Severity Classification

Severity determines which tier handles the escalation and how fast it moves. Misclassification — calling a routine issue critical, or treating a critical issue as routine — is the most expensive error in escalation design.

Four severity levels, each with explicit criteria. S1 (Critical): customer data at risk, revenue impact exceeding $10K, system-wide agent failure, reputational exposure. Response: immediate, all-hands. S2 (High): SLA breach imminent, quality regression detected, cross-team coordination failure. Response: within 15 minutes, dedicated handler. S3 (Medium): single-agent task failure, non-critical deadline at risk, handoff validation failure. Response: within 2 hours, standard queue. S4 (Low): optimization opportunity, non-urgent process improvement, informational flag. Response: within 24 hours, backlog.

We audit severity classification accuracy monthly. Current accuracy: 91.24%. The remaining 8.76% are predominantly over-classifications (calling S3 issues S2) — which is preferable to under-classification but still wastes priority attention.