The returns that never shipped.

86 refunds were approved in 30 days against a six-month average of 22. 74 relate to one product, a €95 three-month pack. 61 of the 86 were approved by one agent, Dario, in a team of six. All 61 used the ‘return label issued’ path, which releases the refund when the label is generated rather than when the parcel is scanned. The carrier shows no scan on any of the 61 labels. Nine of the other 25 refunds have scans.

The ledger shows where the money went and through which path, and it establishes a concentration worth explaining. It does not show intent: the agent may be following the workflow as designed, and an unscanned label can mean a parcel in a cupboard rather than a parcel that never existed.

Dario works the evening shift, which handles 55% of refund tickets because subscription renewals bill at 18:00. A naive allocation would assign about 47 of the 86 approvals to that shift, but ticket mix and approval opportunities may differ; this is not a validated expected count. His approval rate on tickets he touches is 94%, against a team range of 70% to 81%. He joined four months ago; refunds on this product began rising in his third week. Two other agents also approved refunds for accounts in the cluster.

Workload explains why raw agent counts are an unreliable comparison. Ticket mix, decision authority and the denominator of eligible requests still need review. The 94% approval rate is a lead to examine alongside those factors, not evidence of intent or a causal estimate of excess refunds.

The 61 refunds belong to 23 customer accounts. 19 of those accounts share four delivery addresses and two device references, and 17 were created within one week in August. Each placed between two and four orders of the same pack before requesting a refund. The 61 refunds total €5,795. Two accounts in the cluster are long-standing customers with ordinary order histories and one previous refund each, both scanned.

The shared attributes create an investigation cohort, not a verified ring. The two long-standing accounts need individual review. Neither the links nor the support-agent concentration establish collusion.

The fictional policy permits a second review of discretionary early refunds. It does not permit delaying statutory refunds or overriding customer rights. A carrier scan or alternative dispatch evidence may support release where appropriate. Existing completed refunds cannot simply be reversed. The policy owner must approve any change before the batch runs.

The decision concerns a discretionary early-release workflow, not a general rule that customers must wait for a carrier scan. Preserve applicable deadlines and accessible alternative evidence.