Features
Every refund decision, grounded and reviewable
Lumtry pairs rules that never drift with reasoning you can inspect, wrapped in durable workflows and a complete audit trail.
Decisioning
Policy decides, AI advises, people approve
Deterministic policy engine
Author versioned rules that decide the predictable cases the same way every time. Versions are immutable once published, and in-flight refunds pin the policy version they started on, so a rule change never rewrites a decision already in motion.
AI-assisted decisioning, always governed
For the cases policy cannot settle on its own, model-assisted reasoning weighs the context through the typed model router and proposes a disposition, with the rationale attached. A new or changed policy can run in shadow mode alongside the live decision path first, so you see what it would have done before it decides anything.
Approvals wherever your team works
Approve from the dashboard, Slack, Gorgias, Zendesk or a voice call. Every approval is checked against the case first.
Operations
Execution you can audit
Stripe and PayPal execution, with compensation
Approved refunds run as a durable workflow from intake through validation, policy, approval, and execution. If a downstream step fails, an explicit compensation step reverses what already happened, so nothing is left half-done.
Audit trail and correlation
Every decision, signal, and override is append-only and tagged with a correlation ID that follows the case end to end. Auditors get a complete, immutable trail without asking engineering to reconstruct it.
Return-receipt gating
When a refund is contingent on a physical return, execution waits for the return to actually arrive before money moves, closing the gap between approving a refund and paying one out early.
Usage-metered billing
Refund volume is metered against your plan's entitlements, with usage held rather than silently dropped when a limit is reached, so nothing processes without your workspace being able to account for it.
Storefront and marketplace intake
Shopify, WooCommerce and BigCommerce stores, plus Amazon and eBay, feed one case queue with integration health on every connection.
AI you can audit
Every AI-assisted decision, recorded and reviewable
Every AI-assisted decision is recorded
Every plan writes a record of the signal, the reasoning and the policy rule behind each AI-assisted decision, so nothing a model touches is a black box.
Shoppers can contest an AI-assisted decision
A shopper can contest a decision a model helped make. The contest moves through three states, from opened to resolved, and the outcome joins the case record.
A remediation proposer drafts fixes for review
When a pattern of cases points at a policy gap, the proposer drafts one candidate rule change for a person to review. It never executes anything on its own.
Low-confidence cases retry on a stronger model
When a case's AI-assisted signal comes back low confidence, it retries once on a stronger model before reaching a person. The cascade runs in shadow, alongside the live decision path.
Outcomes feed back into the record
Each AI-assisted run is joined to the outcome it led to, so the record shows what a model suggested and what actually happened next.
A monthly AI allowance, with a warning before it runs out
Your workspace gets a monthly AI allowance and a warning at 90 percent used. You can buy a top-up, and running out pauses only the optional AI agents, never refund processing.
A platform-wide brake for AI
If something looks wrong with a model vendor or a result, the platform can pause AI assistance across the board while refund processing keeps running on policy alone.
Model vendor failover
Five model endpoints are registered in the sub-processors register, and three route live traffic today. If a routed vendor fails, the request fails over to another registered endpoint instead of stalling the case.
Simulate a policy before you publish it
Replay a draft policy against your recent decisions and see where the outcome would change, with no model call and no cost, on every plan.
An AI-assisted rule explainer
When a policy gap keeps producing the same manual review, the explainer drafts one candidate rule as a pending suggestion. It never publishes on its own; a person reviews and publishes it.
See what comes next
What we are building after launch. Nothing here is live today.
Next up
- ERP and accounting syncPost refunds and credits to the ledger your finance team already closes.Coming soon
- More regionsAdditional hosting regions, including Azure and managed deployment lanes.Coming soon
Further out
- Digital goodsRefund decisions for licenses, subscriptions and downloads.Planned
- Shopper receipt agentShoppers prove purchase through an assistant that proposes, never decides.Planned
- Bring your own model keyRoute AI tasks through your own provider account.Planned
- Data residency optionsChoose where case data is stored.Planned
See it on your own refunds
Connect a store and watch policy and AI handle the volume while your team handles the exceptions.