Conversation Review
Inspect the thread, AI reasoning trail, CRM context, and approval actions before a reply leaves the console.
test
Agent Studio accepted the inbound Chatwoot message.
Discovery with Medium risk.
Retrieved Markdown KB/SOP/QA/compliance context.
HITL-only preview policy requires approval before send.
# Discovery Templates For vague greetings, ask one short probing question. Approved template: "Hi. What can we help with today: HVAC, plumbing, electrical, cleaning, or something else?"
# HITL Compliance Policy All live Chatwoot sends require human approval in the first preview. Refund, cancellation, complaint, payment, identity, and account-change conversations are medium or high risk. They must remain supervisor-gated and must not auto-send.
# Supervisor QA Rubric Score every AI draft for accuracy, SOP adherence, tone, compliance risk, and next-step clarity. Pass criteria: the draft answers the user, avoids overpromising, uses approved language, asks for verification when needed, and clearly identifies supervisor-g...
# Account Verification SOP Before sharing appointment, invoice, billing, or account-specific details, ask the customer to verify a phone number, email, ZIP code, invoice number, or booking reference. Do not expose private account details until verification is complete. If the u...
Draft includes retrieved KB/SOP context.
Live Chatwoot replies require supervisor approval.
- Provider
- Twenty CRM
- Status
- Twenty external
- Source
- Agent Studio adapter
- Mode
- Disabled or dry-run until Agent Studio env is configured
- Lifecycle
- Chatwoot visitor
- Last job
- n/a
- Value
- Unknown
- Risk
- Medium
- Owner
- Intake Pod
- Area
- Web chat