Outsourced CompanyPowered by StealthAgents

Philippines staffing research · Published · Updated

Why outsourced quality scores drift when volume rises

A quality score that moves with volume is measuring the month, not the team. Review sampling, definitions, mix, and rework together before drawing a conclusion.

A quality reviewer comparing queue reports with a colleague at a desk

Research findings

A quality score is usually stable in a steady month and then moves when volume rises. The movement is often read as a drop in effort or a decline in skill. Those explanations are possible, but they are not the first place to look. Volume changes the mix of work, the time available for review, and the way errors are counted. A score that ignores those changes measures the month, not the team.

The first mechanism is sampling. When volume is low, a reviewer can inspect a large share of the work. When volume doubles, the same reviewer inspects a smaller share and the sample shifts toward the cases that are easiest to check. The score then reflects the sample rather than the population. A stable review rate is more informative than a stable headcount.

The second mechanism is definition. Busy teams apply a quality definition faster and with less discussion. A borderline case that would have been marked as a defect in a quiet week is passed in a busy one. The direction of the drift depends on the definition, but the drift itself is predictable. Write the definition with examples so the boundary survives pressure.

The third mechanism is the mix. High volume often brings a larger share of routine work and a smaller share of complex work, or the reverse during a campaign. A score computed across the whole queue will move when the mix moves, even if each segment performs the same. Segment the score before comparing months. The total can hide two opposite trends.

The fourth mechanism is rework. When volume rises, rework competes with new work for the same capacity. A team that prioritizes new items will leave rework in the queue, and the quality score may improve because the rework is not counted as an error. The improvement is an artifact of the backlog. Count rework in the quality measure or the score will reward the wrong behavior.

The fifth mechanism is the review itself. A reviewer under pressure may check the fields that are easy to verify and skip the judgment that requires context. The error escapes the review and appears later as a customer complaint. The quality score looks better at the moment of review and worse at the moment of complaint. Measure both points to see the gap.

Discussion

A score that drifts with volume is not necessarily wrong. Some drift is real because speed and accuracy trade off under pressure. The question is whether the drift is understood and managed. A team that knows its score moves at high volume can plan a review surge or a temporary scope limit. A team that treats the score as fixed will misread the same movement as a people problem.

The first practical step is to report the score with the volume and the review rate beside it. A score of ninety-five percent means little without knowing how many cases were checked and how many arrived. Put the three numbers in the same view so the reader cannot separate them. The pairing prevents the most common misreading before it starts.

The second step is to define the score at the level where the work is done. A single queue-wide number hides the segment that changed. Break the score into two or three segments that match the customer promise and the process. Then a rise in volume that affects one segment is visible rather than averaged away. The segments should be stable across months so the comparison means something.

The third step is to separate detection from prevention. Detection is the share of errors found by review, and prevention is the share that never reached the customer. A team can improve detection by reviewing more, without preventing anything. A team can improve prevention by fixing a step, without finding more errors. Report both so the score reflects the work rather than the inspection.

Implications and limits

The fourth step is to review the score with the people who do the work. Ask which cases felt different during the busy period and which rule was hardest to apply. Their answers usually identify the mechanism faster than the dashboard. Record the answer beside the score so the next review starts with context. A score without a story is easy to misread.

Avoid the temptation to reset the baseline after a busy month. A reset hides the drift and makes the next comparison meaningless. Keep the same definition and add a note about the volume. If the definition must change, publish the change and restate the prior months. Continuity is what makes a quality score useful over time, across teams, and across reporting cycles.

The final question is what the score is for. If it is a customer promise, measure the customer's experience. If it is a process control, measure the step that fails. If it is a provider comparison, measure the same work in the same way. A score that tries to be all three will drift whenever any one of them changes, so choose the purpose and design the measure around it.

Reading a score with volume

Start with the ratio of checked cases to arrived cases. A review rate that falls while volume rises explains part of the movement before any judgment about effort. Compare the rate across months and across segments. If the rate is stable and the score still moves, the cause is in the work rather than the sample. That distinction saves a long and unproductive conversation.

Then compare the mix of case types. A month with more exceptions will produce a different score even if the team performs at the same level. Show the mix beside the score so the reader can see the shift. Do not adjust the score silently; show the components and let the reader follow the reasoning. Transparency is more useful than a single corrected number.

Definitions under pressure

A quality definition should include three examples: a clear pass, a clear fail, and a borderline case with a stated decision. The borderline example is the one that keeps the definition stable under pressure. Without it, each reviewer resolves the boundary differently as the queue grows. Review the examples each quarter and replace any that no longer match the work.

Train new reviewers on the examples rather than the prose. A short session with five real cases will align them faster than a document. Then compare their decisions with an experienced reviewer for the first month. Record the disagreements and use them to refine the examples. A definition that produces consistent decisions is ready to survive a busy period without a rewrite.

Segmenting the queue

Choose segments that match how the work is assigned, not how the report is organized. If one person handles two customer types, the segment should follow the customer type. If the process changes at a step, the segment should follow the step. The segmentation should be stable enough to compare across months and simple enough to explain in one sentence.

Report the segment scores with the same review rate and volume detail as the total. A segment with ten cases is not comparable to one with a thousand, so show the count. Avoid ranking segments by score alone. The purpose is to find where the process changed, not to create a league table that invites competition between teams or shifts.

Counting rework and repeat contacts

Rework is the clearest signal that a quality score is incomplete. A case that had to be redone consumed capacity and often reached the customer twice. Count it as a quality event even if the final output was correct. Then compare the rework count with the error count to see which is driving the score. A team with few errors and much rework is not performing well.

Repeat contacts are the customer-side version of the same signal. A customer who asks the same question twice is telling you that the first answer did not resolve the request. Track repeat contacts by reason and by segment. The pattern usually points to a template or a step that needs a change. Add the count to the quality review so it is not treated as a separate support metric.

Keeping the measure honest

Publish the definition, the review rate, the volume, and the segments together. A reader who sees only the score will draw a conclusion the data does not support. A reader who sees the components can ask a better question. The extra context is small compared with the cost of a wrong decision. Make the full view the default rather than an appendix.

Review the measure itself once a year. Ask whether it still matches the customer promise and the process. Retire fields that no longer change a decision and add fields that the work now requires. Then restate the prior year with the new definition if possible. A measure that is never reviewed becomes a ritual rather than a control, even when the reports look tidy.

Questions

Does a falling score always mean worse work?

No. Check the review rate, the mix, and the rework count first. A score can fall because more complex cases arrived, not because the team performed worse.

How many segments should we use?

Two or three stable segments are enough for most queues. More segments produce small samples that cannot be compared month to month.

Should rework count as an error?

Count it as a separate quality event. Rework shows capacity lost and customer contact repeated, even when the final output was correct.

When should the definition change?

Only when the process or the customer promise changes. Publish the change and restate prior months so the comparison stays honest.

Philippines staffing intake

Define the position before hiring begins.

Share the tasks, tools, schedule, and approval limits for your Filipino team member. The intake turns those details into a practical staffing brief.

Contact Us