Testing an AI Refund Workflow: Allow, Deny, Modify, and Escalate

Description and use

Testing an AI refund workflow with ACR Control Plane, demonstrating ALLOW, DENY, MODIFY, and ESCALATE outcomes.

Problem

Support agents need to perform refund operations while ensuring each action respects fine‑grained permissions and data‑handling policies.

Outcome

The test demonstrates that the gateway correctly allows, denies, modifies, and escalates refund requests, and that only authorized, sanitized requests reach the executor.

Workflow

• 1. Agent submits a customer lookup request. 2. Gateway evaluates and returns ALLOW; executor performs lookup. 3. Agent submits a delete_customer request. 4. Gateway returns DENY; executor receives no call. 5. Agent submits a $50 refund with an email in the reason. 6. Gateway allows, filters the email to [REDACTED], returns MODIFY; executor receives sanitized payload. 7. Agent submits a $150 refund. 8. Gateway returns ESCALATE and creates a pending approval. 9. Operator approves via approval API. 10. Gateway forwards approved request; executor performs refund. 11. Record decisions, approvals, and execution outcomes for audit.

Evidence

The recorded run uses synthetic customer data, a mock business executor, and an assistant-directed test harness calling running control-plane services.

Source

The support agent has registered permissions and a declared purpose. Customer lookup is allowed, customer deletion is prohibited, and refunds above $100 require approval.

Source

The refund passes the policy check. The PII filter then replaces the email address with [REDACTED], and the gateway returns MODIFY.

Source

The gateway returns ESCALATE and creates a pending approval.

Source