Call Center LogsAudit-ready taxonomy
A unified event model across voice, queues, agents, and governance
AfriRoute treats contact center telemetry as a first-class platform surfacenot an afterthought. The taxonomy below enables consistent analytics, QA, and compliance across human agents, AI agents, and telecom operator planes.
Operational consistency
One schema across channels and tenants so dashboards, exports, and alerts don't break per integration.
Compliance evidence
Policy decisions, approvals, and supervisor interventions emit auditable events with correlation IDs.
AI transparency
AI decision logs capture model/version, confidence, rationale, and hand-off semantics for audits.
Core principles
- Events are immutable, append-only, and tenant-scoped.
- Every event includes correlation fields (tenantId, traceId, conversationId/sessionId, callId when applicable).
- Supervisor and governance actions are first-class events (not hidden admin actions).
- AI actions emit explainability-friendly payloads and explicit human handoff markers.
- PII is minimized by default; sensitive payloads are tokenized/redacted with access controls.
Minimum event envelope
This is the shared contract across all log categories. Individual event types add their own payload fields.
eventEnvelope
eventId name tenantId timestamp source severity traceId conversationId/sessionId callId? queueId? agentId? policySetId?
Primary categories
Voice logs
Call lifecycle, media, recording, transcription, consent markers, and quality metrics (where available).
Queue logs
Enqueue/dequeue, SLA timers, overflow, reroutes, abandonment, and assignment decisions.
Agent logs
Presence, joins/leaves, transfers, notes, and explicit handoffs between AI and humans.
Wrap-up logs
Disposition codes, tags, outcomes, follow-ups, QA flags, and post-call actions.
Supervisor actions
Listen/whisper/barge, reassignments, SLA overrides, and coaching interventions.
Governance + AI decision logs
Policy allows/denies, approvals, kill-switch events, and AI decisions with rationale and confidence.
Backend contracts (to finalize)
If you are implementing the ingestion pipeline, define stable endpoints and schemas for: call lifecycle events, queue/assignment events, supervisor actions, governance decisions, and AI decision traces. This page documents the intended model; the specific API endpoints and event schemas should be published alongside the API reference when available.