Fatal parameters: designing auto-fail checks that protect the brand
Some mistakes should fail a call no matter how good the rest was. How to pick them, word them and keep the list short.
A weighted scorecard tells you how good a call was. A fatal parameter tells you whether it should have happened at all. Treating the two differently is what keeps QA honest.
Keep the list short
Three to five fatal checks is plenty. Typical ones are identity verification before discussing an account, no threats or abusive language, no misleading statements about consequences or pricing, and consent where it is required.
Word them as observable behaviour
"Was respectful" is not auditable. "Did not threaten legal action, visits to the workplace or contact with family" is. The more concrete the wording, the closer AI and human scores will agree.
Act the same day
A fatal flag is only useful if someone sees it before the next shift. Route flagged calls straight to the team lead's review queue, with the timestamped evidence attached.