Back to all posts
support qa

What Is a QA Scorecard, and What Does the Score Mean?

TJ Baker 6 min read

A QA scorecard is the form a support or contact-center team uses to grade a conversation. It lists the criteria an agent is expected to meet, such as verifying the customer, answering the question that was asked, and following the refund policy, and it gives each criterion a point value. The QA score is the share of available points the conversation earned, usually shown as a percentage.

That percentage is what gets reported upward, averaged by agent and by month, and compared with other teams. It carries less information than it appears to, because rules such as how not-applicable points are counted change it without appearing on many scorecards.

What goes on a scorecard

Criteria are the individual things being checked. Some are binary (the agent verified the account or did not). Others use a scale, such as one to five for tone or clarity.

Point values, or weights, set how much each criterion contributes. Giving resolution 40 points and greeting 10 says resolution matters four times as much. That ratio is a prediction about the customer, and it is usually set once and not revisited.

Critical items, sometimes called auto-fail items, override the total. If an agent discloses account details to the wrong person, the conversation scores zero however well everything else went. These are usually compliance or security checks.

Not-applicable rules decide what happens when a criterion does not apply to a conversation. A billing conversation has no troubleshooting step to grade. The scorecard has to do something with those points, and the choice changes the score.

How the score is calculated

Take a scorecard with four criteria: greeting worth 10 points, identity verification 20, resolution 40 and tone 30, for 100 in total.

One conversation earns the full 10 for greeting, 20 of 40 for resolution and the full 30 for tone. Verification was not needed, because the customer asked a general question that touched no account. That is 60 points earned.

The score for that conversation depends on the not-applicable rule:

  • If the 20 verification points are removed from the total, the score is 60 out of 80, or 75 percent.
  • If not-applicable counts as a pass, it is 80 out of 100, or 80 percent.
  • If not-applicable counts as zero, it is 60 out of 100, or 60 percent.

The conversation is identical in all three cases. The whole 20-point spread comes from one rule, and a report that shows only the percentage does not say which rule was used. Removing not-applicable points from the total is the most defensible of the three, because it grades the conversation only on what the conversation contained. It is not the only practice in use, and a scorecard that does not state its rule leaves each evaluator to choose one.

Critical items add a second rule of the same kind. If an agent's conversations average 90 and one in twenty hits an auto-fail item, the monthly average falls to about 85. The average then mixes two different facts, how well the agent handled conversations and how often a compliance line was crossed, into one number.

Why there is no useful average QA score benchmark

Teams search for an average QA score to compare themselves against. A published average only means something if every team behind it used identical criteria, weights, not-applicable rules, critical items and scoring practice. Published averages rarely say which rules produced them, and the example above moved one conversation by 20 points with a single rule. Two programs that both report 85 percent cannot assume they measured the same thing, so a team at 78 against a published 85 learns little about its own customers.

The comparisons that hold up are internal. The same scorecard, applied the same way, can be compared across months, across teams and across agents. When the scorecard changes, the series breaks, and the change should be dated in the report.

What the single number hides

COPC Inc., which publishes a widely used contact-center operations standard, published a worked example in 2016 of a program reporting 90 percent overall quality while scoring 70 percent on the attributes that mattered most to customers. The gap came from criteria such as correct openings and hold procedures, which agents nearly always pass and which lift the total. COPC's recommendation was to report customer-critical, business-critical and compliance-critical accuracy as three separate metrics instead of one rolled-up score. Its 2022 Global Benchmarking Series on quality assurance, a survey of more than 900 customer-experience executives, found that 22 percent of organizations said they do not analyze the relationship between QA data and customer satisfaction at all.

In COPC's example, the criteria closest to the customer's experience carried less of the total than the procedural ones, which is why the total and the customer-critical score came apart.

What to report instead of one number

Report the pass rate for each criterion alongside the total. A team can report 84 percent overall while its agents answer the customer's actual question in 62 percent of conversations, and only the per-criterion figure shows that.

Separate compliance from quality. Report critical-item failures as a count, and keep them out of the quality average, so that a single breach does not read as a month of weaker conversations.

Write the not-applicable rule on the scorecard, and apply it the same way for every evaluator and every automated scorer. The post on designing a QA rubric covers the other rules a scorecard should state in writing, and how to check that evaluators read them the same way.

Test the weights against what happened to the account afterwards. A criterion that scores the same for customers who renewed and customers who left is adding points without separating the two groups. The earlier post on scorecard weights sets out how to run that comparison.

The last change is the hardest to make with a monthly sample of a few conversations per agent, because a handful of scores per criterion shows almost nothing, and because the score and the outcome sit in different systems. Resonant IQ scores every conversation against the same scorecard and ties each score to the conversation and the account it came from, so each score sits on the account's timeline beside what the account did next.

What this does not settle

Separating the score into its parts shows where the points come from. It does not show which criteria matter to customers. That still requires the comparison against outcomes, rerun as the product and the customer base change. A per-criterion report also depends on consistent scoring: if two evaluators read the same conversation differently, the pass rates carry that disagreement, and calibration between evaluators comes before any of the reporting above.

Sources

Stop guessing which accounts are slipping.

Join the founding cohort and lock your rate for 24 months while we build the evidence layer with you.