Call Center Outsourced evidence brief · Desk review · Published
When does a rare disposition sample support a quality conclusion?
A sampling approach for uncommon dispositions: know the population, review high-risk cases as a census, report the count, and test reviewer agreement.

On this page
- Key takeaways
- Question and scope
- What the evidence supports
- Define a rare disposition
- Start from the population
- Separate zero from unseen
- Set a minimum evidence rule
- Stratify by risk
- Review every critical case
- Test reviewer agreement
- Widen the window
- Combine related dispositions
- Report the uncertainty
- Avoid a percentage from three cases
- Sample the near misses
- Use the finding for training
- Recheck after a process change
- Handle a client demand for a rate
- Limits of the evidence
- Manager review questions
- Conclusion
- Put this into a support lane
- Related operating guides
- FAQs
- Sources
Key takeaways
- Count the population before deciding whether to sample.
- Treat high-risk and critical cases as a census, not a sample.
- Report the count and the uncertainty beside any rate.
- Test reviewer agreement and recheck after a process change.
Question and scope
This brief asks when a sample of rare dispositions is large enough to support a quality conclusion. The population is one queue during a declared period, and the unit is a case with a disposition that occurs infrequently. The question is whether a small number of reviewed cases can justify a rate, a trend, or a coaching decision. The analysis covers sampling and reporting, not a universal minimum. Record the threshold and the reason it was chosen for this queue.
What the evidence supports
Federal data-quality guidance calls for documented sources, tested reliability, and stated limitations, and contact-center standards call for defined quality processes. Neither source sets a minimum sample for a rare outcome. The operating inference is to report the count and the uncertainty with the rate. A conclusion should name the evidence it rests on. A population count is the first number in every sample report.
Define a rare disposition
A rare disposition is one that appears in a small share of cases but carries a meaningful customer or compliance risk. The threshold should be set by the queue, not by a generic percentage. Record the disposition definition and the period. A category that is rare in one month may be common in another. Report the number reviewed beside every zero and every rate.
Start from the population
Count every case with the disposition before drawing a sample. The population tells you whether a sample is possible and what it can support. If only four cases exist, reviewing all four is better than sampling. A percentage from a population of four is a description, not a stable rate. The rule should name who can approve a census or a wider window. The rule should also say who reviews the sample when the regular reviewer is absent.
Separate zero from unseen
A month with zero reviewed errors may mean the work was clean or that the sample missed the problem. Report the number reviewed beside the zero. Do not present a zero error rate as proof of quality. The reader needs to know how much evidence produced the zero. Show the strata in the report so the reader can see where the effort went.
Set a minimum evidence rule
Write a rule that says what the queue will do with a small sample: review every case, widen the period, or report the count without a rate. The rule should depend on the risk of the disposition. High-risk work deserves a census, while low-risk work may wait for a larger sample. A critical case should have its own containment record, not only a quality score.
Stratify by risk
Group the rare dispositions by customer impact, regulatory exposure, and reversibility. Review every case in the highest-risk group and sample the lower-risk groups. This concentrates effort where a missed case matters most. The strata should be visible in the report. If agreement is low, rewrite the definition before reviewing more cases. The highest-risk group should be named in the report even when no case was found.
Review every critical case
Some outcomes are critical regardless of frequency: a privacy exposure, a wrong payment, or a missed safety step. Treat them as a census, not a sample. Report the count and the containment action for each. A critical case cannot be offset by a clean sample elsewhere. State the combined window and the process version for the whole period.
Test reviewer agreement
Ask two reviewers to score the same small set and compare their decisions. A rare disposition often depends on a judgment that is not written down. If the reviewers disagree, fix the definition before expanding the sample. Agreement is part of the evidence. Keep the subtype counts visible even when the codes are combined. A reviewer who cannot explain the decision should not be counted as agreement.
Widen the window
When the monthly count is too small, combine several months with the same process and mix. State the window and check that the process did not change inside it. A wider window can produce a usable count without pretending the month was representative. A range should name the method used to produce it. The wider window should exclude any month when the disposition definition changed.
Combine related dispositions
Some rare codes describe the same underlying event. Combine them only when the customer impact and the required action are the same. Record the combination rule so the next reviewer can reproduce it. A combined category can hide a serious subtype, so keep the subtypes visible. Say what the queue will review next when the count is too small for a rate.
Report the uncertainty
Use a range or a simple confidence statement rather than a single rate. Explain that a small sample can produce a wide range. The client may still want one number, so give the count and the range together. An honest range is more useful than a false precision. Near misses belong in the learning set but not in the critical-case count. A range with no stated method is only a guess with a wider label.
Avoid a percentage from three cases
Three reviewed cases with one error do not establish a thirty-three percent error rate. Report three cases and one error, and say what the queue will review next. The percentage can be added later when the count supports it. A rate without a denominator context invites the wrong decision. Use the example set in the next coaching cycle and note whether it changed behavior.
Sample the near misses
A near miss is a case that could have become a critical outcome but was caught. Review near misses alongside the rare disposition because they show where the control almost failed. They are often more numerous than the critical cases. Include them in the learning set, not in the critical count. Compare the new sample with the old definition before publishing a trend.
Use the finding for training
Turn the reviewed cases into a short example set for coaching. Remove customer identifiers and keep the decision point. Show the correct action and the consequence of the wrong one. A training example from a real rare case is more memorable than a general rule. Offer the count, the window, and the limitation in the same response.
Recheck after a process change
A new script, system, or authority rule can change how a rare disposition is used. Review a fresh sample after the change and compare it with the old definition. Do not carry the old rate forward without checking. The process change may make the category more or less rare. State what the evidence cannot prove before the client draws a conclusion. The client summary should show the count and the window before the percentage.
Handle a client demand for a rate
When the client asks for a monthly percentage, explain the count behind it. Offer a rolling window or a census of critical cases instead of a misleading rate. Put the limitation in the report, not only in the meeting. The client can make a better decision with the count visible. Record the answers with the population count and the review list.
Limits of the evidence
This brief cannot set a universal minimum sample or a statistical standard for every queue. The right approach depends on the risk, the volume, and the definitions. A small sample can guide a review without proving a rate. State the limitation and use the evidence to direct attention. The sample is defensible when the reader can see how it was chosen.
Manager review questions
Managers should ask how many rare cases existed, how many were reviewed, and whether the highest-risk group was a census. They should also ask whether the reviewers agreed and what changed after the last process update. The remaining question is which rare case will be reviewed next. Record the answers with the sample. A quality conclusion should rest on a known population and a tested definition.
Conclusion
A rare disposition sample supports a quality conclusion when the population is known, the high-risk cases are reviewed, and the count is reported beside any rate. Widen the window or combine related codes only with a stated rule. Test reviewer agreement and recheck after a change. The goal is a defensible learning set, not a precise number from too few cases. Review the sampling rule at the same time as the quality scorecard.
Put this into a support lane
Bring one queue, its reporting extract, and the manager who owns the operating decision.
Review the queueRelated operating guides
FAQs
Is a zero error rate from five cases a good result?
It is a limited result. Report the five cases beside the zero and say what the queue will review next before drawing a conclusion.
Can related rare codes be combined?
Only when the customer impact and required action are the same, and only with a written combination rule that keeps subtypes visible.