Skills Quality assurance
QA coverage economics
Coverage conversations usually start with a number someone heard — "industry standard is X%" — or with whatever volume the current team can physically grade. Neither is a strategy. Coverage is a purchase decision: you are buying precision on a measurement for a specific use, at a cost in reviewer time and agent attention.
npx rulebase-skills install cx-qa-coverage-economicsWhen to use it
Reach for this when someone says any of these — they are the phrases the skill itself triggers on:
- “how many QA evaluations do we need”
- “what coverage should we aim for”
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.
Step 1: Separate uses — they need different coverage
One coverage target cannot serve every use. Split them explicitly.
Step 2: Write the precision requirement in plain language
Before arithmetic, answer: Over what period? One bad week vs a quarter of evidence?.
Step 3: Work from agent exposure, not ticket share
For any use that names individuals: Start from headcount and eligible conversations per agent per period.
Step 4: Account for diminishing returns
The first few evaluations per agent teach you a lot; the fiftieth teaches less unless the rubric is very granular or the agent's work is heterogeneous.
Step 5: Price the programme honestly
Coverage cost is not only grader hours: Grading time per conversation, Calibration and adjudication, Agent and manager time, Opportunity cost.
Step 6: Say no to benchmark percentages
"Review 5% of tickets" is not a plan until you know: 5% of which frame (channel, queue, window)?, 5% distributed how (by ticket vs by agent)?, 5% supporting which decision?.
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
