Wynter Comfort (Internal)
AI Client Acquisition Agent
Tech Stack
| Layer | Technology |
|---|---|
| Runtime | Node.js 18+ (ESM modules) |
| AI Brain | Anthropic Claude API (Sonnet) for qualifying + copywriting |
| Prospecting | Google Maps Places API (local business discovery) |
| Website Scoring | Google PageSpeed Insights API + load-time fallback |
| Email Finding | Playwright (contact page scraping) + Yelp scraping |
| Email Sending | Gmail API via Google Workspace OAuth2 |
| Social DMs | Playwright (LinkedIn, Instagram) with cookie-based auth |
| Logging | Google Sheets API (full prospect + outreach tracking) |
| Scheduling | node-cron (prospecting, follow-ups, reply detection) |
| Dedup | Local JSON persistence (.prospected.json) |
The Problem
As a solo developer, I was spending hours every week manually searching for potential clients, researching their businesses, writing personalized outreach emails, and tracking who I'd contacted. The process was effective but didn't scale. I needed a system that could prospect, qualify, write, and send outreach on my behalf while I focused on building. The challenge was making AI-written outreach actually sound like me, not like a template or a chatbot.
The Approach
I designed a 7-module pipeline that mirrors exactly how I prospect manually, but runs autonomously. The system searches Google Maps across multiple cities and niches, scores each business's website using PageSpeed Insights, then passes qualified prospects to Claude with my complete brand voice framework (built on Human Design principles) so every message sounds like I actually wrote it. The entire pipeline logs to Google Sheets so I have full visibility into who was contacted, what was said, and when to follow up.
The Solution
The agent runs on a cron schedule, prospecting twice daily across 6 Michigan cities and 17 business niches. It finds local businesses with underperforming websites (scoring under 70/100), enriches them with contact information scraped from their sites, then qualifies each one through Claude with specific disqualification rules (too large, too small, already has a dev team). Qualified prospects get a personalized cold email written by Claude using my exact voice, tone rules, and messaging formulas. The system handles 3-day and 7-day follow-ups automatically, monitors Gmail for replies, and only notifies me when a prospect responds. Every touchpoint is logged to Google Sheets with the full message, prospect details, and status.
The Results
The system replaced what used to take 10+ hours per week of manual prospecting and outreach. In its first live run, it identified 209 qualified local businesses, wrote personalized outreach for each, and sent 20 emails (the daily cap) in a single cycle. The outreach quality matches what I'd write manually because the AI prompt is built on my actual voice framework, not generic templates. I now only open my laptop for client acquisition when someone has already replied.
Key Features
7-module autonomous pipeline: Prospect, Enrich, Qualify, Write, Send, Follow-up, Notify
Google Maps API integration scanning 6 cities across 17 local business niches
PageSpeed Insights website scoring with load-time fallback for faster analysis
Claude AI qualifier that scores prospects 1-10 with specific disqualification logic
Custom brand voice framework built on Human Design principles (Generator type, 5 Gate energies)
Three outreach formulas: The Reframe, The Conviction Drop, The Quiet Authority
Contact email discovery via website scraping, Yelp listings, and Hunter.io integration
Gmail API cold email sending with rate limiting and daily caps
Automated 3-day and 7-day follow-up sequences for email prospects
Reply detection that only flags responses from contacted prospects (not all inbox)
Cookie-based authentication for LinkedIn and Instagram DM prospecting
Full Google Sheets logging with prospect data, outreach copy, status, and notes
Key Outcomes
200+ per cycle
Prospects Found
20 personalized emails
Daily Outreach
10+ per week
Hours Saved
7-stage pipeline
Modules
6 Michigan markets
Cities Covered
17 business categories
Niches Scanned