How the calculation works
- z = 1.645 at 90% confidence, 1.96 at 95%, 2.576 at 99%
- Initial sample = z² × 0.25 ÷ (margin of error)²
- Corrected sample = initial sample ÷ (1 + (initial sample − 1) ÷ interaction population), rounded up
Why the 50% assumption
The 0.25 in the formula comes from assuming a 50% pass rate, which gives the largest sample for a given margin of error. It is the conservative choice when you do not yet know the rate you are measuring.
The finite-population correction reduces the sample when the population is small, so a team reviewing a few hundred interactions a month is not asked to review more than it handles.
Assumptions and limits
- Interactions are drawn at random from the whole population.
- The sample estimates one overall rate. Conclusions about individual agents need a sample for each agent.
- Stratify high-risk and rare contacts and sample them separately.