Route Approval Requests: GPT-4o-mini + Supabase + Gmail

Automatically route approval requests with AI risk scoring—and one-click approvals

This n8n workflow captures incoming approval requests from an n8n form, uses GPT-4o-mini to summarize and score risk, logs everything in Supabase, emails the correct approver through Gmail with approve/reject links, and finally notifies the requester once a decision is made.

What this workflow does

• Receives a new request from an n8n form (type, title, amount, department, justification, and requester email).
• Summarizes and scores risk with OpenAI (GPT-4o-mini), producing a short summary plus a risk score, priority, and suggested approval level.
• Selects the approver based on amount thresholds and the AI risk score, then merges the AI output with the original form data.
• Stores the request in Supabase by creating a new row in the approval_requests table, including the request details, AI fields, and a generated request ID set to pending.
• Sends decision emails via Gmail to the approver, including one-click approve/reject URLs that point to a webhook.
• Handles the approver’s action: when the approver clicks, the workflow updates the Supabase record to approved or rejected with audit details, then emails the requester the final outcome via Gmail.

Use cases

• Route purchase or expense approvals to the right manager based on amount and AI risk.
• Standardize approvals for departments (e.g., Finance, Engineering) using consistent risk summaries.
• Provide approvers with a fast, auditable approval process using email-based links.

Technical details

• OpenAI: GPT-4o-mini for summary, risk score, priority, and suggested approval level.
• Supabase: writes to approval_requests and updates status/audit details.
• Gmail: emails approvers with approve/reject links and notifies requesters of results.
• Webhooks: webhook endpoints to receive approve/reject clicks.
• n8n logic nodes: if, set, code, plus supporting flow control.