How does the AI decide to auto-reply or hand off?

For every incoming guest message, the AI runs a route decision: auto-reply, handoff, or ignore. The route depends on two things — what kind of question the guest asked, and whether the answer is grounded in knowledge you've provided. This page explains the three routes.

The decision

When a message arrives, the AI evaluates it in a split second:

1. What is the guest asking? — classify the intent: check-in, wifi, amenities, directions, a complaint, a pricing question, and so on.
2. Can I answer from known facts? — is the answer present in the facts pack built from your listing and rules?
3. Is it safe to answer? — does the question need your approval, a staff action, or sensitive data?

The answers to those three questions pick the route. The more complete your knowledge, the more often the answer to question 2 is "yes" — and the more messages auto-reply.

Auto-reply route

The AI replies on its own when the guest asks a routine question and the answer is grounded in your facts.

Typical auto-replies include:

• Check-in time and check-out time.
• Wifi name and password.
• Directions and how to get to the property.
• Amenity questions ("Is there parking?" / "Do you have a washer?").
• Simple confirmations ("Yes, your booking is confirmed for Friday").

The critical rule: the answer must be in the facts pack. The AI does not auto-reply to a question it can't ground — otherwise it would be guessing, which it is never allowed to do.

Handoff route

The AI hands off to your team when an automatic answer would be unsafe or incomplete:

• Needs your approval — financial matters, refunds, pricing, policy exceptions, emergencies.
• Needs a staff action — "bring a towel", "fix the AC", "the wifi isn't working".
• Facts are missing — a question the AI genuinely can't answer from what you've given it.

When it hands off, the AI sends the guest a holding message, creates a handoff ticket, and notifies your team with full context — so the right person picks it up fast.

Ignore route

Some messages don't need any reply at all. Pure social closings — "thanks!", "ok", "great" — trigger the ignore route so the AI doesn't chase the conversation into an endless loop of pleasantries.

This route is conservative: if a closing message actually contains a question, the AI routes it normally rather than ignoring it.