An AI score can start a conversation. It cannot replace a thoughtful manager or the context behind a call.
Look at patterns, not isolated moments.
A bad call can happen for many reasons: a difficult customer, technical issues, conflicting instructions, or an unusual case. Use trends and multiple examples before concluding an agent needs support.
Make the feedback observable.
“Improve quality” is hard to act on. “Ask one clarifying question before offering a solution” is concrete and coachable.
Separate the person from the metric.
Discuss behaviors and process rather than identities or assumptions. Ask the agent how they understood the interaction before offering a perspective.
Agree on one experiment.
Set a simple practice goal for the next week. Review a few follow-up examples together and discuss what changed.
A useful test: can the agent explain exactly what they will try differently on their next call? If not, the feedback may still be too vague.