Guests do not rate your software. They rate how it felt to stay with you. That is worth holding onto, because most "AI for guest experience" conversations drift immediately into features and away from the handful of things guests actually notice.
They notice three: how fast you answered, whether the answer was right, and whether anyone seemed to know who they were.
Start with what guests actually complain about
Read a few hundred reviews across any portfolio and the negatives cluster tightly. Very few are about the AI-shaped problems people expect. They are about:
- Waiting. A question at 9pm answered at 9am next day. The issue is rarely the answer — it is the silence.
- Being told something wrong. Confidently wrong check-in instructions do more damage than a slow reply.
- Being treated as a booking reference. The guest on their fourth stay who is asked, again, for their arrival time.
- Friction at the moment of need. Wanting a late checkout at 7am and having no way to get one without a conversation.
Each of these is a response-time or a knowledge problem. Both are now solvable.
Strategy one: answer in seconds, in their language
Speed is the single highest-leverage change available, because it converts a complaint into a non-event. A guest who asks about parking and is answered in under a minute does not remember asking.
Language matters more than most operators assume. If a guest writes in Portuguese and is answered in English, they have been served, but they have not been hosted. Boom's guest messaging handles 40+ languages with a median first response measured in seconds rather than hours, and roughly 80% of messages are resolved without a human.
The point of that number is not the saving. It is that the fifth of conversations that do need a person now get one immediately, because your team is not working through a backlog of wifi passwords.
Strategy two: give it the property, not just the booking
This is where most implementations quietly fail. An assistant with access to the reservation can tell a guest when they arrive. An assistant with access to the listing — the door code, the parking bay, whether the pool is heated, which bin day it is — can actually host.
Before you evaluate any tool on its conversational quality, ask what it knows. A fluent model with thin context produces confident, wrong answers, which is the worst of the options available to you.
Strategy three: treat the upsell as a service, not a sale
Early check-in, late checkout, mid-stay cleans and airport transfers are things guests already want. The reason they go unsold is timing: offered at booking they are premature, offered at checkout they are irrelevant.
Handled well, the offer arrives at the moment the guest is likely to want it and can be accepted in a single tap. Handled badly, it is a discount email. The test is simple — if the offer would annoy you as a guest, it will annoy them.
Strategy four: close the loop on reviews
Most operators treat reviews as the end of the stay. They are more useful as an input.
A review mentioning shower pressure should not just receive a courteous reply — it should open a maintenance task before the next guest arrives. That is the difference between managing your reputation and improving the thing your reputation is about. Answering every review matters too, and it is exactly the sort of consistent, low-judgement work that should not depend on someone having a quiet afternoon.
Where to keep humans, deliberately
Automate the repetitive and the immediate. Keep people for:
- Anything involving money in dispute.
- Anything where the guest is already unhappy.
- Anything touching safety, neighbours, or a property issue mid-stay.
- Your highest-value or returning guests, where the relationship is the product.
Make that routing explicit rather than hoping the system escalates well on its own.
Measuring it honestly
Track median first response time, the share of conversations closed without a human, and review score — together. The first two without the third can be gamed by answering fast and badly. Watch the review text, not only the score: guests will tell you in words what a star rating flattens.
The operators who get this right are not the ones who automated the most. They are the ones who worked out which moments deserved a person, and then removed everything standing between their team and those moments.