Client Overview

During an automation audit, we determined the most significant time sink for a client. A service business with a shared inbox was losing time and focus to email triage and repetitive replies. The inbox mixed genuine inquiries (that needed fast, accurate responses) with newsletters, automated notifications, and marketing emails that created noise.

How the Client Approached Prosperus

The client reached out after two operational symptoms became hard to ignore:

• Manual inbox scanning made response times inconsistent.
• At first glance, the team mistook some legitimate inquiries for low-priority messages and replied later than they should have.

They wanted an AI email assistant, but with a strict requirement: no automated sending without human review. The goal was speed and consistency without sacrificing control.

---

The Problem: Email Triage and Drafting Were Bottlenecks

Before implementation, the workflow looked like this:

1. Open inbox
2. Decide what needs a reply
3. Draft response from scratch
4. Send and hope nothing was missed

This created two predictable failure modes:

• Context switching and interruptions throughout the day
• Variable quality and tone, depending on who replied and how busy they were

---

The Solution: AI Drafting With Human Approval

We implemented a two-stage workflow that mirrors a proven “human in the loop” email response pattern:

1. Receive emails
2. Use AI to determine whether an email needs a reply
3. Draft a professional response with AI
4. Request human approval before anything is sent

Why Human-in-the-Loop Was Non-Negotiable

The workflow is designed so AI accelerates triage and drafting, while a human remains accountable for final approval and sending. This reduces the risk of replying to automated emails, misreading intent, or sending inaccurate information.

---

Automation Flow Breakdown

1) Intake: Email Trigger (IMAP)

• Watches the mailbox and ingests the email subject and body text into n8n.
• Keeps the pipeline lightweight by focusing on what the AI needs to classify and draft.

2) AI Triage: “Assess if message needs a reply”

• Prompt includes the subject and message content.
• The model returns only a structured response: { "needsReply": true | false }
• Rule-based guidance in the prompt suppresses noise: marketing, delivery confirmations, and automated system emails do not require a response.

This stage is run deterministically (temperature 0) to keep decisions stable and auditable.

3) Gate: “If Needs Reply”

• If needsReply = true, proceed to drafting.
• If needsReply = false, stop and do nothing. No distractions, no drafts.

4) AI Draft: “Message a model”

• Drafts a reply in a professional, business-casual tone.
• Avoids fabrication and uses placeholders where needed.
• Responds in the same language as the inbound email.

5) Human Review Output

You have two review-friendly output options, both consistent with human-approval-first design:

• Gmail Draft creation so the user can review and send from the existing thread.
• Internal approval email containing the proposed reply for quick review and manual send.

AI drafts, then a human approves and sends the final reply.

---

Implementation Highlights That Matter in Production

Structured outputs for triage

Using a strict schema for needsReply prevents the automation from giving ambiguous outputs and reduces routing errors.

Minimal, explicit prompting for drafting

Clear constraints improve reliability: no special formatting, no invented facts, and language mirroring the inbound email.

Auditability

Each step produces artifacts the team can review: classification output plus the draft response.

---

Results

• Time saved on triage: from ~60 minutes/day to ~25 minutes/day (Microsoft)
• Time saved per reply draft: from ~8 minutes/email to ~3 minutes/email (Talkdesk)
• Median response time improvement: from ~16 hours to ~3 hours. (Superhuman Blog)
• Quality consistency: 30% to 50% fewer major rewrites and 20% to 35% fewer missed-question follow-ups.

The values are estimates based on industry standards.

---

Why This Email Automation Converts

Faster responses without losing control

Prospects want proof that automation improves speed while keeping humans in charge of outbound communication.

Consistency in tone and structure

The workflow enforces a repeatable reply style and reduces variability across team members.

Reduced cognitive load

The team spends less time deciding what matters and more time handling real conversations.

---

FAQ

Torm Erik Raudvee

28.01.2026