The problem
A refund queue mixes routine requests with exceptions. Reviewing every request manually is slow; approving everything automatically ignores policy and customer history.
How the workflow decides
Clear requests pass the configured rules. Larger or late requests wait for a person, duplicate refunds are blocked, and changes to an approved amount require a new decision. The reviewer sees the reason and supporting records.
This public demo is a policy rules engine running in the browser on eight sample requests. It illustrates the approval controls that an agent workflow can use; it does not call an AI model or issue real refunds.
Work through the queue
Run the queue with the default rules, then open a request to inspect the decision. Change a limit and run it again to see which requests need review.
2The Queue · click a row to open the record
| Request | Customer | Reason | Amount | Status |
|---|
3Audit Log
What this demonstrates
The important behavior is a clear boundary between automatic decisions and human judgment. Each decision leaves a readable record. Connecting the workflow to a help desk or payment provider would be a separate integration.