Document classification: from one mailbox with everything, to one process per document

The reality of shared mailboxes

At many mid-sized organizations the back office runs through a handful of shared mailboxes. One for procurement, one for sales, one for customer service. Everything comes in through those mailboxes: orders, invoices, complaints, internal email, newsletters, quotes, service tickets, sometimes even personal correspondence.

Someone, often an experienced colleague, reads every email, figures out what it is, and forwards it to the right handler or takes the first steps themselves. At a few hundred emails a day that work becomes invisible but dominant; the best person on the team spends a big chunk of the day on triage.

What an AI coworker does here

The classification AI coworker reads every incoming email and every attachment. Based on content, structure and context it decides the document type, the urgency and the right downstream process. From there one of three things happens:

• The document goes to another AI coworker that runs the specific process (order processing, invoice processing, service tickets).
• The document is routed to the right person or department, with a summary of what it is and what likely needs to happen with it.
• The document is archived or ignored, internal email, marketing, duplicate receipts, with an audit record so nothing gets lost by accident.

Where it's sensitive

Classification looks simple, but the cost of error is asymmetric. An invoice mistaken for a quote costs time. A complaint lost in the general stream costs a customer. That's why the AI coworker isn't only aimed at high precision, but also at deliberately visible doubt: if the document type can't be determined with enough confidence, it comes to a person with a proposal.

That's what makes the system reliable: not because it never doubts, but because the doubt is handled cleanly.

A concrete example

At a large home-goods retailer our AI coworkers together process around 600,000 documents a year, spread across five processes. The starting point is classification: one incoming stream, dozens of document types. After classification each document moves to the matching process, order processing, invoice processing, complaints, tickets, each with its own validations and actions.

Projected annual net saving: €150,000 to €185,000. Not from the classification step alone, but because every downstream process gets automated too. Classification is the invisible building block that makes the rest possible.

Where you start

A classification engagement starts with an audit of one or two critical mailboxes. Which document types come in? At what volumes? Which get the right attention today, which fall through the cracks? From there we define the classification rules and decide which process gets hooked in first. For most customers the first connected process is also the largest volume, that's where the business case sits.