Inbox Triage & Auto-Reply Drafting Copilot
An Outlook copilot that classifies, drafts, and learns — giving senior staff ~5 hours a week back.
Before
Senior team members spent 60–90 minutes a day classifying and replying to repetitive stakeholder emails.
After
An Outlook add-in classifies incoming mail by intent, drafts a tone-matched reply, and learns from accept/reject signals. It saved ~5 hours/week per user, reached a 78% draft-acceptance rate, and held 91% triage accuracy across 12 categories.
Senior team members spent 60–90 minutes a day classifying and replying to repetitive stakeholder emails.
Outlook add-in that classifies incoming mail by intent, drafts a tone-matched reply, and learns from accept/reject signals.
- 1
Intent classification
Phase 1Built a 12-category intent classifier for incoming mail — triage first, because routing the email correctly is half the time saved.
- 2
Tone-matched drafting
Phase 2Generated reply drafts that match the user's tone, surfaced in Outlook where they already work rather than a separate app.
- 3
Learn from accept/reject
Phase 3Fed accept/reject signals back in so drafts improved over the first two weeks — acceptance climbed to 78%.
Draft, don't send
Why · Keeping a human in the send loop is what makes an email assistant safe. The copilot proposes; the person approves.
Live inside Outlook
Why · Adoption dies if people have to leave their inbox. An add-in met users where they already are.
- Saved ~5 hours/week per user across the pilot group
- Acceptance rate on drafted replies climbed to 78% after 2 weeks
- Triage accuracy held at 91% on 12-category classification
- 01 · Never auto-send. A tone-matched draft with a human approving is the sweet spot.
- 02 · Meet users in their existing tool or adoption collapses.
- 03 · Accept/reject feedback is a cheap, powerful signal — capture it from day one.
Ask anything about Inbox Triage & Auto-Reply Drafting Copilot
AI scoped to this project · GPT-OSS 120B