The business inbox, finally under control.
Shared inboxes (info@, orders@, hr@, support@) are the black hole of Italian companies. An AI agent classifies incoming email, routes to the right person, directly replies to first-line requests, and keeps a structured log.
What happens today.
The shared inbox is a black hole: hundreds of emails a day, a mix of customer requests, suppliers, spam, internal comms. Replies are picked up late, some get lost, priority requests dilute into ordinary ones.
Manual routing breaks down past a certain scale. Building rules with static filters (subject contains X, sender is Y) fails on the first off-pattern case. Modern AI instead classifies by intent and content, not fragile keywords.
The inbox isn't a black hole. It's a structured workflow no one has had time to design yet.
The solution, broken into parts.
-
Intent-based classification
The agent reads every incoming email, identifies the intent (info request, order, complaint, supplier, spam, etc.), the implicit priority, and the natural recipient.
-
Routing + automatic reply
Emails are routed to the right recipient (person, team, or category). First-line ones (FAQs, confirmations, routine follow-ups) are answered directly by the agent in your tone of voice.
-
Human hand-off for sensitive cases
Complaints, low-confidence cases, requested escalations are forwarded to a person with pre-classified context. The human receives a pre-digested email, not a wall of text.
The typical profiles who benefit.
-
SMEs with overloaded shared inboxes
info@, orders@, support@, hr@: more than 100 emails a day, shared management, variable reply quality. AI standardizes quality and cuts time-to-response.
-
Multi-client professional firms
Accounting, legal, consulting firms: email arrives from different clients, on different matters, with different deadlines. AI routes by client/matter and prioritizes by deadline.
Transparency on what the client does.
Before we start we need a few accesses and decisions. All reasonable, no surprise asks.
-
Access and data
- IMAP/API access to shared inboxes (Microsoft 365, Google Workspace, IMAP)
- 500-2,000 historical emails with sample classification (for initial training)
-
Operational decisions
- Taxonomy of routing categories/intents
- Auto-reply vs human hand-off rules
- Tone of voice for first-line replies
Indicative numbers, not quotes.
- TIME
- Typically 4-8 weeks from discovery to go-live. Simple cases (routing only, no auto-replies) in 3-5 weeks.
- COST
- Range €12,000-35,000 for the initial version. Operating LLM cost €100-600/month depending on volume.
- MODEL
- Fixed milestones. Soft launch on a percentage of email, progressive scaling.
Indicative numbers. For an accurate quote, let's talk.
Answers to the most common questions.
Does the agent read every email? And the privacy?
Yes, the agent reads incoming email in the configured inboxes. Email is processed with in-transit and at-rest encryption, data is not used for training (enterprise providers: OpenAI, Anthropic, Mistral with no-training options), and the log is structured and auditable. Your DPO can review the entire data flow.
What happens if the agent classifies or replies wrong?
The system learns from corrections. Every email can be reclassified or re-answered by a human, and the agent records the correction to improve. Monitoring identifies error patterns. Auto-replies start in soft launch on a percentage of email, scaling only after validation.
Does it support certified email (PEC)?
Yes, it supports Italian PEC (Aruba PEC, Legalmail, PEC.it, etc.) via IMAP. Important: PEC has legal implications (registered-letter value), so auto-replies on PEC are enabled only after case-by-case review, and never for sensitive categories (legal notices, tax deadlines).
Recognize your case?
Write a couple of lines about your context. We'll reply within 24-48 hours with an initial assessment and a first orientation on time and cost.
Let's talk