Define rework before counting it
Rework is work that had to be done again because the first pass was wrong or incomplete. A corrected typo, a resubmitted order, and a refund reopened after a bad entry are common examples. Record the definition with two examples and one non-example. A broad definition turns every follow-up into rework and hides the real defects.
Decide the unit you will count. An order, a line item, a customer record, and a correction event are different units with different rates. Record the unit and the source system. A rate that mixes units cannot be compared across weeks.
Separate rework caused by the desk from rework caused by the source data. A missing price from the client is not the same as a mistyped quantity by the specialist. Record the cause category on every entry. The split decides who can fix the problem.
Exclude changes that the customer requested after the first pass. A customer who changes an address is not a quality defect. Record the exclusion rule and the count. The log should measure avoidable repeat work, not normal change.
Agree the review owner and the meeting where the log is read. A log with no reader becomes a record of blame. Record the owner and the cadence. The team should know that the entries lead to a change, not a penalty.
Capture the cause at the moment of correction
Open the log when the correction is made, not at the end of the week. The person correcting the order knows what was missing and why. Record the case, the field, the original value, and the corrected value. A later reconstruction loses the detail that makes the cause useful.
Use a short cause list that the team helped write. Missing intake field, wrong system value, unclear instruction, and copy error cover most cases. Record one cause per entry and a note when two causes combine. A list of twenty causes will be used inconsistently.
Record the time between the first pass and the correction. A correction found in minutes may be caught by review, while one found after a customer complaint has already reached the customer. Record both times. The gap shows where the quality check is failing.
Note whether the correction changed a customer-facing promise, such as a price or a delivery date. Record the promise and the new message. A quiet ledger correction can leave the customer with the wrong expectation. The log should link to the message that fixed it.
Keep the log in the same tool as the order or link it from the case. A separate spreadsheet that nobody opens during the work will be filled in late or not at all. Record the link and the owner. The nearest form wins.
Separate source data errors from desk errors
Source data errors begin with a client system that is missing, stale, or contradictory. Record the field, the system, and the value the desk saw. The client needs this evidence to fix the input. A desk blamed for a client export will stop reporting the export problem.
Desk errors begin with a choice the specialist could have made correctly with the information available. Record the instruction, the choice, and the correct path. The entry should show what a reasonable person would have needed to know. This keeps the review focused on the process rather than the person.
Some errors sit between the two, such as an instruction that is technically correct but easy to misread. Record the ambiguity and the example that caused it. The fix may be a template change or a clearer field label. A category called unclear instruction is more useful than a general other.
Count the source errors and the desk errors separately in the weekly summary. A high source-error count points to the client, while a high desk-error count points to training or review. Record both counts and the volume. The comparison should change the next action.
Share a source-error example with the client without the customer's personal details. Show the field and the effect, not the person who noticed it. Record the example and the requested fix. A concrete case gets a faster response than a percentage.
Review the log weekly and choose one fix
Read the week's entries together and group them by cause and by field. A repeated missing field is a stronger signal than a single dramatic error. Record the count and one example per group. The group is the unit of improvement.
Choose one fix that removes the most repeated cause. Add a required field, change a template, or move a check earlier in the process. Record the fix, the owner, and the date. One fix is easier to verify than a list of five.
Check whether the fix changed the count in the following weeks. Use the same unit, definition, and time window. Record the before and after counts. A drop in one field with no change elsewhere is a useful result.
Keep a short list of fixes that did not work and why. A failed fix is evidence about the process, not a reason to stop measuring. Record the attempt and the result. The next team should not repeat the same change.
Give the team the weekly result in a form that respects the work. Show the trend and the fix, not a ranking of individuals. Record the summary and the questions it raised. People report causes honestly when the log is used to improve the work.
Close the loop with the client and the queue
Send the client a short monthly note with the volume, the rework rate, the top cause, and the fix in progress. Record the recipients and the date. A client who sees the cause can correct an input that the desk cannot change. A short note with one request is more likely to be read than a full report.
Update the work instruction when a fix changes how the task is done. A log that improves the process but leaves the instruction stale creates a new source of rework. Record the instruction version and the effective date. The person doing the work should not have to remember which version was right.
Check the fix in the actual queue rather than only in the log. Sample a few orders after the change and confirm the field or check is being used. Record the sample and the result. The log shows what people report, while the sample shows what happens.
Retire a cause category when it no longer appears for a full quarter. A category kept for history can make the log look more complex than the work. Record the retirement and the date. The cause list should stay short enough to use at the moment of correction.
Review the log's own cost once a quarter. If each entry takes more than a few minutes, the form is too long or the data is hard to find. Record the median time and one simplification. The log should pay for itself in fewer repeats.
For the next step, compare this workflow with Build a peer review rotation for a small offshore quality team and Build a daily reconciliation between a help desk and a billing queue. For access and oversight, use the authoritative planning reference alongside your own policy.
