What Is Speech Analytics? What Customer Calls Can Tell You
Speech analytics turns recorded calls into searchable evidence and patterns. Learn how it works, what its signals mean, and where human judgement still matters.
Read moreBuild a call center QA scorecard with clear criteria, evidence-based ratings and fair exception rules. Follow a practical process from scope to calibration.
A useful call center QA scorecard measures observable behaviour, separates serious failures from coaching opportunities, and makes every rating explainable. This guide to a call center QA scorecard: how to build one takes you from choosing the right call type to testing a scorecard that reviewers and agents can use consistently.
Before writing questions, decide what the scorecard should help someone do. Identifying a compliance concern, coaching an agent and diagnosing a broken process are different decisions. Your scorecard can support all of them, but it should not merge them into an unexplained overall grade.
Define these foundations:
Start with a specific call type, such as an existing customer's billing enquiry. A single scorecard covering billing, complaints, sales and vulnerable-customer support is likely to contain irrelevant questions or overlook important differences.
For the wider operating process around your scorecard, see the guide to building an ongoing call center quality management system.
List what must happen for the call to be handled correctly. Use approved policies, service procedures, applicable obligations and recurring customer problems—not just the current script.
Separate requirements into three groups:
Ask the policy owner to confirm mandatory controls. A QA team should not have to interpret legal obligations independently.
Also identify what sits outside the agent's control. An unavailable system or restrictive policy may prevent resolution even when the agent follows the correct process. Record that context without automatically treating it as poor agent performance.
Write each criterion so a reviewer can point to evidence in the call. Avoid labels such as “professional attitude” or “excellent ownership” unless you define the behaviours behind them.
Use this structure:
Criterion → expected behaviour → acceptable evidence → exceptions.
For example, replace “showed empathy” with “acknowledged the customer's stated concern before explaining the next action.” That gives reviewers something observable without requiring agents to repeat a stock phrase.
This starter framework shows how to translate broad categories into assessable questions. Adapt it to your call type rather than treating it as a universal template.
| Category | Scorecard question | Evidence to look for |
|---|---|---|
| Verification | Were the required checks completed before protected information was discussed? | Sequence of checks and disclosure |
| Understanding | Did the agent establish the reason for contact? | Relevant questions and confirmation of the issue |
| Accuracy | Was the explanation consistent with the approved policy? | Advice compared with the applicable policy version |
| Action | Was the appropriate action or escalation taken? | Stated action and supporting record, where available |
| Communication | Was the explanation clear and responsive to the customer's concern? | Plain language and relevant acknowledgement |
| Closure | Were the outcome and next steps made clear? | Summary, responsibility and any applicable timeframe |
Keep each question focused. “Was the agent polite, accurate and efficient?” combines unrelated behaviours and makes a single rating difficult to defend.
A simple rubric is easier to calibrate than a detailed scale with unclear distinctions. A practical starting point is:
Define partial credit separately for each criterion. Do not leave it to a reviewer's general impression.
Keep “not applicable” separate from “unable to assess.” A missing recording segment is an evidence problem, not proof that a requirement did not apply. Establish how both statuses affect any aggregate result, and make that treatment visible.
If you use weighted scoring, give greater importance to criteria tied to material harm or the call's purpose. Test whether changing the weights changes the assessment in sensible ways. Avoid rewarding a polished greeting enough to conceal incorrect advice.
Some failures need escalation regardless of the rest of the conversation. Define these explicitly with compliance, legal and operational owners.
For each critical-failure rule, document:
Do not make every missed script phrase a critical failure. Distinguish exact wording that is genuinely required from guidance where equivalent language is acceptable.
Preserve the underlying findings even when a critical failure changes the call's overall status. The agent may still need separate coaching on diagnosis or clarity. The guide to call center regulatory compliance standards and evidence explains why traceable evidence matters alongside the rule itself.
Before using the scorecard for formal performance decisions, ask reviewers to assess the same varied set of calls independently. Include straightforward calls, escalations, interruptions and cases where a criterion does not apply.
Compare disagreements at criterion level rather than debating only the final grade. Ask:
Revise the rubric, then test it again. Keep examples of clear passes, failures and borderline cases as reviewer guidance, with appropriate access controls for customer information.
Calibration should improve the scorecard—not simply train everyone to agree with the most senior reviewer.
Each finding should lead to a suitable response. A knowledge gap may need coaching; an unclear policy needs an owner; a system limitation needs a process or technology fix. Do not send every issue to the agent as a training task.
Before launch, establish an appeal route and version control. Record which scorecard and policy version applied to each assessment. When a requirement changes, communicate the change rather than silently judging older calls against new expectations.
Automation should follow these definitions, not replace them. Ask any provider to demonstrate how ratings map to your criteria, how evidence is presented and how uncertain assessments receive review.
CX Insights supports Auto QA against an organisation's own scorecards, alongside compliance checks, performance insights and audit trails across voice, email and chat. A clearly defined rubric gives teams a sound basis for evaluating automated scoring and deciding where human judgement remains necessary.
Include requirements for verification, understanding, accuracy, appropriate action, communication and closure where relevant. Every criterion needs a rating definition, evidence requirement and exception rule. Keep critical-failure handling distinct from routine coaching findings.
Use shared definitions for behaviours that genuinely apply across teams, but adapt criteria to the call type. A sales call and a complaint escalation have different purposes and obligations. Too many irrelevant questions make a universal scorecard harder to use fairly.
Treat customer satisfaction as complementary evidence, not a substitute for reviewing the interaction. A customer may dislike a correct policy decision, while a friendly conversation may still contain inaccurate advice. Compare the signals without assuming they measure the same thing.
It is ready for evaluation when its criteria are observable, exceptions are documented and reviewers can apply it consistently. Test automated findings against reviewed examples before relying on them for consequential decisions. Retain a process for disputed ratings and unclear evidence.
Find out how it fits your team’s workflow.
Offices
| | |