QA Metal Detector
Sweep every scored call; spend human ears where the grader is probably wrong.
The job
When AI scores every recorded call but human review time is short, sends reviewers to the calls where the grader is most likely wrong or the stakes are highest. Proves it against a random sample, so more critical errors are caught per hour.
The moment
AI scores every recorded call, but the team only has a few hours for human review. The app sweeps the scored calls like a metal detector, looking for calls where the AI grader is probably wrong.
The beep rises as the sweep crosses a cluster of perfect scores followed by angry callbacks.
The lead points out that the team reviewed 400 random calls last month and found 6 wrong scores.
This afternoon, the detector's 40 picks found 17.
What it does
- Build the dig list (detector + control arm)
- Record a verdict
- Send an overturn to calibration
- Correct the official QA score
- Send the yield summary for the QBR
What you see
Metal-detector sweep with a rising beep, a dig list, and a buried-treasure shelf
What it moves
Critical errors and wrong scores caught per hour of QA review, compared with random checks, plus QA hours saved and errors caught per full-time QA reviewer.
Built for
- QA, training & knowledge
- Contact-centre operations
