Every refund is a small decision about money, policy and a customer. Most stores make thousands of them a year by hand, and almost none of them can say afterwards why a given refund was paid.
Today we are launching Lumtry, agentic refund orchestration for e-commerce. Lumtry decides refunds with deterministic policy plus AI reasoning, routes edge cases to your team for approval in the dashboard or Slack, executes the refund on Stripe or PayPal, and keeps a full audit trail. It is the second agent from our studio to become a product, after MarginTide.
The refund problem is hard to audit.
Refunds look simple from the outside. A customer asks, someone checks the order, someone clicks a button. In practice each request pulls in the order history, the store's policy, the payment processor and, often, a judgement call about whether the request is reasonable.
That creates three problems that show up again and again. Decisions are slow, because every request waits for a person with the right access. They are inconsistent, because two people read the same policy differently on a busy Monday. And they are hard to audit, because the reasoning lives in someone's head or a chat thread, not next to the refund.
Fully automatic refunds fix the speed and make the other two worse. Fully manual refunds keep control and give up the speed. We wanted the routine cases handled by rules you wrote, the unusual ones handled by people you trust, and a record of both.
How it works: policy, reasoning, approval, payout, trail.
Every refund request in Lumtry takes the same path.
- Policy decides first. Your refund policy is written as deterministic rules you can read: amount, currency, reason and more. The policy is versioned, so you always know which version made a call.
- AI reasoning fills in the gray areas. A model reads the request and the context and explains cases the rules do not settle cleanly. It informs the decision. It never overrides your policy.
- People approve the edge cases. Anything outside the rules is routed to your team for approval in the Lumtry dashboard or in Slack, with the reasoning attached, so a decision takes one look rather than a new investigation.
- Lumtry executes the refund. Approved refunds are paid out on Stripe or PayPal, so nobody has to switch tools to finish the job.
- Everything lands in the audit trail. Each step records what was decided, by which rule or person, and why. When finance, a processor or a customer asks about a refund, the answer is already written down.
This is the pattern we use for every agent we ship: rules where rules work, a model where judgement is needed, a person where the stakes are high, and telemetry on all of it. Lumtry packages it for one job.
Who it is for.
Lumtry is built for small and mid-sized e-commerce merchants, and for the four people who feel refunds most. Owners want control over what money leaves the business. Ops leads want requests to stop piling up. CX leads want customers to get an answer without waiting on a manager. Finance wants a record that reconciles.
If your store handles more refund requests than one person can review properly, or you have ever been unable to explain a refund after the fact, Lumtry is for you. If you want to put a number on the problem first, our refund cost calculator uses your own volumes.
Works with the tools you already use.
Lumtry connects to Shopify, WooCommerce and BigCommerce stores and to Amazon and eBay. Refunds execute on Stripe and PayPal. Approvals happen in Slack as well as the dashboard, and it works alongside Gorgias and Zendesk so support teams stay in their helpdesk.
Lumtry also includes an MCP server, so AI agents and assistants can work with refund cases under the same policy, permissions and audit trail as your team. An assistant works inside your rules; it cannot skip them.
Start with your own policy.
Lumtry is launching now. The fastest way to see whether it fits is to write your refund policy into it and watch how it treats the requests you already get.
Set up your refund policy and see Lumtry decide real cases.
Get started with Lumtry