Small-business outsourcing research

What share of weekly cash-flow reports need a manual correction?

A protocol for measuring corrections in a small-business cash-flow report, with the source, the stage, and the owner decision affected.

What share of weekly cash-flow reports need a manual correction?

Research question and scope

This protocol asks what share of weekly cash-flow reports need a manual correction before the owner can use them. The unit is one weekly report for one business, identified by the reporting date and the version. The study measures the number of corrections, the stage where each was found, and whether the correction changed an owner decision. It does not judge whether the owner made the right decision with the report.

Set the scope to one business, one reporting template, and one source system. Record the accounts included, the reserve rules, and the approval thresholds at the start. If the template or the source changes during the window, analyze the cohorts separately. A new template can reduce corrections without any change in the underlying data.

Define the population as every weekly report issued during the window, including reports that were corrected before or after the owner read them. The unit is the report, not the correction, because one report can contain several errors. Test the definition on two recent reports before collection begins. Record the test result and the boundary cases.

Definitions that make the measure reproducible

Define a correction as a change to a number, a label, a date, or a classification after the first draft was shared. A formatting change that does not alter meaning is not a correction. A change made before the report was shared is a draft edit, not a correction. Record the sharing time so the two can be separated.

Define the report as shared when it reaches the owner through the agreed channel, not when it is saved. Define the owner decision as the recorded choice to pay, delay, hold, or investigate. Define a decision change as a reversal or material change after a correction. Record the original and the revised decision so the effect is visible.

Record the report date, the preparer, the reviewer, and the time the report was shared. Add whether the report was prepared from the live system or a downloaded file. A downloaded file can be stale before the report is finished. Note the source and the export time beside the report.

Evidence standard and data dictionary

For each correction, capture the field, the original value, the corrected value, the person who found it, and the time it was found. Record whether the correction came from a bank feed, an invoice, a supplier message, or the owner's memory. Add the stage: draft, owner review, or after a decision. Keep the draft and the final version so the change can be checked.

Record whether the correction changed the cash available, the committed amount, or the open risk line. A correction to a label may matter less than a correction to the reserve calculation. Classify the correction by its effect so the count is not dominated by cosmetic edits. A second reviewer should be able to rebuild the report from the record.

Record the time spent finding and fixing each correction. The time is part of the cost even when the final number is right. Add the number of people involved and whether the owner had to answer a question. A correction that pulls the owner into the detail defeats the purpose of the report.

Sampling and collection

Draw the sample from the report archive, not from a list of reports the owner liked. Include reports that were corrected, delayed, or replaced. Record the export date, the filters, and the number of reports excluded. If the archive is incomplete, state the gap rather than filling it from memory.

Collect the report as it existed at the time, not after a later cleanup. If a value was fixed days later, record the fix date and analyze the original delay separately. Do not infer a correction from a bank transaction that was later reclassified. Note the collection date separately from the report date.

Record the preparer's shift, the owner's review time, and any system outage. A report prepared during an outage may have more corrections for reasons outside the preparer's control. Add the holiday calendar and the payroll week. These fields help separate a data problem from a staffing problem.

Analysis and reporting

Report the number of reports, the share with at least one correction, and the median number of corrections per report. Show the counts by field, stage, and effect on a decision. Present the median time spent on corrections and the share found after the owner read the report. Use the same denominator for every rate and show reports with incomplete records.

Inspect the reports with the most corrections and describe the source without naming the preparer. A cluster in the reserve line may point to a rule change, while a cluster in supplier payments may point to a missing approval. Compare the result with the payroll week and the system outage flag. Report reports that were replaced entirely separately.

A useful additional view is the number of corrections the owner found personally. An owner who finds errors at review is spending time on quality control instead of decisions. Show the owner-found share beside the preparer-found share. That view helps the business decide whether to fix the source or add a review step.

Limits of the evidence

The record cannot show whether the report was useful or whether the decision was correct. It can show only the corrections, the timing, and the recorded decisions. A report with no corrections is not proof of accuracy if nobody checked it. The protocol also cannot capture a correction made verbally and never written down.

A small business may have too few weekly reports to support a stable share. Describe the reports and the correction types instead of ranking the preparer. If one week dominates the sample, report that concentration. A general rate would hide a single payroll event.

A correction that is made before the owner reads the report may still be a quality problem. Record the draft edit separately so it is visible without inflating the owner-facing rate. A high draft-edit count may point to a source that is hard to use. That evidence should lead to a source fix, not a longer review.

Give the result to the owner and the preparer and ask them to choose one observable change. The change might be a required source field, a cash rule written in the template, a second review before sharing, or a different export time. Record the owner, the effective date, and the measure that will show whether the change worked. Repeat the protocol with the same definitions after the change.

If corrections fell but review time did not, the report may still be hard to read. If review time fell but decision changes rose, the report may be hiding uncertainty. Report the two movements together before declaring a success. Close the review when the owner accepts the finding and the next cohort is scheduled.

Record whether the preparer had access to the live bank feed or only to a downloaded statement. A statement that is a day old will produce corrections that the live feed would have prevented. Note the source and the time of the last refresh. That detail separates a skill problem from a data-access problem.

Ask the owner which correction changed their decision and which one only changed the presentation. A presentation fix can wait, while a decision-changing error needs a source control. Record both and report them separately. A single correction count treats a typo and a wrong reserve as the same event.

Ask the preparer which number was hardest to source and why. A concrete answer is more useful than a general complaint about the bank feed. Fix that source or add the missing guidance to the template. The next measurement should test whether the same correction recurred. Note the date each correction was made so a later reader can see the sequence. Add the source refresh time so the owner knows whether the report used live data.

Sources

Frequently asked questions

Does a favorable sample prove the routine caused the result?

No. The design is observational and supports a bounded operating decision, not a universal causal claim.

What should an owner review first?

Start with high-consequence defects, missing denominators, exceptions, and cases that required the owner to reconstruct the evidence.

Related research