Email, PEC and triage

A shared inbox — info@, accounts@, the certified-email box — is where work enters a business, and it is almost always handled by hand: someone opens, understands, forwards, retypes, replies. As volume grows, that is where the queues form, the lost requests, the "I thought you had taken that one". It is a process that automates very well — provided you decide seriously what the system may do on its own, and what it may not.

What the system does

• It classifies. Each incoming message is recognised for what it is — an order, a complaint, a request for a quotation, a reminder, an invoice, a formal notice — and routed to the right person or queue, at the right priority.
• It extracts. From the body and the attachments come the data needed to work the request: who is writing, on whose behalf, on which matter, with what deadline. Linking to the existing file happens automatically when the reference is there.
• It prepares. For recurring categories the system proposes a draft reply, built on your own wording and on the data in the file. The draft waits for a person: correcting it takes seconds, where writing it from scratch took minutes.
• It records. Every message has a tracked outcome: when it arrived, how it was classified, who took it on, when it was closed. That is the end of evaporating messages — and the beginning of real numbers on response times.

Certified email is a case apart

On PEC — the Italian certified-email channel — the automation has to be designed under one extra constraint: what arrives there includes notifications, formal demands, served documents — messages with legal effect and deadlines that start running on delivery. So the system helps where help carries no risk: it registers, classifies, extracts deadlines into the right calendars, and alerts immediately whoever must see a given class of documents at once. It does not reply, does not discard, does not decide alone what matters: it proposes, and a person disposes. Acceptance and delivery receipts are preserved intact alongside the message, because the receipts are the evidence — a subject I also know from the forensic side.

Where it starts

With your real traffic: the mailbox is taken as it is and the system runs read-only for a period, its classifications compared against the ones people make by hand. When the numbers on your actual traffic — not on a laboratory sample — show the sorting holds, routing is switched on, and draft replies come last. It is the same principle as the whole section: measure first, trust after.

Frequently asked questions

Does the system reply to clients on its own?

Not by default. The system classifies, extracts the data and prepares a draft; a person sends it, and may correct or rewrite it first. Automatic sending is enabled — if at all — only for narrow, low-risk categories such as acknowledgements of receipt, and that choice is written into the project scope.

Does this apply to PEC (Italian certified email) too?

For reading and sorting, yes, with one extra caution: PEC carries documents with legal effect and deadlines that start running on delivery. The system can therefore classify, register, extract deadlines and alert the right person immediately — but it never replies on its own, and the configuration treats acceptance and delivery receipts as records to preserve, not summarise.

Does it work with our current email?

With the common systems — Microsoft 365, Google Workspace, ordinary IMAP mailboxes and the main Italian PEC providers — yes, through their standard interfaces. The mailbox stays where it is and remains yours: the system reads and organises it, it does not replace it.

What about the confidentiality of the messages?

The framework described on the section's main page applies: where messages may travel is decided first, model providers are used in configurations where your data does not train their systems, and for the most sensitive mailboxes the analysis can run entirely on your own machines. Every read and every classification is logged.