Building an AI email triage system that saved us 10 hours a week
We were drowning in client emails. So we built an AI system to categorize, prioritize, and draft responses automatically.
As AXI grew, so did our inbox. Between client requests, feedback rounds, vendor communications, and new inquiries, we were spending over two hours a day just triaging email. That is time we could have spent designing, building, or thinking strategically. So we automated it.
How the System Works
The system works in three stages. First, every incoming email is analyzed by Claude to extract the sender, intent, urgency, and associated project. Second, it gets automatically categorized and routed: new business inquiries go to our sales pipeline, client feedback gets attached to the relevant project in Linear, and vendor emails get filed for weekly review.
Third — and this is the part that saves the most time — the system drafts responses for routine emails. Scheduling requests, status update inquiries, and simple acknowledgments all get pre-written replies that we can approve with a single click or quickly edit before sending. About 60% of our outbound email is now AI-drafted.
Building It
We built this with a combination of Gmail API, Claude API, and a simple Next.js app that serves as our review dashboard. The total development time was about three days, and the monthly API costs run under $50. The ROI was positive within the first week.
The unexpected benefit: consistency. Every client now gets a prompt, professional response regardless of how busy we are. Before the system, response times varied wildly depending on the day. Now, 95% of emails get at least an acknowledgment within 30 minutes, even on our busiest days.
Share this article
