Research question and scope

This protocol measures how often a help desk record and a billing queue record disagree at the end of the client's business day. The unit is one linked case that exists in both queues. A disagreement means a required field has a different value or is missing on one side. The measure describes record alignment, not the correctness of either system.

Set the scope to one client, two queues, and a fixed period. Record the fields that must agree and the reference that links the records. If a field or reference changes during the period, split the sample at the effective date. A different rule creates a different denominator.

Keep the measure separate from a billing error count. A disagreement can be a display delay, a timing difference, or a label mismatch rather than a wrong amount. The purpose is to test whether the two teams describe the same case. Report financial corrections separately.

Name the decision the result will support before collection starts. The client may use it to add a required reference, change a status label, or move a check earlier. A report that only lists mismatches leaves the teams with the same problem. Write the intended action beside the scope.

Record the systems and the export times before drawing the sample. A billing export taken at noon and a help desk export taken at five cannot answer an end-of-day question. Note the time zone and the cut-off. The measurement is only as good as its timing.

Population and unit of analysis

Define the population as every case linked across both queues during the period. Include cases created by the desk and cases created by the client team. The unit is the linked case, not the ticket, the invoice, or the customer. A case with two invoices is one linked case when the reference is the case.

Include cases that were closed on one side and still open on the other. Mark the state on both sides. Exclude cases that were never expected to appear in the second queue. Report the excluded count and the reason. A closed-open pair is often the most useful finding.

Exclude duplicate links created by a retry or a reopened ticket. Link the duplicate to the original and count the case once. Duplicate handling is a separate efficiency question. State the rule before sampling so a second reviewer applies it the same way.

Decide whether a case with no billing record at all counts. Include it as a missing-record finding and report it separately from a value difference. A missing record can mean the work never reached billing, which is more serious than a stale status. The report should name the queue that should have created the record.

Decide how to treat a case that was corrected during the day. Score the end-of-day state and record the correction time. A case that was wrong at noon and right at five is a resolved difference, not a current one. The history is still useful for the weekly review.

Definitions that make the measure reproducible

Define a disagreement as a required field with a different value, a blank value, or a status that implies a different stage. Define a resolved difference as one corrected before the end-of-day export. Define an open difference as one still present at the export. Publish one example of each.

Define the required fields by the customer promise they support. The amount, the refund status, the promised date, and the owner are common. Record the field list beside the reference. A field that no customer sees may still matter for reporting, but it should be a separate group.

Define the authoritative queue for each field in writing. Billing usually owns the amount, while the help desk owns the request and the next step. Record the owner and the rule. A field with two owners will always produce a disagreement.

Define the end-of-day cut-off in the client's time zone and use the same cut-off every day. A moving cut-off changes which cases are counted. Record the time and the export reference. The rule should be visible in the report.

Agree how to classify a difference caused by a timing lag. A payment posted after the export is a lag, not a control failure. Record the lag and the time the value changed. Report lags separately so the team does not chase normal processing.

Sampling and data collection

Draw the sample from both queue exports for the same cut-off. Record the export date, the filters, and the row count on each side. Do not build the sample from cases that caused a complaint. A complaint-based sample overrepresents the failures that were noticed.

Capture the reference, the case, the customer, the amount, the status, the owner, and the last update time from each side. Add the promise and the customer-facing message time. Store both exports with the report. The reference and the two update times are the core evidence.

Check a small overlap sample with a second reviewer. If the reviewers disagree about whether two statuses mean the same stage, the definition is too vague. Rewrite the status map and retest before the full sample. Agreement matters more than sample size.

Record the time spent matching the exports. A report that requires reading every case note will not be repeated. Note the fields that cannot be matched automatically. The collection cost belongs beside the result.

Take a pilot of one day before the full period. Use the pilot to test the reference, the field list, and the export timing. Record what was missing and how long the match took. A pilot prevents a full sample from discovering a broken reference at the end.

Measuring disagreements

Report the share of linked cases with no difference, a resolved difference, and an open difference. Show the counts behind each share. The three groups describe different levels of alignment and should not be collapsed. A single mismatch rate hides the difference that was fixed before the customer noticed.

Break the result down by field, queue, and case type. An amount mismatch may cluster in refunds, while a status mismatch may cluster in plan changes. Compare only groups with the same reference and cut-off. A field with the same mismatch every day may need a shared status map.

Report the share of open differences that affected a customer-facing promise. A wrong internal status is different from a refund date the customer was told. Include the promise and the message time. The customer link shows where the risk becomes real.

Report the median time from the end-of-day export to the resolution. A same-day fix is a different control from a three-day fix. Include the longest open difference and its age. The distribution shows whether the daily check is actually closing work.

Compare the disagreement rate on days with a known release or payment run against ordinary days. A spike around a system event points to a timing or integration issue. Record the event and the count. The comparison helps separate a routine lag from a design problem.

Limitations and next action

The measure cannot show which system holds the correct value or whether the customer was harmed. It also cannot show a case that was never linked at all. Treat the result as a measure of record alignment. Note the systems and case types that were not included.

Give the result to the service lead and the billing owner with one change to test. The change might be a required reference at intake, a shared status map, or an earlier check. Record the effective date and repeat the measure with the same definitions. A controlled change is easier to judge.

Review the boundary cases before adding more fields. A few timing lags may explain a large share of open differences. Fix the export time or the status map first if that is the pattern. The goal is a shared view of the case, not a lower number by definition.

Repeat the measure after the change with the same scope and cut-off. Report the before and after counts together, including the sample size. Record any new system or queue that entered the period. A comparison with a changed denominator is not a comparison.