Most inbound forms die in a CRM. A human is supposed to follow up inside five minutes. In practice it takes two days, the lead has gone cold, and the meeting never books.
This is the entire GTM workflow that closes that gap, exported from a live n8n build and free to import. Not a diagram of one. 149 nodes, 11 AI agents, 11 integrations, every prompt left in place. You import the JSON, connect your own credentials, and it runs from form submission to booked meeting.
What you get
- The full workflow export (
gtm-system-workflow.json), importable into any cloud or self-hosted n8n instance. - 11 AI agent nodes with the prompts already written, covering research, email bodies, subject lines, call analysis, and reply handling.
- A Pinecone knowledge base wired to a Google Drive doc, so every message is grounded in your own offer and case studies instead of a generic template.
- 29 Google Sheets nodes that turn one spreadsheet into both the system of record and the reporting layer.
- A 10-section implementation guide: setup, deployment, optimization, troubleshooting, and customization.
How the workflow actually runs
1. Lead intake. An n8n form titled Lead Capture & Analysis collects seven fields. Full Name, Email, and Company are required. Job Title, Company Website, Pain Points, and Lead Source are optional and feed the research layer when present. A parallel webhook accepts submissions from any external form, so your existing site form works without a rebuild.
2. Email verification. Every address hits Reoon before anything else fires. An IF gate splits valid from invalid, and a Slack alert flags bounce risk. A dead address never reaches the sender, so your domain reputation survives the volume.
3. Research, from four sources in parallel. This is the layer most builds skip.
vdrmota/contact-info-scraperpulls company contact dataapify/website-content-crawlerreads the site copy, the recent wins, and the tone of voicedev_fusion/linkedin-profile-scraperpulls the profilesupreme_coder/linkedin-postpulls their recent posts
A Gemini 2.0 Flash agent compresses all four into a single research brief that every downstream message reads from.
4. Opener assignment. A switch node splits every lead randomly across three different call openers. You are A/B/C testing your scripts by construction, not by intention, and the results land in the sheet.
5. The voice call. Retell AI places the call. A webhook catches the transcript, the recording URL, and the combined call cost. Then gpt-4.1 classifiers do the work a human usually does listening back: human or voicemail, booked or rejected or later, the reason for the rejection, and the meeting date and time pulled straight out of the transcript.
6. Staged email follow-ups. If the call is not picked up, goes to voicemail, or books nothing within 24 hours, the email sequence takes over. Three emails, spaced T+0, T+4 days, and T+8 days, driven by a Stage field and held apart by Wait nodes. gpt-4.1 writes the body and the subject line separately, both retrieving from the Pinecone knowledge base, so email three still knows what email one said. Each one comes from a different angle: the call recap, the pain point, then the social proof.
7. Back and forth. A Gmail trigger watches for replies. The workflow pulls the full thread before answering, then replies in-thread. When a lead replies or books, the sequence cancels itself instead of firing the next scheduled step at someone who already said yes.
The stack
n8n, Retell AI, Reoon, Apify, Pinecone, OpenAI, Google Gemini, Gmail, Google Sheets, Google Drive, Slack.
Model routing is deliberate. gpt-4.1 does the writing and classification across 16 nodes, gpt-4o-mini handles cheap parsing across 8, and gemini-2.0-flash-001 does the web extraction.
What it tracks
Every run writes back to one sheet: calls sent, cost incurred, human calls picked up, voicemails, appointments booked, offers rejected, recording URL, call summary, last email sent, latest thread ID, and the raw back-and-forth of every thread.
Cost per booked meeting becomes arithmetic instead of a guess.
Who this is for
Founders and revenue leads with inbound flow they cannot keep up with, who are comfortable working inside n8n. If leads are slipping because the human in the loop is the bottleneck, this is the layer you remove.
You will need your own paid keys for Retell, Apify, Reoon, OpenAI, and Pinecone. The workflow is free. The infrastructure it calls is metered by the vendors.
Open it below.