Research question and scope

This protocol measures how often an invoice or payment request fails a duplicate-payment check. The unit is one candidate payment that enters the check during the observation period. A failure means the check placed the payment on hold for a possible duplicate. The measure describes the check and the records, not the honesty of any vendor or specialist.

Set the scope to one accounts queue, one payment system, and a fixed period. Record the matching rules, the review window, and the statuses used at the time. If the rules change during the period, split the sample at the change date. A new rule can change the failure rate without changing the underlying records.

Keep the measure separate from a fraud investigation. A duplicate hold may come from a data error, a reissued invoice, or a genuine second charge. The purpose is to find where the records allow a duplicate to enter. State that purpose in the report.

Population and unit of analysis

Define the population as every candidate payment that entered the check, including payments that cleared without a hold. The unit is the candidate payment, not the invoice document. One invoice can generate a correction and a second candidate, and each candidate should be counted. A hold rate without the cleared payments has no denominator.

Include payments that were held and later released as valid. A release is part of the check's result because it shows a false positive or a legitimate second charge. Count the reason for the release. Excluding releases would overstate the accuracy of the rule.

Exclude payments that never reached the check because they were rejected at intake. Report those separately as an intake failure. The duplicate check cannot catch a record it never sees. Keeping the two stages apart makes the result easier to interpret.

Definitions that make the measure reproducible

Define a hold as a candidate payment that the check prevents from entering the scheduled batch. Define a true duplicate as a second payment for the same obligation and period. Define a recurring bill as a separate obligation that happens to share an amount or a description. Publish one example of each.

Define a release as a held payment that an authorized reviewer allows to proceed. Record the reviewer, the evidence, and the date. A release without evidence is a control gap rather than a valid result. The definition should make that distinction clear.

Set the review window before reading the sample. Use the same window for every vendor unless the vendor has a documented billing cycle. Record the window with the result. A short window will miss late duplicates and a long window will produce more false positives.

Decide whether a credit note or a reissued invoice counts as a duplicate attempt. A reissue may be legitimate when the original was voided. Record the status of the original before classifying the candidate. A clear rule prevents the same case from being counted differently by two reviewers.

Sampling and data collection

Draw the sample from the payment system rather than from a list of known problems. Record the export date, the filters, and the row count. Include every vendor and every payment method in the period. A sample built from holds alone cannot produce a failure rate.

Capture the candidate identifier, the vendor identifier, the invoice number, the amount, the period, the bank record, and the hold status. Add the original payment that triggered the match. Store the evidence with the record. A count without the matched records cannot be audited.

Check a small overlap sample with a second reviewer. If the two reviewers disagree about whether a payment is a true duplicate, the definition is too vague. Rewrite the definition and retest before the full sample. Agreement matters more than sample size for a rule that will be reused.

Record the time from hold to release or cancellation. A hold that is resolved after the payment cutoff has no effect on that run. Report the timing with the result. The rate alone does not show whether the check changed the outcome.

Measuring duplicate-check failures

Report the share of candidate payments that were held, the share that were true duplicates, and the share that were released as valid. Show the counts behind each share. A high hold rate with few true duplicates may mean the rule is too broad. A low hold rate with a known duplicate may mean the rule is too narrow.

Break the result down by cause, such as same invoice number, same amount and period, changed bank account, or vendor reissue. A cause that appears often points to a specific fix. Keep the list short and stable across the comparison period. Add a short note to each cause so a new reviewer can classify the next case.

Report the vendors with more than one hold and the oldest unresolved hold. A repeated vendor may need a cleaner invoice process rather than a stricter check. The oldest hold may show a decision that no one owns. Both examples make the aggregate result actionable.

Limitations and next action

The measure cannot show duplicates that were paid without entering the check or that were caught by the bank after release. It also cannot show a duplicate that the vendor never reported. Treat the result as a measure of the check, not of all payment risk. Note the systems that were not included.

Give the result to the accounts owner and the payment approver with one rule to test. The change might be a required invoice-period field, a tighter bank-change review, or a shorter matching window. Record the effective date and repeat the measure with the same definitions. A controlled change is easier to judge.

Review the holds after the change and confirm that the true duplicates still appear. A rule that reduces false positives by missing real duplicates is not an improvement. Look at both sides before keeping the change. The goal is a check that the team can trust and use.