The night auditor at a boutique hotel knows that room 314 has a compressor that hums, and that you don't put light sleepers in it. This is not in the PMS. It is not in a maintenance ticket. It is in her head, and in a sticky note on the monitor, and in the fifteen seconds she spends fixing a reservation at 9 p.m. on a Thursday.
She also knows that the corporate account from the medical device company always sends people who need early check-in and never says so on the booking. She knows which OTA guests are going to ask for an upgrade. She knows that when the wedding block checks in on Friday, you want your calmest person at the desk.
None of this exists anywhere in your technology stack. All of it is why your reviews are good.
The temptation, and why it fails
The temptation, when you start thinking about AI in hospitality, is to reach for the guest-facing thing. A chatbot. A voice agent that takes reservations. Something that talks to customers, because that's the part of the business that's visible.
It fails, reliably, and the reason is instructive: you are pointing your most exposed surface at a system that doesn't know anything.
The bot doesn't know about 314. It doesn't know about the medical device account. It has been given your rate table and your policies and a friendly tone, and it will use them to give confidently wrong answers to guests, in public, at scale — and every one of those interactions is a review.
Meanwhile the actual, defensible, hard-to-copy asset of your property — the accumulated operational knowledge of the people who work there — remains locked inside four humans, two of whom will leave this year.
Capture the knowledge before you automate the surface
The order matters enormously, and almost everyone gets it backwards.
Start by making what your team knows legible to a system. That means, concretely: a structured record of room-level quirks, not a sticky note. Account-level preferences captured at the account, not remembered by whoever's on shift. Service recovery history attached to the guest, so the second complaint from the same person is handled by someone who knows there was a first. Vendor performance recorded, so the sixth time the HVAC guy no-shows is not a surprise.
This is unglamorous data work. It has nothing to do with AI. It's the thing you have to do before AI is worth anything, and skipping it is why so many properties end up with an expensive chatbot and worse reviews.
Then let the machine do the machine work
Once the knowledge exists, the automation opportunities are enormous and almost all of them are internal, not guest-facing.
Guest messages — email, SMS, OTA, the review site — arrive in six channels and get triaged by whoever's around. A system can read all of them, classify by urgency and type, attach the guest's history and their room's history, and route to the right person with the context already assembled. The "my AC is loud" message from 314 arrives at the desk with a note: known compressor issue, three prior complaints, engineering ticket open since March, comp offered last time was a drink voucher.
That's not replacing your front desk. That's making your newest front desk employee as smart as your night auditor on day one.
Housekeeping dispatch, maintenance triage, group block prep, arrival lists flagged with preferences, the daily flash report that somebody currently builds by hand at 6 a.m. — all mechanical, all currently consuming the time of people you'd rather have on the floor.
The number
A 90-room property. Front office and management spend, conservatively, three hours a day across the team on message triage, status chasing, and report assembly. At a blended loaded rate of $28, that's about $30,000 a year.
Then the part that's bigger. Service recovery that happens before the review instead of after it. A single review's difference in rating, sustained, moves rate and occupancy in ways that are well documented in this industry and that any GM can feel in their bones — the ADR you can hold when you're the 4.7 property versus the 4.4 property is not a rounding error.
We won't put a fake number on that. But if you know your rate elasticity, you can, and you should, because it will dwarf the labor line.
What it's for
There's a version of hospitality technology whose goal is to reduce the number of humans a guest encounters. You've stayed in those hotels. You know how they feel.
We're arguing for the opposite. The reason to take message triage, report assembly, and status chasing off your team is so that when a guest walks up to the desk, the person standing there is not mid-task, not behind, not distracted — and knows, because the system told them, that this guest had a rough experience last time and that we're going to fix it.
That's the whole business. Everything else in a hotel is a commodity. The bed is a bed. The Wi-Fi is Wi-Fi. The only thing you own is the thirty seconds at the desk, and right now you're spending that person's attention on typing.
Get it back. Give it to the guest.