Working Data skill
Measure support quality
Build a balanced support-quality measurement system covering accuracy, resolution, effort, safety, communication, and learning.
Measure whether customers receive correct, safe, durable help. Combine outcome signals, structured review, and customer evidence instead of optimizing only speed or satisfaction.
Inputs
- Gather support goals, channels, case types, service levels, reopen and escalation data, feedback, and quality reviews.
- Identify privacy, labor, accessibility, language, outsourcing, and regulated-work constraints.
- Define the decisions the scorecard should support and decisions it must not make alone.
Procedure
- Define quality dimensions such as correctness, completeness, resolution, effort, safety, ownership, clarity, and accessibility.
- Write observable rubric anchors with examples and counterexamples for each dimension.
- Design a representative sampling method across channel, reason, risk, customer state, language, and outcome.
- Combine review results with reopen, repeat contact, escalation, defect, effort, and customer feedback signals.
- Separate agent-controlled behavior from product, policy, staffing, and dependency failures.
- Train reviewers and run blinded calibration until material disagreements are resolved.
- Record evidence and uncertainty for each review while minimizing customer and employee data.
- Build team and system views that reveal patterns without relying on simplistic individual rankings.
- Create a correction and appeal path for mis-scored or context-missing cases.
- Link findings to coaching, documentation, product, policy, tooling, and staffing actions, then verify change.
Boundaries
Do not use hidden surveillance, protected traits, private message content beyond legitimate need, or one metric as an automatic employment decision. Never reward speed that sacrifices safety or accuracy. Obtain required consent and review for employee monitoring and recorded interactions.
--- name: measure-support-quality category: data description: Build a balanced support-quality measurement system covering accuracy, resolution, effort, safety, communication, and learning. Use when a team needs a fair scorecard, calibration process, or evidence for support improvement. --- # measure-support-quality Measure whether customers receive correct, safe, durable help. Combine outcome signals, structured review, and customer evidence instead of optimizing only speed or satisfaction. ## Inputs - Gather support goals, channels, case types, service levels, reopen and escalation data, feedback, and quality reviews. - Identify privacy, labor, accessibility, language, outsourcing, and regulated-work constraints. - Define the decisions the scorecard should support and decisions it must not make alone. ## Procedure 1. Define quality dimensions such as correctness, completeness, resolution, effort, safety, ownership, clarity, and accessibility. 2. Write observable rubric anchors with examples and counterexamples for each dimension. 3. Design a representative sampling method across channel, reason, risk, customer state, language, and outcome. 4. Combine review results with reopen, repeat contact, escalation, defect, effort, and customer feedback signals. 5. Separate agent-controlled behavior from product, policy, staffing, and dependency failures. 6. Train reviewers and run blinded calibration until material disagreements are resolved. 7. Record evidence and uncertainty for each review while minimizing customer and employee data. 8. Build team and system views that reveal patterns without relying on simplistic individual rankings. 9. Create a correction and appeal path for mis-scored or context-missing cases. 10. Link findings to coaching, documentation, product, policy, tooling, and staffing actions, then verify change. ## Boundaries Do not use hidden surveillance, protected traits, private message content beyond legitimate need, or one metric as an automatic employment decision. Never reward speed that sacrifices safety or accuracy. Obtain required consent and review for employee monitoring and recorded interactions. ## Done - The scorecard has observable rubric anchors and a representative sampling plan - Reviewer calibration, disagreements, corrections, and evidence are recorded - Customer outcomes and system causes are visible alongside handling behavior - Improvement actions have owners and later checks show whether quality changed