Research question and scope
This protocol asks how many approval steps a delegated invoice queue actually requires and which steps add control rather than delay. The unit is one invoice, identified by the vendor, the amount, and the submission date. The study measures the number of approvals, the time each step adds, and the share of steps that change the outcome. It does not evaluate whether the invoice should have been paid.
Set the scope to one accounts payable lane, one approval policy, and one fixed window. Record the system of record, the approval roles, and the threshold rules at the start. If the policy changes during the window, keep the earlier definition or analyze the change separately. A new threshold can look like a change in invoice quality when it is only a change in the rule.
Population and unit of analysis
Define the population as every invoice that reached a final decision during the window, including invoices that were rejected or returned. The unit is the invoice, not the approval step, because one invoice can pass through several reviewers. Record each step as an event with its own time and outcome. A resubmitted invoice remains one unit with a revision history.
Exclude invoices that were cancelled before any approval and record the exclusion count. Include invoices below the threshold that still required review, because those are the cases where an unnecessary step is most visible. Test the unit definition on several ordinary invoices before collection begins. Keep the excluded invoices with the sample so the denominator is auditable.
Definitions that make the measure reproducible
Define an approval step as a recorded decision by a named person that is required before the invoice can move forward. Define a substantive approval as one where the reviewer checks the invoice against the purchase order, the receipt, or the vendor record. Define a rubber stamp as an approval with no evidence of review beyond opening the record. Define cycle time as the elapsed time from submission to final decision.
Do not count an automated system check as a human approval step, but record it separately because it can replace a manual review. Do not count a notification as an approval. Each definition should include one positive and one boundary example. Publish the definitions with the sample so a second reviewer can classify consistently.
Evidence standard and data dictionary
For each invoice, capture the vendor, the amount, the purchase order, the receipt, the approval roles, the decision times, and the final outcome. Add whether each reviewer changed the invoice or only approved it. Record the reason for any rejection or return. Keep the packet with the invoice record so a later reviewer can replay the path.
Also capture whether the approver had access to the supporting documents at the time of review. An approval made without the receipt is different from one made with the full packet. If the approval was verbal, record the note that was written afterward and mark it as reconstructed. A reviewer should be able to rebuild the path from the packet.
Sampling and collection
Draw the sample from the invoice system rather than from a list of large or late invoices. Record the export date, the filters, the row count, and the selection rule. Include invoices from every amount band and every vendor type so the result is not dominated by one source. If the population is ordered by date, a fixed interval spreads the sample across the window.
Collect the approval record as it existed at the decision, not as it looks after later edits. If an approver was added after a delay, record the change and analyze the original path separately. Do not infer the number of steps from the final status alone. Note the collection date so the reader can see how much of the record has aged.
Analysis and reporting
Report the number of invoices, the mean and median number of approval steps, and the share of steps that changed the outcome. Show the time added by each step and the share of invoices that waited more than one business day at a single step. Break the result down by amount band and vendor type. Show the number of invoices behind each step so a rate is never based on a handful of cases.
Compare invoices with two steps and three or more steps on cycle time and error rate. A third step that catches real errors may be worth the delay, while a step that never changes anything is a candidate for removal. Report the exceptions and missing evidence separately. If the sample is small, show counts rather than percentages.
Limitations and sources of error
The record cannot show an error that was caught informally before the invoice entered the system. It also cannot show the risk that a removed step would have prevented. A step that rarely changes an outcome may still be valuable as a deterrent. The study measures observed review behavior, not the full control value of the step.
Approval records can be incomplete when a reviewer approves by chat or in person. If the same person prepares and approves an invoice, note the independence limitation. Amounts can change after approval, which complicates the cycle-time comparison. Record these cases and their treatment rather than dropping them.
Review cadence and next action
Give the result to the finance owner and the accounts payable lead and ask for one change. The change might be a higher threshold, a sample review instead of a full review, or a required receipt at the first step. Record the owner, the effective date, and the measure that will show whether the change worked. Change one rule at a time.
Repeat the protocol after the change and compare the same fields. If cycle time fell but errors rose, the removed step was doing work that the policy did not describe. If cycle time did not change, the bottleneck may be document collection rather than approval count. Close the review when the owner accepts the finding and the next sample is scheduled.
Threshold design and sampling
An approval threshold is a policy choice, not a natural boundary. Test the current threshold against the distribution of invoice amounts and the errors observed near it. A threshold set below the common legitimate amount creates routine escalations. A threshold set far above the evidence gives away control. Show the distribution before recommending a change.
Sample invoices just below and just above the threshold. If the error rate is similar, the threshold may not be doing the work the policy assumes. If errors cluster just below it, the queue may be splitting invoices to avoid review. Record the pattern and investigate before changing the number.
Vendor and document variation
Approval steps may be necessary for some vendors and redundant for others. Compare cycle time and error rate by vendor type and document quality. A vendor that always sends a matching purchase order and receipt may not need the same review as a new vendor with incomplete documents. The rule can vary if the variation is written and justified.
Record whether the invoice arrived with the purchase order and receipt or required collection. Document collection is often the real delay, not the approval. If the step is blamed for a delay caused by missing documents, removing it will not improve cycle time. Measure the parts separately.
Reporting to finance
Report approval count, cycle time, and outcome-changing steps in a format the finance owner can compare across periods. Show the share of invoices that waited at each step. Avoid presenting a single average that hides a long tail. The owner should see where the delay concentrates.
Include the errors that the removed step would have caught, even if they were caught elsewhere. A control decision without the error evidence is only a speed argument. State the assumptions and the sample size. Offer a trial period with a sample review before removing a step permanently.
Replication and version control
Keep the policy version, the threshold, the system export, and the classification rules with the report. Repeat the measurement after the change on the same fields. If the system changes, publish a new baseline. A second analyst should be able to replay the approval path from the packet.
Set the next review date and the sample size before closing the report. Assign the finance owner and the accounts payable lead. A control review that is not repeated becomes a one-time opinion. The comparison is what shows whether the change held.
