n8n Workflow: Log AI Refund Decisions as Execution Data

Log AI refund decisions as Execution Data—so you can audit every run in n8n

This n8n workflow receives a refund request, uses a demo refund policy to choose refund_approved, refund_declined, or needs_human_review, asks an AI agent to explain the reasoning, and then saves the full case and final outcome directly into custom execution data for later search and inspection.

What this workflow does

• Accepts refund input via Webhook: handles a POST request with orderId, customerTier, and message.
• Looks up the order (simulated): returns a demo order record for this exercise.
• Selects a policy decision: uses a “Policy Decision” step with an Edit Fields node to set refund_approved, refund_declined, or needs_human_review.
• Prevents oversized execution tag values: checks tag length; if values exceed 255 characters, it routes to “Reject Oversized Tag,” then uses a Stop and Error node.
• Saves case details to Execution Data: a dedicated Execution Data node stores the case for traceability.
• Explains the decision with an AI agent: “Refund Support Agent” generates an explanation using an OpenRouter chat model.
• Stores the final decision as custom execution data: “Record Decision” writes the outcome so you can open the saved execution later and filter by custom data (n8n Cloud Pro/Enterprise required for advanced filtering).

Use cases

• Auditing customer support refund decisions across environments.
• Building traceable AI workflows for SaaS operations and incident review.
• Training your team on audit & trace patterns from the n8n Production AI Playbook (Audit & Trace, Exercise 2).

Technical details

• Trigger: Webhook (POST)
• Logic/controls: If, Set, Code
• Guardrails: Stop and Error for oversized tag values
• AI integration: OpenRouter (template uses openai/gpt-4.1-mini, but any supported chat model works)
• Traceability: Execution Data + custom execution fields