Skills Quality assurance
Outcome validation for QA scores
Leadership eventually asks whether the QA programme predicts anything that matters. Without an answer, the scorecard is an expensive habit. With a dishonest one — a correlation on forty evaluations, or CSAT treated as ground truth — it becomes an expensive habit that also misdirects coaching.
npx rulebase-skills install cx-outcome-validationWhen to use it
Reach for this when someone says any of these — they are the phrases the skill itself triggers on:
- “do QA scores predict anything”
- “validate the scorecard”
- “correlate QA with CSAT”
How it works
The method, in the order the skill runs it. The full procedure — tables, worked examples and the edge cases — is in the skill itself.
Start with the outcome, not the score
Pick one or two outcomes the business already tracks and would act on.
Assemble the analysis set
You need evaluated conversations linked to downstream outcomes on the same population and time window.
When the sample is too thin
There is no universal minimum, but several conditions make any correlation uninterpretable: Fewer than a few dozen evaluated conversations in a segment, Outcome events rarer than the evaluation count, Most agents with one or two evaluations, Outcome measured on a different population.
Analyse without invented thresholds
Do not declare success or failure against a made-up correlation cut-off. Report what the data show: Direction, Strength qualitatively, Which score components move with the outcome, Time stability.
CSAT-specific caveats
CSAT is the outcome everyone reaches for and the one most likely to mislead.
Reverse causation and other traps
A relationship in the data is not a recommendation to tighten the rubric.
Related skills
Free and open source, and vendor-neutral — it reads the conversations from whichever helpdesk you already run. Browse all 149 skills · connect your helpdesk over MCP · source on GitHub
