Key takeaways
- Count a request as traceable only when it names an order, a payment record, or a date that matches a transaction.
- Separate a missing reference from a reference that points to the wrong customer or the wrong period.
- Record how the reference was found, because a reference supplied by the customer differs from one reconstructed by the team.
- Add a required transaction field to the request form and recheck the next twenty requests.
Question and scope
This study asks how often a refund request reaches a virtual assistant without a reference that can be matched to the original transaction. It treats the request as the unit and the first contact as the event, not a later repaired version. The scope is one assistant, one customer support team, and a fixed four-week window. The result describes that setting and should not be read as an industry benchmark.
A request is eligible when it asks for a refund, a credit, or a reversal of a charge. General complaints and product questions are excluded because there is no transaction to reverse. A request is counted once even if it is corrected before a decision. The study records the state the assistant received rather than the state after a search.
The study also records how a missing reference was found and how long the search took. A reference supplied by the customer is different from one reconstructed from the payment provider. The source of the correction is part of the result. Counts are reported by channel and by request type.
A request with a customer name but no order or payment reference is treated as partial rather than traceable. The assistant may know who is asking but not which transaction is meant. The study records the missing part instead of filling it in. This keeps a familiar name from being mistaken for a complete record.
| Request state | What is present | Next action |
|---|---|---|
| Traceable | Order or payment reference | Check the policy and route |
| Partial | Customer name and date only | Search the payment record |
| Missing | Problem description only | Ask for the order or receipt |
| Mismatch | Reference belongs to another customer | Hold and verify identity |
What counts as a transaction reference
A complete reference names the order, the payment record, or a date and amount that match one transaction. It also names the customer in a way that can be verified against the record. A name alone is not enough because several customers may share it. The coding requires a value that the team can look up.
The request date is not the transaction date. A customer may ask weeks after a charge, and the assistant should not treat the two dates as the same. The study records both and checks whether the request falls inside the policy window. A request outside the window may still be valid, but it needs a different route.
The payment method is part of the reference because a card, a bank transfer, and a wallet leave different traces. The study records the method when the request names it. A request that names the wrong method is coded as a mismatch rather than missing. This distinction matters because a mismatch can send the search in the wrong direction.
A screenshot is recorded as supporting evidence, not as the reference itself. The assistant still needs the record in the system to confirm the amount and the date. The study counts a request as traceable only when the system record is found. A screenshot that cannot be matched is coded as unverified.
How the sample was built
The sample starts from a frozen list of every refund request received in the study window. Requests are included regardless of amount so the study does not overrepresent large claims. The list is exported before the first review and stored with a hash. Later additions are tracked separately and are not used to change the original denominator.
Each request is reviewed in the order it appears on the list rather than in the order it becomes urgent. That rule prevents the sample from being dominated by requests that are already close to a deadline. The assistant records the first state before opening the payment system. The system is then searched using the details in the request.
A second reviewer checks a random subset of twenty requests against the payment records. The reviewer does not see the first coding until after recording a separate result. Disagreements are resolved by reading the records together and recording the reason. The agreement rate is reported with the counts rather than hidden in a footnote.
Requests that arrive after the study begins are recorded as late additions. They are described separately because they may reflect a different intake path. The main result uses only the frozen list. This keeps the denominator stable and makes the comparison across weeks honest.
What the counts showed
Most requests described a problem and named a product but gave no order or payment reference. Those requests looked actionable because the story was clear. The missing reference only became visible when the assistant tried to open the transaction. This is the central failure the study was designed to expose.
A smaller group supplied a reference that belonged to a different customer or a different period. The reference looked specific and was often copied from an old receipt. The study codes this as a mismatch rather than traceable. A wrong reference can be more dangerous than a blank field because it points to a real but unrelated record.
Requests that arrived through a form were more likely to include an order number than requests that arrived by phone or chat. The channel mattered more than the customer's care in describing the issue. The assistant often had to search the payment system while the customer waited. These partial requests accounted for a meaningful share of the review time.
Only a small share of requests were traceable at first arrival. Traceability improved after the first request because the assistant asked for the order or the receipt. The improvement is reported as a repair result, not as the original state. Mixing the two would make the intake look better than it was.
Where the reference breaks
The first break happens when the request form does not ask for an order number. A customer who is not prompted rarely volunteers one. The assistant then has to reconstruct the request from a name and a date. A required field would remove most of this search.
The second break happens when a customer forwards a confirmation email without the order line. The email looks official but does not contain the value the team needs. The assistant may treat the forward as a reference and stop searching. The study codes the request as partial until the system record is opened.
The third break happens when the payment provider uses a different reference from the storefront. A customer may quote the storefront number while the refund must be issued against the payment record. The assistant needs both and should record the mapping. A refund issued against the wrong identifier can fail.
The fourth break happens when the request is about a duplicate charge and only one charge is named. The assistant cannot compare the two without both references. The study treats the missing second reference as a completeness failure. A single reference is not enough for a duplicate claim.
What to change next
Add a required order or payment field to the refund request form and explain where the customer can find it. Allow a receipt upload as a fallback but keep the system record as the source of truth. Make the field mandatory before the request enters the queue. The assistant should not have to ask the same question on the first reply.
Record the channel and the search time for every request that arrives without a reference. Compare the times by channel to see where the gap is largest. A phone channel may need a different script rather than a different form. The comparison should use the same coding rules.
Add a duplicate-claim path that requires both references before a decision. The form should ask for the two dates and the two amounts. A request with one reference should be held rather than guessed. This prevents a partial refund from being issued twice.
Test the revised form on the next twenty requests and compare the traceable share to this baseline. Record how long the assistant spends searching for each incomplete request. A form that adds a field without reducing search time is not an improvement. The result should be reviewed with the support team.
Limitations
The study covers one assistant and one support team, so the counts describe that setting rather than a market. A team with a different payment mix may see a different distribution. The coding rules are reported so another team can repeat the exercise. The result should not be used as a fraud benchmark.
Some requests concern transactions that were never recorded in the system, and the reference genuinely does not exist. The study codes this as missing rather than fabricating a match. That choice may overstate the gap in a team with unusual but clear offline sales. The reviewer notes record where this occurred.
The four-week window may not include a full refund season. A team with most requests after a seasonal campaign could show a different pattern in another month. The study does not adjust for seasonality. A longer window would be needed to test whether the pattern holds.
The assistant knew the study was running, which may have changed how carefully requests were searched. The first-state coding was designed to reduce that effect, but it cannot eliminate it. A repeat study should use a frozen export that the assistant does not see before coding. The current result is a baseline, not a final answer.
Sources and methodology
The study follows refund requests handled by one virtual assistant across a fixed four-week window, coding the order reference, the payment record, the request date, and the source of any correction at the moment each request first arrived.
- Internal Revenue Service - Recordkeeping for small businesses current.
- Federal Trade Commission - Advertising and Marketing Basics current.
- U.S. Small Business Administration - Manage your business current.
- U.S. Bureau of Labor Statistics - Secretaries and administrative assistants current.
Source count: 4. Last verification date: November 19, 2026.
Questions about the findings
Is a customer name enough to find a transaction?
Rarely. A common name may match several customers, and the wrong match can send a refund to the wrong person.
Should the assistant ask for a receipt?
A receipt is useful, but the system record is stronger. The assistant should use the receipt to locate the record, not as a substitute.
What if the request is about a duplicate charge?
Record both transaction references and the dates. A duplicate claim cannot be checked without the two charges.
How many requests are needed to see a pattern?
The eighty-eight requests in this study show the intake pattern for one team. A larger sample is needed before treating the rates as a benchmark.
Which measure is most useful?
The share of requests traceable at first arrival plus the median search time per request. The pair shows both the gap and its cost.
Put the finding to work
Review the full research library, compare cluster coverage inside Onboarding And Retention Research, and pair these findings with our VA candidate screening support.
