When the AI stops
TravelPal's AI is built to stop. A complaint or a refund request halts it in that conversation and alerts a person on your team, topics you name are never handled without a human, and a door code is never guessed. This page is the full list: what stops the AI outright, and which gates you hold yourself.
TravelPal is an AI guest messaging and upsell tool for vacation rental property managers, connecting to 33 property management systems including Guesty and Hostaway, priced at $25 per unit per month. What happens after a stop is covered in how escalation works.
A complaint or refund request stops the AI in that conversation
When a guest seeks redress, a refund, compensation, a formal platform case, or an escalation threat, the AI stops replying in that conversation entirely and your team is alerted two ways at once: a team notification and an email to your inbox, both saying a person needs to pick the thread up.
The stop is sticky: the AI stays out of that thread until a person takes it over and does not creep back in when the guest says "ok thanks". The details live in what happens on refund requests.
TravelPal has stopped replying in this thread and alerted your team by notification and email. A person takes it from here.
An illustrative conversation; the guest and unit are invented.
The guardrails you set yourself.
Four controls, all in your dashboard, all in your own words and your own numbers.
Topics you name go to your team
You list topics the AI must never handle alone, in plain language. A reply touching a named topic goes to your team as an internal note instead of to the guest, no matter how confident the AI was.
A guest asked how the pool renovation will affect their stay. The reply was held for your team instead of sent to the guest.
Styled after the dashboard's internal notes; the guest and property are invented.
The confidence bar
The AI scores its own reply, and a reply below the bar you set becomes an internal note for your team instead of a message to the guest. A reply that could not be scored at all is never sent.
Your cleaning crew's notice
You set how much notice your crew needs, and the AI will not sell a mid-stay cleaning inside that window, measured on the property's local clock. More in working with your cleaning schedule.
When each upsell may be offered
Every upsell has its own switch, off until you turn it on, and you set when it may be sold: how far ahead, and how close to the event is too close for housekeeping to react.
The stops that are always on
Some lines the product draws for you, on every account.
Door codes are never guessed
When a guest asks for a code you withhold, the AI files a request to your team and tells the guest the team has been notified. One unit's codes are never used to answer another unit's guest. See how door codes are handled.
No AI-initiated messages overnight
Messages the AI initiates are held overnight and sent at a civil hour, measured on the clock at the property. Replies are different: a guest who writes at 3 AM gets an answer at 3 AM. Across a portfolio that spans time zones, each property runs on its own clock; see multi-market portfolios.
It does not talk over your team
If a person on your team answers the guest first, the AI's pending reply is withheld, including replies already queued. If a teammate is already active in the thread, the AI stands down and writes nothing.
The calendar is checked before any sale
A late checkout is refused when another guest arrives that day, and an early check-in when another guest leaves that morning: the hours the cleaning crew needs for a turnover are never sold.
The same stops, as your team sees them fire:
A guest asked for a code you withhold. The AI stated none and guessed none, and told the guest the team has been notified.
A message the AI wanted to start overnight waits, on the clock at the property. A guest who writes at 3 AM still gets an answer at 3 AM.
Someone on your team answered the guest first, so the AI's pending reply never went out and it stood down in the thread.
The hours the cleaning crew needs for the turnover were never offered for sale.
Styled after the team's notifications; the guests and properties are invented.
Pause it in one click.
You can pause the AI's sending from your dashboard in one click, no engineer and no ticket. The pause catches replies already queued, because each send re-checks the setting at the moment it fires.
What your team still sees while it is paused
Your team still sees, privately, what the AI would have said to each guest, so switching it off costs review time, not coverage.
Every fact it uses is yours to search, correct, or delete in the knowledge base.
Held replies stay readable by your team, and nothing reaches the guest while sending is paused.
Styled after the dashboard's pause switch.
Safety FAQ
What happens when a guest asks for a refund?
The AI stops replying in that conversation and your team is alerted by a team notification and an email, both saying a person needs to pick the thread up. The stop holds until someone on your team takes the conversation over; it does not resume on its own when the guest says thanks.
Can I keep the AI away from certain topics?
Yes. You name topics in your own words, like anything about refunds or the pool renovation, and a reply touching one goes to your team as an internal note instead of to the guest, no matter how confident the AI was. The list is yours to edit from the dashboard, and naming no topics restricts nothing.
Will it ever guess a door code?
No. When a guest asks for a code your team keeps to itself, the AI does not state one and does not guess: it files a request to your team and tells the guest the team has been notified. One unit's codes are never used to answer another unit's guest.
Can I pause it?
Yes, in one click from your dashboard, no engineer and no ticket. The pause catches replies already queued, because each send re-checks the setting at the moment it fires. Your team still sees, privately, what the AI would have said, so pausing costs review time, not coverage.
Updated