Lead Automation System
An end-to-end AI-powered lead conversion pipeline built for a US-based HVAC company receiving 40–60 leads per day. The system captures, qualifies, routes, responds, and follows up — entirely without human intervention.
Form Submission to Follow-Up, Automatically
The moment a lead submits a Tally form, the pipeline fires. A webhook triggers n8n, which passes the lead to Groq's LLaMA3 model for instant qualification — generating an urgency score from 1–10, a lead quality rating, sentiment analysis, and a plain-language summary. The entire qualification cycle completes in under a second.
High-urgency leads — scored 7 or above — are immediately escalated to the sales team via Telegram and sent a priority email. Everyone else receives an AI-written, contextually personalised response. Every lead, regardless of path, is logged to Google Sheets as a lightweight CRM entry.
A System That Knows When to Stop
Two follow-up workflows run on a 15-minute schedule, scanning the sheet for leads that haven't booked and haven't been followed up yet. A 1-hour email goes first. A 24-hour email goes second. The moment a lead books via Calendly, a separate workflow updates their sheet status — and the follow-ups stop automatically.
No manual tracking. No missed leads. No duplicate emails.
A Foundation Ready for Production
The current build handles the full lead lifecycle cleanly — capture, qualify, route, respond, follow up, and close. The known constraints are practical ones: n8n's free plan processes leads sequentially, follow-up timing is approximate to 15 minutes, and the AI has no memory of prior interactions with the same lead. Each of these has a clear upgrade path — queue mode, SendGrid open tracking, a fine-tuned scoring model — none of which require rebuilding the core architecture.
What This Project Actually Demonstrates
This isn't a toy automation. It was designed around a real business problem — a company losing leads to slow response times and inconsistent follow-up — and built to solve it end to end. The stack is deliberately unglamorous: no custom code where a webhook works, no database where a spreadsheet is enough. The goal was a system a small business could actually run, not a portfolio piece that only works in a demo.
