Loading…
AI performance — containment, escalation and reasons, and draft edit rate — computed directly from the interaction and staff-question logs. Every table exports to CSV.
Share of the lead cohort the AI advanced to a later stage with zero human turns. Cohort = non-spam, non-manual leads created in range and matured ≥ 7 days; a staff outbound message breaks containment (AI-02; approvals leave sender_kind='ai_agent' so they don't count as turns).
No data for the selected range.
Share of escalatable AI interactions the agent escalated to a human, by agent (AI-03a). Denominator = interactions with kind in reply_draft/qualification/follow_up/staff_answer.
| Agent | Interactions | Escalated | Escalation % |
|---|---|---|---|
| lead_concierge | 30 | 4 | 13.3 |
| staff_coach | 1 | 1 | 100 |
Distribution of escalation reasons across AI interactions and staff-coach questions (AI-03).
| Reason | AI interactions | Staff questions | Total |
|---|---|---|---|
| low_confidence | 3 | 0 | 3 |
| unexpected_stop_reason: max_tokens | 1 | 0 | 1 |
| agent_escalation:no_kb_coverage | 1 | 0 | 1 |
Share of sent AI drafts a human edited before send, by agent and kind (AI-05). A falling rate is the earn-auto-send signal.
No data for the selected range.