What Manual Call Review Actually Samples
Manual call review samples a fraction of a fraction. A team leader listens to five or six calls per agent per month, chosen by whoever pulled them, and scores them against a form. Everything not sampled is unexamined, and the calls that went wrong are not more likely to be in the sample.
The sampling problem is the argument, and it is stronger than any accuracy claim about automated scoring. A quality programme that reviews a tiny, non-random slice cannot detect a pattern, coach reliably, or defend a complaint. What stays out of the published version is your scoring accuracy claims: agreement rates between automated and human scoring vary by form and by language and should be presented as a calibration result from the customer's own data.
The template walks one call through seven scenes: one on how calls are currently selected for review, two on what full coverage changes about what can be seen, one on what is actually scored automatically and what is not, one on how a flagged call reaches a team leader, one on coaching from patterns rather than incidents, and one on where human review stays essential.
