Trace Refund Approvals with OpenRouter & OpenTelemetry in n8n

Trace refund approvals end-to-end in n8n—using OpenRouter + OpenTelemetry

This workflow enriches a refund request, drafts an approval proposal with an OpenRouter AI agent, and then pauses for a human decision—while keeping both the initial run and the approval step connected via OpenTelemetry tracing.

What this workflow does

• Receives an approval request via a POST webhook with requestId, amount (in cents), reason, and an incoming traceparent header.
• Enriches and validates the request in the Enrich Request sub-workflow (including simulated customer context).
• Calls a Policy Service using an HTTP request, carrying the trace context to an echo service.
• Drafts an approval proposal with the Approval Proposal Agent using an OpenRouter chat model.
• Returns the Approval Request with the proposal, executionId, and a resumeUrl for the next step.
• Waits for Human Approval, pausing the workflow until a decision is posted to the resumeUrl.
• Validates Approval and records the decision. Invalid decisions stop at Reject Approval.

Use cases

• Audit and traceable refund handling for SaaS support teams with human-in-the-loop approvals.
• Testing and demonstrating production-grade observability in n8n workflows using OpenTelemetry.
• Ensuring AI-generated approval proposals (via OpenRouter) are tied to the same trace across “wait” boundaries.

Technical details

• Nodes/steps: webhook, if, set, http request, wait, and support nodes like sticky note.
• OpenRouter integration: OpenRouter - Approval credential for the proposal agent.
• OpenTelemetry: optionally enable in Settings > OpenTelemetry and search traces by n8n.execution.id to find both run parts.

From the n8n Production AI Playbook on Audit & Trace (Exercise 5), this workflow helps you follow a single execution through approval—without losing observability.