Process routine refund and return requests
The problem
Simple refund requests require an agent to check order data, policy, and then click through several systems.
How the agent works
The agent verifies the order, checks eligibility against your written policy, and either issues the refund within limits or prepares it for approval, then replies to the customer.
Setup steps
- Write the refund policy as explicit rules
- Give API access scoped to lookup and refund only
- Set a dollar cap for auto-approval
- Log every action with reasoning
- Route exceptions to a human
Tools that can do this
- Decagon
- Sierra
- Custom agent with Stripe/Shopify APIs
- Intercom Fin (with actions)
Not ranked and not sponsored. Check each vendor's current features and pricing.
Human in the loop
Human approval above the cap or outside policy.
Risks and guardrails
Money movement: start in draft-only mode, use tight permission scopes and an audit log.
ROI notes
Removes manual clicks from high-volume, low-judgment requests.
Also useful: reconcile bank transactions with your books.
Related playbooks
Triage and tag every support ticket
Tickets land in one queue unlabeled, so urgent issues wait behind password resets.
Answer common customer questions from your help docs
The same questions are answered manually dozens of times a day.
Turn support tickets into a weekly voice-of-customer report
Product teams don't see patterns hidden across hundreds of tickets and reviews.