Protocol premise

The study would test whether the bank feed changes how quickly and how accurately a receipt reaches the right invoice.

Protocol status and question

This is a proposed research protocol, not a report of completed findings. It asks whether two bank feeds produce different cash application outcomes for comparable receipts. The result would help a finance team decide whether a feed change needs a different review rule. Keep the protocol and its approval date with the data file. State plainly which decisions the protocol cannot support.

The protocol does not compare the banks as businesses. It compares the evidence and the handling of that evidence in one team's workflow. The sponsor should define the decision the result could inform, such as a matching threshold or a training change. The sponsor should record that decision before collection starts. Record who approved the question and when.

Population and observation unit

Define the population as receipts from two named feeds during a stated window, limited to customers that appear in both feeds. The unit is one receipt, with the feed, the remittance fields, the application outcome, and the time to application recorded. Receipts from one-off customers should be excluded and counted. Write the frame as a dated list so a later reviewer can rebuild it. If the frame changes mid-study, keep the earlier version.

Decide how to treat a payment that is split across feeds. Link the parts to one economic receipt and record the split. The analysis can compare the feed-level records and the linked receipt separately, but it should not double-count the same money. A count that cannot be rebuilt is not a population. Name the system that produced the frame and the export time.

Definitions and measures

Define accuracy as the receipt being applied to the correct customer and invoice without a later reversal. Define timeliness as the elapsed working time from receipt availability to application. Define completeness as the share of required fields present in the feed's remittance data. Each definition needs a documented source and a review rule. Record the definition version with the data so the check can be repeated.

Report the measures by feed and by customer segment. A feed that is slower overall may be faster for a subset of customers, and that difference may matter more than the average. Keep the counts visible beside every rate. A definition that two reviewers read differently is not ready. Record the definition version with the data so the check can be repeated.

Evidence standard

Link each application to the bank record, the remittance, and the ledger entry. A correction should link to the original application and the reason for the change. Use timestamps from the systems where possible, and mark any reconstructed time as an estimate. Link each classification to a document a second reviewer can open. Keep the original record even when a later correction is added.

If a receipt was applied manually, record the analyst and the evidence used. Manual work is not a failure by itself, but it needs to be visible because it may explain a difference between feeds. The study should distinguish a feed problem from a workflow problem. State what the evidence does not show. Use stable references rather than personal folder names.

Sampling and collection

Use a census of the shared-customer receipts if the volume is manageable. Otherwise stratify by feed, customer size, and amount. Select the sample before looking at application outcomes and keep a log of replacements. Do not replace a slow or reversed application with a cleaner case. State the final sample size and how it was reached.

Train reviewers on a common set of receipts and compare their classifications. Clarify how to treat a receipt that was applied and then moved to another invoice. Record the first application and the final application separately so the analysis can see both. Compare two reviewers on an overlap sample. Pilot the collection on ten records before the main pass.

Analysis and reporting

Report accuracy, timeliness, and completeness for each feed, with the denominator and window. Break the results down by remittance quality and by customer type. Show the reversals separately because a fast but reversed application is not a good outcome. Report counts with rates and keep the denominator visible. Show the distribution, not only the average.

Avoid claiming that one bank is better. The result describes the interaction between the feed, the customer, and the team's process. A recommendation should name the specific field or rule that would reduce the difference. Name the counterexamples as well as the main pattern. Show the distribution, not only the average.

Limits and sources of error

The two feeds may serve different customer mixes, which can explain part of the difference. The study should report the mix and avoid comparing averages without it. A feed change during the window can also disrupt the comparison. Keep the limitation in the same sentence as the claim it qualifies. Name the records the study could not see.

Manual reconstruction of timestamps and the exclusion of one-off customers can bias the result. State both limits and keep the excluded records visible. The reader should be able to see how the population was built. Do not generalize beyond the documented frame. Describe the missing-data rate and its likely direction.

Review cadence

Repeat the review after a feed change, a matching-rule change, or a significant customer migration. Use the same definitions and the same shared-customer frame. A second review can test whether a process change reduced the gap. Set the next review date before the current one ends. Close the review with one question for the process owner.

Assign the follow-up to the cash application lead and set a date. The protocol should produce a specific decision about which feed fields to require or which matching rule to test next. Compare the result with the prior period and explain the change. Version the protocol when the method changes. Set the next review date before the current one ends.

Sources

  1. Financial Accounting Standards Board
  2. IRS Publication 583, Starting a Business and Keeping Records
  3. U.S. Small Business Administration
How many customer statements disagree with the ledger?

Put the protocol into operational context.

Discuss the queue, evidence, ownership, and safeguards that apply to your workflow.

Discuss your workflow