AI behavior
Configure supported assistance while preserving explicit human review.
On this page
Reply thresholds
Off, high confidence and aggressive describe supported reply behavior choices. Explicit human approval and automatic-reply-off take precedence; aggressive mode is not permission to bypass them.
Approval and escalation
Review auto-run and approval preferences together. Risky requests can be escalated. Refund/damage keyword handling is not a universal risk-off switch; it changes the supported keyword behavior, not your responsibility to review exceptions.
Settings that are not enforced
Sentiment preferences marked Not yet enforced do not create an operational guarantee. Do not assume a stored toggle changes classification or delivery. Check the specific label in the product.
Use AI responsibly
Verify dates, amounts, policies and public claims before customer use. Avoid entering credentials or sensitive guest details into marketing facts. Credits and channel readiness are separate from behavior preferences.