Help center Reporting and performance Use KPI scorecards to improve support performance Updated Wednesday, October 28, 2026 · Marcel Gravelle Scorecards combine configured performance measures into a consistent view of support work. Use them to investigate service commitments, workload, response quality, and customer feedback. For B2B operations, the value comes from connecting the numbers to account context and case complexity—not from ranking people on a single unexplained score. At a glanceFor: support leaders, supervisors, team leads, and reporting users. Applies to: Supportbench Core.Start here: Reporting → KPI Scorecards → Agent scorecards, Team scorecards, Division scorecards, or Organizational scorecard Agree what the scorecard means · Open and compare a scorecard · Understand common metric definitions · Understand the overall weighted score · Turn a trend into a useful decision · Run a weekly team review from the score to the case evidence · Use different measures for different operational questions · Common interpretation problems Agree what the scorecard means Reporting permissions control which case, agent, and team data you can inspect. Your administrator configures scorecard metrics, targets, weights, and optional measures. A scorecard is therefore meaningful only with its configuration, scope, and period.Decide whether you are reviewing an agent, a team, a division, or the organization. Do not compare different scopes as if their case mix were identical.Agree how to interpret service times, satisfaction, and AI quality measures. Confirm business-hours/SLA behavior where it affects the recorded values.Pair a scorecard with a case view when you need to identify the actual work behind a trend. A high-level score does not explain its cause. Open and compare a scorecard Open Reporting → KPI Scorecards. Choose Agent scorecards, Team scorecards, Division scorecards, or Organizational scorecard, then select the permitted scope.Choose Current day, Current week, Current month, Current quarter, or Current year. Use the applicable date, week, month, quarter, or year selector and confirm the period before comparing results.Read the metric values alongside their configured targets and weights. Check eligible observations and configured empty-value behavior. Some calculations return zero when there are no observations, and a zero can count as success for a metric configured that way; it is not proof of good performance.Open a trend for a supported metric row to see how it changes over time. Compare like periods and the same scope; not every row exposes a trend.Use Spreadsheet or Download PDF when you need to share the result. Record the scope and dates with the export so it remains interpretable later.Inspect representative cases or feedback behind an unexpected value before choosing an operational change. Understand common metric definitions MeasureHow to interpret itAverage close timeElapsed creation-to-closure minutes for cases closed in the reporting period. It is not simply the age of cases created during that period.80th-percentile close timeA nearest-rank percentile of eligible close durations. It describes the slower portion of the distribution; it is not the average.Closed within 7, 14, or 30 daysThe share of closed cases whose creation-to-closure duration falls within the applicable threshold.First response timeUses recorded SLA response minutes when available, otherwise the recorded response seconds converted to minutes. Verify the underlying service-time context before comparison.Activity response timePrefers available SLA-adjusted durations, with recorded raw durations as a fallback.Positive SLAMet observations divided by eligible observations with actual and expected values. It should not be interpreted as a count of all cases regardless of measurement eligibility.Survey response rateSurveys sent in the period that have a received-response timestamp, divided by surveys sent in that period. CSAT and NPS instead use responses received in the period. Small samples can be volatile.NPSPercentage of promoters (9–10) minus percentage of detractors (0–6), using valid 0–10 answers. Passive answers (7–8) remain in the valid-response population.CSATFavorable responses according to the configured survey scale. Confirm the scale rather than assuming every survey uses the same threshold.AI quality percentageEligible positive stored QA scores are evaluated against the configured metric definition; the implemented favorable threshold for the quality percentage is a score of at least 4. It is not a human-reviewed verdict on every case.Only the metrics enabled in your scorecard apply. A label alone is not enough to establish a calculation: compare its configuration and source population when investigating a discrepancy. Understand the overall weighted score The overall score is based on included metric weights. A metric that meets its configured target earns its full weight; a true failure earns zero. A warning-band or indeterminate result earns half its weight. Empty measures are excluded only when that metric is configured to exclude empty scores. The result is the rounded percentage of earned weight out of included weight, rather than a simple average of the displayed metric values.Example: three included measuresSuppose the configured weights are 50, 30, and 20. The first meets its target, the second misses, and the third is indeterminate. The earned weight is 50 + 0 + 10, giving 60% of the included total. This is an illustration of the scoring method, not a recommended configuration or a built-in target.Changing a target or weight can change the overall score without any change in the underlying work. Record configuration changes when interpreting a trend or comparing reporting periods. Turn a trend into a useful decision If response performance worsens, inspect queue coverage, assignment, and the work awaiting a first response. Choose an action with an owner.If close time rises, examine case complexity and customer dependencies. A long-running integration issue should not be judged as if it were a password question.If satisfaction changes, review the response count and the actual feedback. A few answers can move a percentage sharply.If AI quality is weak, inspect the relevant answer evidence and knowledge gaps. Improve and retest the source material instead of treating the number as an explanation.Review the result after the operational change using the same scope, period, and definitions. Run a weekly team review from the score to the case evidence A scorecard is a starting signal. Suppose a team’s first-response result worsens while its overall weighted score remains similar. The useful question is which work changed and whether customers experienced a missed commitment—not whether a single score is “good.”Open Team scorecards, choose the intended team, and select the week you are reviewing. Record the period and scope before exporting or comparing.Inspect the response metric, its target/weight and eligible observations. Check whether a changed metric configuration or a very small population explains part of the movement.Where the metric row supports a trend, select that row to open it. A trend shows movement; it does not automatically produce the case-level cause of that movement.Use a separate case view to inspect representative work from the relevant scope. Compare queue coverage, assignment, customer waiting time and actual activity. Do not assume every long duration indicates agent inactivity.Record one supported finding and a practical action with an owner. For example, if cases waited for ownership during a shift gap, agree a queue-coverage checkpoint. If the sample is too small, record that limitation instead of declaring a trend.At the next review, compare the same scope and definitions. Note any change in targets, weights, case mix or response population so the comparison remains meaningful.A completed review note“Scope: integration support team, selected reporting week. Response measure below configured target; eligible count reviewed. Sampled cases show a gap before assignment during handoff. Action: shift lead checks the unassigned queue before leaving. Review next week with the same metric definition and inspect exceptions.” This is an example recommendation, not a conclusion about your team. Use different measures for different operational questions Close duration can reveal long-running work, but it also reflects case complexity and customer dependencies. Survey percentages can reveal customer experience, but response count and survey timing matter. SLA results concern eligible measured commitments, which may not be every case. Read the definitions above before putting these measures side by side.For individual coaching, use evidence from comparable work and discuss context with the agent. For staffing, combine workload views, coverage and the relevant response measures. For knowledge improvement, read actual unsuccessful answers and missing procedures. A high overall weighted score can coexist with a serious issue in one measure because weights and inclusion rules determine the total. Common interpretation problems ProblemChecks and correctionTwo reports disagreeCompare metric definition, date basis, scope, permissions, filters, and snapshot time.A score changed but the team did notReview targets, weights, optional empty metrics, and eligibility changes.A small team’s result swings sharplyCheck the denominator and sample size before treating the movement as a stable trend.A metric is missing or indeterminateCheck enabled metrics, permissions, available observations, and the reporting period.A trend does not openOnly supported metric rows provide a trend; use the available report/export or an appropriate data view. Continue learning Configure KPI scorecardsReview KPI trendsInvestigate the underlying case work with viewsSchedule operational reportsImprove knowledge and AI answers Was this article helpful? Yes No Thank you for your feedback.