1. 01A scanner or scanning service that saves one letter per file to a folder in Microsoft 365.
  2. 02A client list with the references used for matching, such as UTRs, company numbers and VAT numbers, and your client folder structure.
  3. 03Claude API access with data terms that suit client information, and the Teams channels or task lists each client’s work belongs to.

What needs careful handling

A letter can relate to more than one client, such as a director and their company, so anything that matches several records goes to the review queue. Scans can be skewed or faint, and dates can be misread, so every suggested deadline is shown beside the image of the letter for a person to confirm. Client post contains sensitive financial details, so processing is limited to what each step needs, access follows your existing permissions and copies are not kept outside your Microsoft 365 tenancy.

Questions people ask.

Does HMRC post still arrive on paper?

Some of it does, and the mix varies by client and by tax. The same approach can read PDFs that arrive by email, so it stays useful as more correspondence moves online.

Could it miss a deadline?

It could misread one, which is why it never replaces a person’s check. Every task shows the original letter beside the suggested date, and anything unclear is flagged for a person to read in full.

All automation examples