Quality Assurance Score

A support team handles more calls and messages than supervisors can review one by one, so it selects interactions and checks them against a scorecard. The resulting quality assurance, or QA, score summarises that review.

Criteria might cover accuracy, listening, policy steps, resolution and tone, and each item should describe observable behaviour because vague labels such as good service invite inconsistent judgements. NiCE describes a quality score as a measure of how well an interaction meets defined standards, but the standard is chosen by the organisation, so scores from different scorecards are not automatically comparable.

A basic score divides earned points by available points; if a reviewer awards 44 out of 50, the result is 88%. Weighting and critical failures can change the final score.

Zendesk's scorecard tools allow criteria and rating structures to be configured, and a team should document how not-applicable items are treated. Otherwise two reviewers may calculate different percentages for the same interaction.

Weighting should reflect customer risk, so a fictional travel support desk gives higher weight to factual booking accuracy than to greeting style and explains the weights to agents; the aim is not to make a target easy to hit. Track critical errors separately when they carry legal, safety or privacy risk, and keep any automatic-fail rule clear, proportionate and reviewed for false positives.

A fictional broadband team checks whether agents verified identity before discussing an account, separately from courtesy, because a warm tone cannot offset a failed critical safeguard. A fictional pharmacy help desk likewise checks whether medical advice is escalated under policy, so a courteous agent can still fail that item and the team treats it as a training and risk issue.

Sampling matters, because escalated calls may understate typical performance while easy cases may overstate it, so use a selection method that fits the question being asked. Calibration brings reviewers together to score the same interaction and discuss differences, which Zendesk describes as a way to improve consistency.

In a fictional case one reviewer marked an email down for a missing closing phrase and another did not, and calibration revealed that the scorecard never required it, so the team clarified its rule and recorded the interpretation. A single score is not a complete picture of the employee, since case complexity, channel mix and system problems matter, so look at patterns across enough comparable interactions and show the agent the observed behaviour and a specific alternative, because a bare 72% tells little about what to change.

When many agents fail the same confusing policy item, the procedure or knowledge base may need revision; a fictional delivery company that found low scores on refund explanations updated its customer-facing script, trained agents and checked later whether explanations improved, not just whether the number rose. Compare trends only when the scorecard and sample are stable, check automated review outputs against human review, restrict access to private customer details, and combine QA with customer feedback, repeat-contact rates and operational data, because high speed with poor accuracy is not good service.