Where Should a Small Hotel Start With AI?

Start with the guest inbox. For most small hotels the first AI job worth doing is drafting replies to pre-arrival questions and reviews, using the AI already inside your email, property management system or booking-site extranet, with a person approving each reply. Leave dynamic pricing, voice bots and booking chatbots until that first job is running smoothly.

Messaging wins on plain arithmetic: it is the job that repeats most often, costs little if a draft is wrong (someone reads it before it goes), and needs only information you already have. The answer changes if your hours go somewhere else: a hotel where reception spends half the shift on the phone, or where rates haven't been touched in months, may get more from starting there. So before choosing, spend one week writing down where the time actually goes.

Follow me on Instagram@sagnikteaches

Score your hotel's candidate jobs before buying anything

Vendors will each tell you their product is the place to start. A simple scoring grid, filled in with your own numbers, is more honest. Score every candidate job from 1 to 3 on four things:

Connect on LinkedInSagnik Bhattacharya

• Volume: how often it happens. 3 = many times a day, 1 = weekly or less.
• Time each: 3 = ten minutes or more, 1 = under two minutes.
• Safety if wrong: 3 = a person checks it before anyone sees it, 1 = a mistake reaches a guest or costs money directly.
• Data ready: 3 = the information already exists in one place, 1 = you'd have to build it first.

Here is how the grid might look for an illustrative 18-room hotel with an owner, two receptionists and seasonal housekeeping. Your scores will differ; the method is what matters.

Subscribe on YouTube@codingliquids

Two jobs usually float to the top for properties this size: guest message replies and review replies. Both are high-volume, both let a person approve before sending, and both run on information you already hold. Pricing and booking chatbots score low not because they're useless but because a mistake costs real money and they need clean data first.

Review replies earn their 11 because the drafting is slow and the checking is quick. Give the assistant the review and a few facts, and ask for a reply under 80 words that thanks the guest, answers each specific point and promises nothing you haven't done. A three-star review reads: "Lovely room and friendly staff, but breakfast was lukewarm and nobody told us the car park locks at 11pm, so we had to park on the street." An illustrative first draft:

Thank you so much for your kind words about our room and team! We're
very sorry breakfast wasn't up to standard. We've spoken to our chef
and this will never happen again. We apologise for any inconvenience
with parking and hope to welcome you back soon!

Three fixes. "Will never happen again" is a promise nobody can keep; say what actually changed (plates now go out from a warmed hot cupboard, say, if that's true). "Any inconvenience with parking" dodges the real point, which is that you didn't tell them; the useful line is "we now put the car park closing time in every booking confirmation", and then you do it. And three exclamation marks in four sentences reads as a template. The edited reply is shorter and more believable, which is what the next guest reading it is judging.

The starting point shifts with the size of the property

Tools for the first project, most of which you already have

Before buying a guest-messaging platform, check what sits unused in your existing systems.

• Booking-site messaging. Booking.com's extranet and Pulse app let you set up message templates for each stage of a stay and automatic replies to common guest requests; Booking.com lists parking, check-in and check-out, and bed types as the request types automatic replies currently cover. Airbnb offers scheduled quick replies that send on triggers such as a new booking, check-in or checkout, which you can skip or edit per reservation.
• Your email suite. Google Workspace business plans now include Gemini in Gmail. Microsoft 365 business plans include Copilot Chat, which now works on the Outlook email or file you have open; the paid Microsoft 365 Copilot Business add-on, about $21 per user a month on annual billing, adds reasoning across your emails, meetings and files together. Either route can help draft replies from a guest's thread.
• A chat assistant with a fact sheet. A single ChatGPT Plus or Claude Pro seat at about $20 a month, with a document describing your property pasted into a Project, drafts consistent replies to anything unusual.
• Your property management system. Many now include guest messaging, and some have added AI drafting. Ask your provider what is included in your current plan before looking elsewhere.

If you outgrow these, the dedicated platforms are compared in hotel AI guest messaging tools for independent properties. For the review side, replying to hotel reviews with AI without sounding canned has prompts and examples.

Leave these until later, and why

Dynamic pricing

It can earn more than any messaging tool, but only when you have enough booking history, competitors to watch, and someone who understands why the system moved a rate. A bad pricing month costs real revenue and is hard to spot until it's over. Whether it's worth it at your size is covered in whether AI dynamic pricing is worth it for a small hotel.

A chatbot that takes bookings

Taking a booking means live availability, rates, payment and cancellation terms all connected correctly. A chatbot that only answers questions is much simpler. The difference is set out in whether a hotel chatbot can take bookings or only answer questions.

Voice AI on the phone

Phone calls to small hotels are often the complicated ones: a wedding party, an accessibility question, a complaint. Automate the written channels first and see whether call volume falls before paying for a voice agent.

What the first project costs in money and hours

For the inbox project, cash cost is usually small and time cost is where the budget goes. A realistic picture for a hotel of 10 to 30 rooms:

• Software: often nothing extra, because booking-site templates and the AI in your email suite are already included. If you add a chat assistant for drafting, one or two seats at about $20 a month each. A business plan such as ChatGPT Business or Claude Team costs a little more per seat but keeps guest conversations out of model training by default and requires at least two seats.
• Owner or manager time: four to six hours to write the property fact sheet properly the first time. This is the part people rush, and it is the part that decides whether drafts are usable.
• Reception time: an hour of practice per person with the prompt, then a few minutes a day checking drafts.
• Ongoing upkeep: 15 minutes a month updating the fact sheet when anything changes.

Compare that with a dedicated guest-messaging platform, which typically charges a monthly subscription per property or per room plus setup. That can be worth it later, once you know from your own numbers how much time messaging takes and how much of it the AI handles.

Keeping guest data safe while you test

Hotels hold more personal data than most small businesses: names, addresses, phone numbers, sometimes passport details and dietary or medical notes. While testing:

• Never paste card numbers, passport or ID details into any chat tool, business plan or not.
• Strip the guest's surname and contact details from a message before pasting it into a consumer plan; the first name and the question are enough for a good draft.
• If staff use personal accounts for work, have them switch off the model-training setting in privacy settings, or move to a business plan where training on your content is off by default.
• Keep the fact sheet free of anything private: it should hold only what you'd happily tell any guest who asked.

If you process data on guests from the EU, check with your data-protection adviser before connecting any AI tool directly to your property management system.

A 30-day plan for the illustrative 18-room hotel

Week 1: count. Reception tallies every guest message for a week: channel, topic, and roughly how long the reply took. A sheet by the desk with four columns is enough; the first few rows of an illustrative tally:

Don't try to be exact with the minutes; the point is to see which topics repeat and which ones eat time. Say the tally shows 140 messages, of which about 100 are routine (parking, check-in times, dogs, dinner reservations, directions) and take around three minutes each. That is five hours a week on routine replies.

Week 2: write the fact sheet. One document with everything a guest asks: times, parking with the height of the car park barrier, dog policy with any fee, breakfast hours, the nearest cash machine, what happens if they arrive after reception closes. Switch on the booking-site templates and automatic replies that match it.

Week 3: draft, don't send. For every routine message, reception asks the AI for a draft using the fact sheet, edits if needed, and sends. Use a prompt like this:

You draft guest replies for [hotel name], an 18-room hotel.
Use ONLY the facts in the attached property sheet. If the answer
isn't in the sheet, say "I'll check and come back to you" instead
of guessing. Warm, brief, no exclamation marks. Sign off as
"[first name], Reception".

Guest message:
[paste]

Try it on a real-looking message. The guest writes: "Hi, we're arriving around 9.30pm Friday with our spaniel, is that ok? Also do you do evening meals?" An illustrative draft from the assistant:

Hi [guest name], thanks for letting us know. Reception is open until 10pm,
so 9.30 is fine; if you're delayed, call [number] and we'll leave
your key in the safe by the front door. Dogs are welcome in rooms
3 and 4 for a $15 per night charge, and you're booked into room 3.
Our restaurant serves dinner until 8.30pm, so on Friday it will
have closed by the time you arrive. I'll check and come back to you
about anything nearby that's open late.
[first name], Reception

Mostly good: it used only the fact sheet, and it said "I'll check" instead of inventing a late-night restaurant. Two things to fix before sending. The assistant confirmed room 3 from the booking the receptionist pasted, so check that it really is a dog room today. And the reply ends on a promise ("I'll check and come back to you"), which means someone must actually do it; either answer it now or add a line on late food to the fact sheet so next time the draft can.

One edge case turns up in the first week at most hotels: a message in a language nobody on reception reads. The assistant will happily draft a reply in that language, and it will probably be fine, but nobody can check it. The safer rule is to have it draft in English first, approve that, then ask for a translation and send both, the guest's language on top and the English below. If the English version is right, a small slip in the translation is forgivable; an unchecked promise in a language you can't read is not.

Week 4: measure. If each routine reply now takes about a minute to check and send, that is roughly 1.7 hours instead of five, a saving of around three hours a week. Also count how many drafts needed real correction; if more than one in five did, the fact sheet has gaps.

The mistake that shows up at checkout

The most common failure in the first month is a stale fact sheet, and the AI simply repeats it. A typical sequence: the owner raises the dog charge from $10 to $15 a night in March, updates the booking sites, and forgets the fact sheet. For three weeks every AI draft about dogs quotes $10, reception approves them because they look right, and the problem appears at checkout when a guest points to the message saying $10. The hotel honours the lower price, and it happens four more times before someone traces it.

The fix is dull but effective: a "last updated" date at the top of the fact sheet, one named person responsible for it, and a rule that any change to a price, time or policy is made in the fact sheet first, then everywhere else. Once a month, ask the AI to compare the fact sheet with your booking-site listing text and list any differences. It is good at spotting "check-in from 3pm" in one place and "check-in from 2pm" in another. A typical result (illustrative) looks like this:

Differences between the fact sheet and the listing text:
1. Check-in: sheet says from 3pm; Booking.com listing says from 2pm.
2. Dog charge: sheet says $15 per night; Airbnb listing says $10.
3. Breakfast: sheet says 7.30-10am; listing says 7-10am.
4. Parking: listing mentions "free parking"; sheet says $8 per night.

Treat the list as leads, not verdicts. In this example, number 3 may not be an error at all if breakfast starts at 7 on weekdays and 7.30 at weekends and only one of the two documents says so; the fix is to make both say the same full thing. Number 4 is the one to correct today, because "free parking" on a listing is a promise a guest will quote at checkout.

A six-room inn with a restaurant scores it differently

At a six-room inn where the owners also run a busy restaurant, the grid usually comes out differently. Room messages are few, perhaps 30 a week, and the owners answer them between services. The job that scores highest is often table enquiries and group bookings for the restaurant, because they arrive by email and social messages at all hours and each needs a menu, a deposit rule and a date check. Scored on the same four measures, room questions might come out at 1, 2, 3 and 3 (9 out of 12), while restaurant group enquiries come out at 3, 3, 3 and 2 (11), losing a point on data only because the set menus and deposit terms haven't been written down in one place yet. The same first-project logic applies (drafts from a fact sheet, approved by a person), but the fact sheet is about the dining room: covers, set menus for groups, deposit terms and dietary handling. That is why the scoring grid comes before the tool: the right starting point depends on where the hours go in your building, not on what a vendor sells to hotels.

Signs the first project is working, and signs to stop

It is working if reply times fall, reception says the drafts are usable as they are most of the time, and guests stop asking the same question twice in a thread. It is not working if staff rewrite most drafts, if a wrong answer reached a guest (a fee that changed, a closed restaurant recommended), or if nobody opens the tool after the first fortnight. In the last case, the problem is usually the fact sheet or the approval step being too slow, not the AI.

Once the inbox is under control, the natural next steps are automated pre-arrival and post-stay emails with an upsell line, then round-the-clock answers for the simplest questions. Each one builds on the fact sheet you wrote in week 2, which is why it is the most valuable document in the whole exercise. Keep it dated, and give one person the job of updating it whenever a price, time or policy changes.

Further reads

• How Small Hotels Use AI to Answer Guest Messages Around the Clock — The next step: answering guest messages around the clock.
• How to Upsell Room Upgrades and Late Checkout With AI — A good second project once messaging runs smoothly.
• How Independent Hotels Win More Direct Bookings With AI — Where AI fits in winning bookings away from the OTAs.
• How to Train Front-of-House Staff to Work Alongside AI — Bringing reception staff along with the change.
• How to Run Your First AI Pilot Project in a Small Business — How to run the first project as a proper pilot.
• Customer-Facing or Back-Office: Where Should AI Go First? — The general case for starting behind the scenes.
• How to Write B&B Room Descriptions and Guest Guides With AI — Room fact cards, prompts for each listing, the words that lead to refund disputes, and a quick way to turn a walk round the house into a guest guide.
• How B&B Owners Can Automate Airbnb and Booking.com Messages — A message-by-message plan for automating B&B guest messages on Airbnb and Booking.com with built-in tools, AI-written templates and a channel manager.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Booking.com for Partners help pages on templates and automatic replies; Airbnb help articles on scheduled quick replies; Microsoft 365 and Google Workspace pricing pages.