Research question and scope
This protocol asks what share of refund requests fall outside a written authority limit and how the queue handles them. The unit is one refund request, identified by the customer, the order, and the request date. The study measures whether the amount, frequency, and evidence conditions were met and whether the request was escalated to the correct owner. It does not evaluate whether the refund was a good commercial decision.
Set the scope to one queue, one refund policy version, and one observation window. Record the policy limits, the approval roles, and the system of record at the start. If the policy changes during the window, keep the earlier definition for the full sample or analyze the change as a separate period. A limit change can look like a change in request behavior when it is only a change in the rule.
Population and unit of analysis
Define the population as every refund request that reached a decision during the window, including requests that were refused or withdrawn. The unit is the request, not the customer, because one customer can make several unrelated requests. A request that is reopened after new evidence remains one unit with a revision history. A separate request for a different order is a new unit.
Record the customer type, the order value, the request channel, and whether the customer had prior refunds in the period. These strata matter because a first-time duplicate charge and a repeated claim carry different risk. Include requests that were approved within the limit, because the base rate is needed to interpret the exceptions. Test the unit definition on two ordinary cases before collection begins.
Definitions that make the measure reproducible
Define the written limit as the amount and frequency stated in the policy version that applied on the request date. Define an excess request as one whose amount exceeds the limit or whose frequency exceeds the permitted count. Define a split request as two or more refunds for the same order within a short window that together exceed the limit. Define a complete request as one with the order, charge, customer request, and reason recorded.
Do not count a request as within the limit when the amount is under the threshold but the frequency is over it. Do not count a request as complete when the reason field says other without a note. A refund that was approved by the owner outside the limit is still an excess request, because the measure is about the request, not the outcome. Each definition should include one positive example and one boundary example.
Evidence standard and data dictionary
For each request, capture the order identifier, the amount, the request date, the reason code, the evidence attached, the approval role, and the decision. Add the customer's prior refund count and whether the request was split across transactions. Record whether the specialist or the owner made the decision and whether the decision was recorded in the system of record. Keep the packet with the refund record for the next review.
Also capture the time from request to decision and the number of follow-up messages. These fields help explain an excess request that was escalated correctly but handled slowly. If the amount was entered in a different currency, record the conversion rate and the date used. A second reviewer should be able to reproduce the classification from the packet without asking the specialist. Keep the packet with the refund record for the next review. Keep the packet with the refund record so the evidence survives a later audit.
Sampling and collection
Draw the sample from the refund system rather than from a list of large refunds. Record the export date, the filters, the row count, and the selection rule. Include requests from every channel and every amount band so the result is not dominated by one source. If the population is ordered by request date, a fixed interval spreads the sample across the window.
Collect the request as it existed at the decision, not as it looks after later edits. If the customer added evidence after the first decision, record the edit time and analyze the first decision separately. Do not reconstruct a missing reason from the final outcome. Note the collection date separately from the request date so the reader can see how much of the record has aged.
Analysis and reporting
Report the number of requests, the share within the limit, the share above the amount limit, the share above the frequency limit, and the share with incomplete evidence. Show the counts by channel and by customer type. Present the median decision time and the share of excess requests that reached the named owner. Use the same denominator for every rate.
Inspect the excess requests and describe the pattern without naming the customer. A cluster in one reason code may point to a product or billing problem, while a cluster in one channel may point to a missing prompt in the intake form. Compare the result with the policy version, because a recent limit change can explain a temporary spike. Report the split requests separately, since they are a different control failure.
A useful additional view is the distribution of amounts just below the limit. A spike near the threshold may indicate deliberate splitting or a pricing pattern that the policy did not anticipate. Show the distribution in bands and describe the largest band without disclosing customer identities. That view helps the owner decide whether the limit is set correctly.
The review should also check the customer's original channel and the agent who first handled the request. A request that began in a phone call may lack the written evidence that an email request carries. Record the channel and the handoff point for every excess request. That detail helps the owner decide whether the intake form or the training needs the first change.
Limitations and sources of error
The record cannot show whether the customer was treated fairly or whether the refund was economically justified. It can show only whether the request met the written conditions and how it was routed. A high excess rate is not proof of poor performance, because the policy may be set below the normal range of legitimate requests. The protocol also cannot capture a refund that was issued outside the system.
Refund systems can omit manual adjustments, chargebacks, and credits issued in a different tool. A request that was resolved by a payment provider may never appear in the queue. Record those cases and their treatment instead of dropping them silently. If the same person approved the request and recorded the evidence, note the independence limitation and analyze that subgroup separately.
Review cadence and next action
Give the result to the finance owner and the support lead and ask them to choose one observable change. The change might be a clearer evidence field, a lower frequency threshold, or a rule that split refunds are reviewed together. Record the owner, the effective date, and the measure that will show whether the change worked. Avoid changing the limit and the intake form at the same time.
Repeat the protocol with the same definitions after the change and compare the same fields. If excess requests fell but decision time rose, the new evidence requirement may be too heavy. If split requests fell but customer complaints rose, the rule may be too rigid for legitimate cases. Close the review when the owner accepts the finding and the next sample is scheduled.
