The Problem

Texas Elite Roofing was losing leads every day. Inbound calls went unanswered after hours. Follow-up emails were sent days late. Sales reps spent hours manually calling cold leads — most weren’t ready to book. There was no system, no consistency, and no way to scale without hiring more staff.

The client needed a solution that could work around the clock, sound human, handle objections, book appointments into Google Calendar, and log everything to a CRM — without adding headcount.

The Solution

I built Alex — a fully autonomous AI Sales Development Representative running on n8n Cloud, powered by Claude Sonnet as the LLM, with Retell AI handling the voice layer. Alex operates across 5 channels simultaneously and makes intelligent decisions at every step of the sales process.

Alex in Retell AI

Alex was configured in Retell AI with Claude 4.6 Sonnet as the LLM, Leland as the voice, and a comprehensive sales system prompt covering BANT qualification, objection handling, and appointment booking. Connected to a central inbound webhook so all calls route through a smart phone router.

Alex configured in Retell AI — Claude 4.6 Sonnet + Leland voice + full sales system prompt

Phone routing — Emily as inbound agent, Parker for outbound, voice-inbound webhook active

Tech Stack

The Main AI Sales Agent Workflow

The core workflow handles all 5 input channels — voice, chat, email, schedule, and Teams — normalizing them into a single pipeline before routing through Alex’s AI brain, CRM lookup, RAG knowledge retrieval, and post-call action branches.

Alex — AI Sales Agent workflow: 5 input channels → CRM lookup → Claude AI Agent → booking / escalation / follow-up

Model Abstraction Layer

A dedicated sub-workflow dynamically fetches the active LLM config from the CRM. This allows swapping between Claude, GPT-4, or Groq without touching the main workflow — just update a single row in Google Sheets.

00 — Alex Model Abstraction Layer: fetches active model config so the LLM can be swapped without any code changes

Outbound Calling Engine

Alex doesn’t just wait for calls — he makes them. Every hour during business hours (9AM–7PM Dallas), Alex pulls prioritized leads from Google Sheets, builds a personalized call script using dynamic variables, dials automatically via Retell AI, and routes into the correct follow-up sequence based on the outcome.

Alex Outbound Calling — hourly trigger → lead prioritization → Retell AI call → outcome analysis → automated follow-up emails

Knowledge Base Upload

A one-time setup workflow converts all roofing knowledge into vector embeddings using Google Gemini and stores them in Supabase. Alex queries this on every call to provide accurate answers about pricing, services, warranties, and insurance processes.

Upload Roofing Knowledge Base — Manual Trigger → format docs → Google Gemini embeddings → Supabase vector store

CRM — Google Sheets

Every lead interaction is logged to a 6-tab Google Sheets CRM: Leads, Interactions, Appointments, Proposals, Escalations, and Dashboard. Alex creates, reads, and updates records automatically — full pipeline visibility with zero manual data entry.

Alex CRM Logs — 6-tab Google Sheets tracking every lead from first contact through appointment booking

How Alex Works

Results & Capabilities