Results at a Glance
- 93% reduction in email drafting time (30-60 min → under 5 min)
- 80%+ first-pass approval rate on AI-generated emails
- 20+ hours/month saved across the team
- 100% adoption: all post-call emails now go through the system
Background
A health coaching company with a 4-person team doing $23K/month in revenue. The founder and primary coach was the bottleneck for every client communication. After each coaching call, someone had to manually write a personalized follow-up email referencing specific details from the conversation. The emails needed to sound like the founder, not like a template.
The Problem
Every coaching call generated 30-60 minutes of email work. The team was already using Fireflies to transcribe calls, but someone still had to read through the transcript, pull out the key details, and manually draft a personalized follow-up in the founder’s voice.
The real cost wasn’t just time. It was the ceiling. They couldn’t take on more clients without either hiring or accepting that follow-ups would get worse. Neither option worked for a lean team with $20K/month in operating costs.
What I Built
An AI-powered email generation system that turns call transcripts into personalized, voice-matched emails in under 5 minutes:
- Transcript intake: Pulls call transcripts automatically from Fireflies
- Voice-matched drafting: Generates follow-up emails in the founder’s authentic voice (trained on 50+ real emails)
- Review workflow: Routes drafts through a structured pipeline in Notion
- One-click send: Creates a ready-to-send Gmail draft after approval
The founder opens Gmail, reads for 2 minutes, maybe tweaks a word, and hits send.
What Made This Different from Generic AI
The emails needed to sound like a specific human, not like ChatGPT. I built a voice profile by analyzing 50+ of the founder’s actual emails: personal pronouns, concrete call references, warm-but-direct tone, zero corporate language.
I also created quality standards that catch common AI tells before they reach the inbox. No em-dashes. No “I’m thrilled to.” No forced inspirational closings. The system checks its own output against these standards.
The Bigger Picture
This was the first automation in a larger engagement. The same voice-matching approach was applied to marketing content (blog posts, emails, social media). The team went from one person being the content bottleneck to three team members creating consistently on-brand material independently.
The business is scaling from $23K/month toward $100K/month without adding headcount. This automation is one piece of how a 4-person team handles the workload of a much larger operation.
Technical Stack
- AI: Claude (via Claude Teams Projects with custom instructions)
- Call Transcription: Fireflies.ai
- Voice Profile: Custom-built from 50+ real emails, loaded live via MCP
- Review Workflow: Notion database (draft → review → approved pipeline)
- Automation: Zapier (Notion status change → Gmail draft creation)
- Calendar Context: Google Calendar API for meeting details
Interested in automating repetitive work in your business?