QA Sample Size: What Four Scored Calls Can Tell You

Most quality programs score four to eight conversations per agent per month. Those scores then decide coaching plans, team rankings and performance reviews.

The range around one agent's score

Take an agent who passes four scored conversations out of four. Treating each conversation as an independent pass or fail, the 95 percent confidence interval around that perfect score runs from 51 to 100 percent. An agent who passes three of four has an interval from 30 to 95 percent. The two ranges overlap across 44 points, so the sample can't say which agent is better.

More conversations narrow the range slowly:

4 of 4: 51 to 100 percent
5 of 6: 44 to 97 percent
18 of 20: 70 to 97 percent
90 of 100: 83 to 94 percent

Those intervals assume every scored conversation is a random, independent draw. In practice samples are rarely random and one evaluator's judgment varies from call to call, so the real ranges are wider.

Measuring quality for the whole contact center to within 5 points at 95 percent confidence takes a random sample of about 400 conversations, largely regardless of the size of the center. In a 50-seat center that is eight per agent. The center-wide number built from all 400 is sound, but each agent's score still rests on eight.

How often teams actually score

COPC surveyed more than 900 customer-experience executives in late 2021 for its 2022 benchmark on quality assurance. 88 percent of organizations monitored each agent at least once a month. The fieldwork is nearly five years old, so it shows monthly per-agent monitoring was standard then, not that the figure holds today.

In a Call Centre Helper discussion on how many calls to monitor, one contributor puts the common range at four to eight a month. Another describes scoring one call per agent per week alongside a one-to-one performance session with the line manager. Garry Gormley of FAB Solutions calls five to six the optimal number, since fewer than five gives an incomplete picture. J Gatenby cites the 400-call figure for center-wide accuracy and adds that sampling theory doesn't carry over to individual agents. These are practitioner opinions, and they range from four a month to fifty.

What a small sample can be used for

Coaching from a specific conversation is sound. A manager who reads a call and finds a commitment made and not kept has evidence of that from the call itself, and the sample size has no bearing on it. The earlier post on scorecard weights placed the value of a handful of calls in the same spot.

Three common uses go beyond what four conversations support:

Ranking agents against each other, when their intervals overlap almost entirely.
Month-over-month trend lines for one agent, when the change between months is smaller than the range around each month.
Pass or fail thresholds that attach a consequence to the agent rather than to the conversation.

The four-call count is usually set by evaluator capacity, not by what the score is asked to decide. A team can keep that sample and stop using it for those three things, with no new tooling.

Scoring every conversation

Automated scoring removes the sampling problem: every conversation is scored, so an agent's rate for the period is counted rather than estimated. Resonant IQ scores each conversation as it arrives against a versioned rubric, and a reviewer can correct any score without erasing the original. Each score links back to the conversation and timestamp behind it, so a disagreement about an agent's rate becomes a question about specific conversations.

Scoring every conversation leaves the rubric's errors in place. A rubric with the wrong criteria, or with wording two evaluators read differently, gets applied to every conversation instead of four, and an automated scorer has its own error rate that has to be checked against human review. None of this shows that better agent scores change what customers experience, which would need its own evidence.

Sources

COPC Inc., "Global Benchmarking Series 2022: Contact Center Quality Assurance", May 2022. Survey of 900+ executives, 1 September to 18 December 2021. Page 26 for monitoring frequency.
Call Centre Helper, "Determining the Right Number of Calls Analysed", Jonty Pearce, published 12 April 2022, updated 22 July 2026. Practitioner responses, not survey research.
Intervals are 95 percent Wilson score intervals, treating each scored conversation as an independent pass or fail. The 400 figure is the standard sample size for a proportion at 95 percent confidence and plus or minus 5 points (384, rounded).