Return and RMA request triage
Sort incoming return requests by reason, eligibility, and urgency, then draft a policy-aligned reply and RMA note for a person to approve.
The use case
Return requests arrive by email, contact form, and helpdesk, each written differently and often missing the order number. Someone has to look up the order, check the return window, read the reason, and decide whether it is a simple return, a warranty claim, or a damaged-in-transit case that needs photos. This agent gathers the order details, compares them with your written return policy, sorts each request into a clear bucket, and drafts both an internal RMA summary and a customer reply. A support person checks the eligibility call, approves or edits the reply, and issues any label or refund through the normal system.
Example workflow
- A new return request lands in the helpdesk or a shared inbox and triggers the workflow.
- The agent extracts the order number, items, and stated reason, then looks up order date, fulfillment date, and item details in Shopify.
- It checks the request against your written return policy and labels it eligible, outside window, needs photos, or warranty review.
- The agent drafts an internal RMA note with the facts it found and a suggested customer reply matching the label.
- A support person reviews the label and draft, edits as needed, and sends the reply from the helpdesk.
- Any return label, exchange, or refund is created by the support person, and the final outcome is logged on the ticket.
Tools
- Shopify
- Gorgias or Zendesk
- Gmail
- Claude
- Google Sheets
- Make
Risks and guardrails
- The agent may misjudge eligibility on edge cases such as final-sale or bundled items — guardrail: every eligibility label is a suggestion and a person confirms it before replying.
- A customer could be promised a refund amount that is wrong — guardrail: drafts never state refund amounts; the agent only notes the order total for staff reference.
- Requests without an order number may be matched to the wrong order — guardrail: low-confidence matches are labeled unmatched and routed to a person to ask the customer.
- Policy text in the prompt can drift from the published policy — guardrail: the agent reads policy from one maintained document and the draft cites the clause it relied on.
Related recipe
Support Ticket and Review Triage with Policy-Cited Draft Replies: An agent tags every helpdesk ticket and public review by intent, urgency, and sentiment, routes it, and drafts a reply that cites your help center. People approve refunds and all public replies.
Want it set up for you? Request a quote.
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Shopify Dev MCP server (AI Toolkit) (official page): shopify.dev/docs/apps/build/ai-toolkit (checked 2026-11-09)
- Gorgias (official page): www.gorgias.com (checked 2026-11-09)
- Zendesk AI (official page): www.zendesk.com/service/ai (checked 2026-11-09)
- Gmail (official page): developers.google.com/workspace/gmail (checked 2026-11-19)
- Claude (official page): claude.ai (checked 2026-11-09)
- Make MCP server (official page): developers.make.com/mcp-server (checked 2026-11-09)