Full-Stack AI Lead Engine: 2 Voice Agents + SMS Text Engine + DB Reactivation + 5-System Booking Chain

The Problem

Goodfellas AC was losing HVAC leads to competitors who called faster. Web form submissions sat uncontacted for hours. Inbound calls during peak season went unanswered. A database of thousands of dormant contacts had never been re-engaged. Every qualification call required a dispatcher manually walking through 5 data points before ServiceTitan could be touched — capping throughput and creating a bottleneck. After-hours calls hit voicemail and were never recovered. No call analytics existed: no sentiment tracking, no routing decisions logged, no visibility into why leads didn't convert.

The Solution

Built six interconnected systems on self-hosted n8n (Coolify/VPS). (1) Outbound Kate: GHL form → n8n normalises phone to E.164 → Retell AI call fires within minutes using a 5-question qualification script (issue type, unit type, duration, urgency, preferred day) with 4-bucket routing (billing/warranty, frustrated customers, general CS → immediate human transfer; service requests → qualification). (2) 5-System Booking Chain: Retell → n8n → GHL (find/create contact, book natively) → Zapier webhook → ServiceTitan (HomeBuddy provider, non-job appointment in Unassigned status with full dispatcher notification) — completes in under 10 seconds. (3) Inbound Kate: same 4-bucket routing and qualification flow, 24/7, with `after_hours_flag` variable changing greeting and downstream routing. (4) Post-Call Analytics: every call logged to PostgreSQL with 15+ fields — duration, recording URL, customer sentiment, routing category, issue type, unit type, urgency, booking status, disconnection reason — with 7-branch GHL follow-up routing on outcome. (5) Text Engine: 4 n8n workflows pulling 60-message conversation history from PostgreSQL, routing to 4 AI agent types (Website, Engagement, Follow-up, KB-only) with 6 booking functions (get slots, book, get/cancel/reschedule appointments, timezone check) — KB stored as vector embeddings in Supabase. (6) DB Reactivation: CSV upload triggers a 26-hour per-lead sequence: SMS → AI call 10 minutes later → SMS #2 two hours post-call → email at 26 hours; any reply exits the sequence.

Tech Stack

Results

• Zero dispatcher time on lead qualification — every booking arrives in ServiceTitan Unassigned queue fully pre-qualified with all 5 data points, service address, and contact record already created
• Inbound and outbound calls handled 24/7 with after-hours detection changing agent greeting and routing flags automatically
• 4,000+ dormant contacts cycled through a structured 26-hour multi-touch sequence: SMS → AI call → SMS → email, with any reply triggering immediate exit
• 5-system booking chain (Retell → n8n → GHL → Zapier → ServiceTitan) completing in under 10 seconds with 2-minute timeout fallback to direct booking link
• 15+ fields of call intelligence logged to PostgreSQL per call — sentiment, routing category, issue type, unit type, urgency, booking status, recording URL
• Text engine routing conversations to 4 specialist AI agents with 60-message memory from PostgreSQL and 6 live booking functions including slot retrieval and cancellation