I Automated 60% of an Executive's Emails — Here's the Exact Setup

100 emails a day — 3 hours of triage

An executive I worked with was drowning in email. 100+ emails daily, 3+ hours just reading and categorizing. Scheduling requests, information queries, follow-ups — all competing for the same inbox.

The constraint

The executive refused to use an AI that sends emails without review. They wanted to see every outgoing message, but they didn't want to write every response from scratch.

The architecture

I built a pipeline:

1. **Ingestion** — emails fetched via IMAP every 5 minutes

2. **Categorization** — OpenAI classifies each email: urgent, scheduling, routine, FYI, spam

3. **Drafting** — For scheduling and routine emails, AI generates a draft using custom prompt engineered for the executive's tone

4. **Review queue** — Drafts appear in a dashboard where the executive can approve, edit, or reject

5. **Learning** — Rejected drafts with corrections are stored to improve future prompts

Prompt engineering detail

The key was the tone prompt:

*"You are drafting responses for [Name], an executive in [Industry]. The tone should be professional but warm. Never use jargon. Keep responses under 150 words. For scheduling requests, default to offering 3 specific time slots based on calendar availability."*

Results

The cost

OpenAI API costs: ~₹3,000/month ($35). The time savings alone made it 10x ROI in month one.