n8n Gmail Auto-Reply Workflow
Budget: $250 – $750 CAD
I run VitroPlus Sainte-Foy in Québec and I’m ready to put a full-scale email assistant into production. The workflow has to live in n8n Cloud, connect to my Gmail account, call the OpenAI API, and return a structured JSON payload that keeps every step transparent for future audits.
Here’s what the flow must accomplish from the moment a message arrives:
• Classify the email as customer request, reply, spam, newsletter, platform notification, Promotions, or Invoices.
• Use sender email-address patterns—rather than keyword lists—to decide whether a message is a fresh customer request or simply a reply inside an existing thread.
• Immediately halt if the mail is spam, a newsletter, or needs manual review, so no tokens are wasted.
For true customer requests the assistant must extract the customer’s name, contact details, plus the vehicle make and model. Glass inquiries default to windshield unless stated otherwise, and it must flag any glass-replacement quote that arrives without a VIN so staff can follow up. If insurance is mentioned, the response should remain soft and non-pushy, in line with Canadian insurance norms.
When a reply is required, OpenAI should craft a natural-sounding message in French only—no English lines, no emojis, no mention of AI or automation. The message is saved as a Gmail draft; the system must never auto-send.
Acceptance criteria
1. Clean, modular n8n workflow with comments and node naming that explain each decision branch.
2. All outputs returned as well-formed JSON for downstream logging.
3. A short Markdown or PDF guide covering environment variables, API keys, and how to redeploy or extend the flow.
4. Successful live test on my Gmail inbox demonstrating correct classification, data extraction, and draft creation for at least five different email types (including Promotions and Invoices).
If you’ve built production automations with n8n, Gmail, and OpenAI before, this should feel familiar. Reliability, clarity, and future maintenance matter more to me than flashy extras, so keep the structure tidy and the logic bulletproof.
Here’s what the flow must accomplish from the moment a message arrives:
• Classify the email as customer request, reply, spam, newsletter, platform notification, Promotions, or Invoices.
• Use sender email-address patterns—rather than keyword lists—to decide whether a message is a fresh customer request or simply a reply inside an existing thread.
• Immediately halt if the mail is spam, a newsletter, or needs manual review, so no tokens are wasted.
For true customer requests the assistant must extract the customer’s name, contact details, plus the vehicle make and model. Glass inquiries default to windshield unless stated otherwise, and it must flag any glass-replacement quote that arrives without a VIN so staff can follow up. If insurance is mentioned, the response should remain soft and non-pushy, in line with Canadian insurance norms.
When a reply is required, OpenAI should craft a natural-sounding message in French only—no English lines, no emojis, no mention of AI or automation. The message is saved as a Gmail draft; the system must never auto-send.
Acceptance criteria
1. Clean, modular n8n workflow with comments and node naming that explain each decision branch.
2. All outputs returned as well-formed JSON for downstream logging.
3. A short Markdown or PDF guide covering environment variables, API keys, and how to redeploy or extend the flow.
4. Successful live test on my Gmail inbox demonstrating correct classification, data extraction, and draft creation for at least five different email types (including Promotions and Invoices).
If you’ve built production automations with n8n, Gmail, and OpenAI before, this should feel familiar. Reliability, clarity, and future maintenance matter more to me than flashy extras, so keep the structure tidy and the logic bulletproof.