Why reviewing a small sample of calls misses the risky ones

Why does sampling miss risky calls?

A sample is random, and risky calls are rare. If an agent makes hundreds of calls in a month and a reviewer listens to a handful, the chance that the one bad call is among them is low. The sample tells you about the average call. It tells you very little about the worst one.

What does that look like in practice?

A lender records every call, because it has to. The recordings pile up. A small QA team picks a few each week, listens, fills in a sheet and moves on. Everything else goes unheard: the call where an agent lost patience, promised something the lender cannot deliver, or said something a borrower will remember. Usually the first sign of trouble is a complaint. By then the call is weeks old, and the coaching conversation becomes a disciplinary one.

What are the options?

• Sample smarter. Pick calls by signals: very short calls, very long calls, calls after a complaint, calls with new agents. This is cheap and better than random, but it still leaves most calls unheard.
• Add reviewers. Works, but cost grows with call volume.
• Score every call automatically and have people review the ones flagged. This covers all calls and keeps human time for the cases that need judgment.

What should you check before trusting automated scoring?

Ask to see the reasoning behind each score, with the point in the call it refers to. Check how short or cut-off calls are handled. Confirm that a person can override a score. If a tool cannot show why it scored a call, do not use the score for coaching or incentives.

Where does Cadence fit?

Cadence is a call quality assurance tool for collections teams at NBFCs, banks, lenders and BPOs. It scores every recorded recovery call against a collections-specific scorecard, with timestamped reasoning, and turns the result into coaching and per-call incentives. See the Cadence site.