QA sampling calculator

AgentsCalls per agent / monthConfidence levelHow often this method lands inside the margin below.Margin of errorHow far the result may sit from the truth, in points.

Compare againstThe conventional workload, calls reviewed per agent per month.

Cochran's formula with a finite-population correction, at p = 0.5. That assumption maximises variance, so every figure here is the conservative one.

margin of error by sample size · 4,200 calls · 95% confidence

Margin of error by sample size
Calls reviewed	Margin of error (points)
100	±9.7
200	±6.8
300	±5.5
600	±3.7
1000	±2.7
1600	±1.9

result95% confidence

353calls to review, out of 4,200

coverage needed8.4%

per agent12

margin of error±5 pts

vs. 10 per agentthe convention

calls reviewed300

coverage7.1%

margin of error±5.5 pts

At this sample, a month-on-month move of less than 11 points cannot be distinguished from noise. Improvements smaller than that are not measurable, in either direction.

Or skip the sampling question entirely and audit everything, which is what most people here end up doing, because at these volumes it costs less than the analyst time the sample was rationing.

How many calls is enough?

Keep reading

Pack planner

Glossary: margin of error

vs. spreadsheets