Hotels run on a property management system (PMS) that holds rooms, guests and bills, a channel manager that syncs rates and availability to booking sites, a booking engine for direct bookings, and payment tools; busier hotels add revenue software. AI earns its place in rate recommendations and guest-message drafts, not in anything that must be exact.
Many small properties buy one product that bundles the PMS, channel manager and booking engine, so the separate systems matter mainly when you outgrow it or when something breaks. The failures that hurt most tend to sit in the links between the parts, and a broken link is what causes overbookings and mismatched rates. Knowing what each part owns also tells you where AI can help safely.
One booking through five systems, with numbers
Picture a 12-room hotel with six Doubles at $140 a night. Every time and figure below is illustrative, but the order of events is how connected systems behave.
- Tuesday 10:02. A guest books a Double for Friday and Saturday nights on an online travel agency (OTA). The total is $280, paid to the OTA in advance.
- 10:02. The channel manager picks up the booking and passes it to the PMS.
- 10:03. The PMS creates the reservation and marks the last Double for those nights as sold. The new availability, zero Doubles, travels back through the channel manager to every other booking site and to the booking engine.
- 10:09. Another guest opens the hotel's own website. The booking engine reads the PMS calendar, shows no Doubles for those dates and offers a Twin at $150.
- Wednesday. The revenue tool sees 11 of 12 rooms sold for Friday and suggests raising the remaining Twin. The manager accepts or edits the suggestion.
- Friday 15:30. At check-in the PMS offers an inspected room. The front desk takes a card, and the payment tool places a $100 hold for incidentals and stores a token instead of the card number.
- Saturday 08:00. Breakfast for two is charged to the room from the restaurant till. $24 lands on the guest's folio, the running bill.
- Saturday night. The night audit posts the day's charges and balances the totals.
- Sunday 10:00. At checkout the folio shows $280 prepaid through the OTA and $24 of breakfast. The terminal captures $24 of the $100 hold and releases the other $76.
- Weeks later. The OTA pays out the $280 less its commission, and someone matches that payout to the reservation.
Three things stand out. The PMS is the source of truth: every other system reads or writes its calendar. The channel manager and booking engine are doors, not rooms; they sell whatever the PMS says exists. And payment status travels with the booking, because a reservation prepaid to the OTA and one that expects a card at the desk look identical on the calendar and very different at checkout.
Bookings: channel manager, booking engine and one shared calendar
The PMS stores every reservation and counts availability by room type. The channel manager works in both directions: it pushes your rates and availability out to booking sites such as Booking.com, Expedia and Airbnb, and pulls their reservations back in. It owns no calendar of its own. Some PMS products include a channel manager; others connect to a separate one, such as SiteMinder's, which syncs with the PMS you already run.
The booking engine sells on your own website. It reads live availability from the PMS, takes the guest's details and payment or deposit, and writes the reservation back. Cloudbeds describes its booking engine as commission-free and Sirvoy calls its own "100% commission free", which is the main reason owners push direct bookings: the sale carries no booking-site commission. Whether a vendor adds a booking fee instead is worth checking. Little Hotelier's pricing page, when checked on 29 October 2026, listed Basics from $39 a month plus a 1% booking fee and Pro from $179, so ask what any percentage applies to.
The budget alternative is calendar-link syncing, where each booking site exports an iCal calendar that another system reads. It is a scheduled pull, not a live two-way connection, so it can lag. Ask the vendor how often it refreshes before trusting it with the last room of a busy weekend.
A common failure is a mapping slip rather than an outage. Suppose the booking site's "Double" listing points at the PMS Family room, which should sell at $190. Every night sold goes out at $140, a $50 loss. Ten nights in a month costs $500 and nobody sees an error, because the sale technically worked. Check mappings when a connection is first set up, whenever you rename or add a room type, and after any change of login. The costs and trade-offs of buying a standalone one are in Channel Manager for a Small Hotel: Cost and What AI Adds.
Rooms: what the PMS does between a booking and a checkout
A booking is for a room type; a stay needs an actual room. The PMS assigns one on or before arrival, keeps a list of arrivals and departures, and tracks each room's state through dirty, clean and inspected. Housekeeping works from that list, and the front desk can only give out inspected rooms. Guest profiles and notes sit here too, along with the folio that collects every charge during the stay.
Out-of-order blocks matter most. If Room 7 has a leaking tap for two nights and maintenance logs the block in the PMS, availability drops from 12 to 11, the channel manager closes that room-night everywhere, and nobody sells it. If the fault lives only in a staff chat group, the room gets sold and the front desk finds out at 15:00 on Friday. Put every unavailable room in the PMS, whatever else you use to chase the repair.
The night audit is the end-of-day routine that posts room charges to in-house guests, checks the day's totals against payments and moves the system to the next date. Some systems run it automatically and some need a person to start it. A written routine for it is in Hotel SOPs Written With AI: Check-In, Housekeeping, Night Audit. Larger hotels also connect door-lock systems, so keys are encoded at check-in, and the restaurant till, so meals post to the room. Mews lists housekeeping, point-of-sale, accounting and an open API on its platform page, which is how those connections are normally made.
Payments: deposits, holds, tokens and the payout that arrives later
Money reaches a hotel in three patterns. The guest pays the booking site, which pays you later. The guest pays you online when booking, as a deposit or in full, through the booking engine. Or the guest pays at the property, usually with a card taken as a guarantee. A single hotel often has all three at once, and the PMS records which applies to each reservation, because it decides what the front desk charges.
The vendors name the same pieces. Mews lists embedded payments, tokenisation, terminals and automated reconciliation. Cloudbeds lists secure payment processing, and SiteMinder lists hotel payments. In plain terms: a payment gateway takes the card, tokenisation stores a stand-in code instead of the real number, a terminal handles the card at the desk, and reconciliation matches what arrived in the bank to what the PMS says should have arrived. None of the pages I read published prices or processing rates, so ask for them in writing.
The sample folio above shows how a hold works. A $100 pre-authorisation reserves funds without taking them. At checkout only the $24 owed is captured and the remaining $76 is released. Confusing a hold with a charge, or the reverse, is a classic source of card arguments at a front desk.
Reconciliation is where a general assistant can help. Say the PMS shows eight stays from one booking site totalling $2,380 gross, so at an assumed 15% commission you would expect a payout of $2,023. The bank shows $1,904, which is $119 short. That gap equals a $140 stay less its 15% commission, and a check of the rows finds a guest who cancelled through the booking site while the PMS still lists the stay as active. Paste the statement rows and the PMS export without guest names, ask the assistant to match them and list differences, then confirm the arithmetic in a spreadsheet yourself.
Revenue tools: when a pricing engine earns its fee
A revenue management system (RMS) forecasts demand and suggests rates. The platform pages show the labels: Mews lists dynamic pricing and demand forecasting, SiteMinder lists market intelligence and demand forecasting, and Cloudbeds has an RMS plus links to third-party ones. Cloudbeds' page gives three ways to treat a suggestion: accept it, edit it, or let Autopilot apply it inside limits you set. That choice is the one that matters, so treat it as a rule: recommend-only first.
A review might go like this. Three weeks before a Saturday, 9 of 12 rooms are sold. The tool suggests raising the Twin from $150 to $165. Your own limits are a floor of $120 and a ceiling of $220. Last year the same Saturday sold out at $150 two days ahead, which supports a rise. The step is 10%, small enough to accept for Twins; leave Doubles alone until you see how the Twins sell. After the date you compare rooms sold and final average rate with the same weekday last year and last month. Four to six weeks of this tells you whether the tool earns its fee before autopilot touches a rate.
Many small hotels do not need one yet. If demand is steady, a weekly look at your own pace beats a subscription; AI for Hotel Revenue Management: Weekly Reports for Small Hotels shows that routine.
Three property sizes, three stacks
These are illustrations of how the parts combine as a property grows, not rules.
| Property | Usually enough | Add when | First AI job |
|---|---|---|---|
| Six-room guesthouse selling direct and on one booking site | One all-in-one PMS with a booking engine; the booking engine's own payment gateway | A second booking site makes manual calendar updates risky | A general assistant for replies, room descriptions and review responses |
| Fourteen-room inn on three booking sites and direct | PMS with a built-in or tightly connected channel manager and booking engine, integrated payments | Prices change weekly because demand swings by date | A pricing tool in recommend-only mode, plus message drafts |
| Forty-five-room hotel with a restaurant and meeting room | PMS with an open API; channel manager and booking engine integrated; till and accounting connected | Staff spend hours reporting across systems | Rate recommendations, request triage and guest messaging with handover |
Sirvoy's pricing page shows the small end: a free plan for one room with a free website and booking engine, with the channel manager on the top paid tier and prices rising with room count. Check any vendor's current page, because tiers and fees change. For a smaller property, whether you need a channel manager at all is usually the first question to settle.
Where AI fits: five jobs to try and three to avoid
Several PMS and channel-manager vendors now name AI products: Cloudbeds has Signals, Mews has Mews AI and SiteMinder has SiteMinder IQ. The public pages range from specific to promotional, so judge a feature by what it does on your data. Cloudbeds' AI page is the most specific of the three I read, and its Signals data layer is described as included, not bought or switched on separately. For the others, the pages I read did not say which plan carries which feature.
| Job | What AI does there today | Verdict and what to check |
|---|---|---|
| Rate recommendations | Forecasts demand and suggests rates; Cloudbeds offers accept, edit or autopilot inside limits | Useful where demand swings by date. Set a floor and ceiling and run recommend-only first |
| Replies to guests | Drafts or sends answers to routine questions in several languages | Useful with staff handover. Check what it can see, whether it sends without approval and when a person takes over |
| Guest request triage | Turns a text into housekeeping and maintenance tasks; Cloudbeds' page shows an example | Promising. Test on twenty real requests before trusting room numbers and urgency |
| Questions about your own data | Answers from live records; Cloudbeds listed Ask Signals as "coming soon" when I read it | Watch it. Ask for a demonstration on your own data |
| Writing | Room descriptions, review replies and pre-arrival emails from a general assistant | Useful. A person checks facts, amenities and policies |
| Availability and rate sync | Nothing: it must be exact | Keep it deterministic. Read the sync log and error alerts |
| Payments and refunds | At most, matching exports to payouts | Decisions stay with staff. Confirm totals in a spreadsheet |
| Photos | Edits that change rooms, views or condition | Avoid. Google and Expedia partner policies bar edits that misrepresent a property |
Request triage is the easiest of these to test. Take a guest text sent to the hotel's number: "Hi, the shower in 12 is only lukewarm and could we have two more towels? Also is breakfast still on till 10?" A triage tool should produce three outputs: a maintenance task for Room 12's hot water marked for the same day, a housekeeping task for two towels to Room 12, and a draft reply giving the breakfast time from your own written policy. Check three things on real requests. The sender must really be in-house in that room, or anyone could create tasks. The urgency must match what your staff would choose, because "lukewarm" is not "no hot water". And the reply must promise no repair time unless a person has set one. Run twenty past requests through it and count the mismatches before it goes live.
For apartment-style units, short-stay rental platforms such as Hostaway and Guesty add messaging automation. Hostaway lets each rule send a draft for approval or reply automatically, and its documentation recommends drafting first. Guesty offers AutoReply and task suggestions. If you use any of them, the draft-first setting is the one to start with. If guests message you on WhatsApp, Meta's own rules on bots and handover apply too, and WhatsApp for Small Hotels: Meta's AI Rules, Templates and Handover covers them.
AI on your own exports: a prompt, a sample answer and the checks
You do not need a vendor's AI to ask questions of your data. Export twelve months of reservations from the PMS, strip guest names, emails and phone numbers, and ask a general assistant.
This is a reservation export from a 12-room hotel. Columns: arrival date, nights, room type,
source, rate per night, status (stayed or cancelled). It contains no guest details.
Work out: occupancy by weekday, share of bookings by source, cancellation rate by source,
average rate by room type. Show the method for each figure, then list three questions the
numbers raise. Say if any month looks incomplete.An illustrative reply gives occupancy of 88% on Fridays, 94% on Saturdays, 61% on Sundays and 55% from Monday to Thursday; direct bookings at 31% of the total against 44% for the largest booking site; and cancellations of 9% direct against 22% on the second booking site. The questions it raises are good ones: why Sunday is so much weaker than Saturday, whether the high-cancellation site justifies its commission, and whether direct guests accept a small midweek discount.
Check it before you act. Recompute one figure by hand from ten rows. Confirm how cancelled bookings were counted, because an assistant can include them in occupancy. Make sure a partial month has not dragged an average down. And remember the output describes the file you gave it: if the export left out walk-ins or phone bookings, the answer does too.
Five terms that blur on vendor pages
- CRS (central reservation system). Holds rooms and rates for a group of hotels and distributes them. One independent hotel's PMS normally does this job for a single property, so a CRS is rarely a separate purchase at small scale.
- GDS (global distribution system). Connects hotels to the booking tools travel agents use. It matters mainly if you sell to corporate or agency travellers.
- Metasearch. Price-comparison listings that send the guest on to a booking site or to your own booking engine. The booking engine decides whether that click becomes a commission-free sale.
- "Hotel management system" and "hotel booking system". Marketing labels that can mean a PMS, a booking engine or both. Ask the vendor to name which of the pieces in this tutorial it actually includes.
- Guest messaging or hospitality CRM. Tools that hold guest history and send emails, texts or chat messages. They normally sit beside the PMS and read its data, which is why they are the usual home for AI replies.
Questions to ask before paying for a hotel system or its AI
- What does it sync, and how fast? Ask for rates, availability and reservations separately, and the maximum delay for each.
- What happens when a connection fails? You want an alert, a log and a named person, not a silent gap.
- Which side is the source of truth? If two systems can both edit availability, they will eventually disagree.
- Is there a commission or booking fee, and on which bookings? One small-property product lists a 1% fee on top of its monthly price; another advertises none on its booking engine.
- Who processes payments, and at what rate? Get the rate in writing, along with where payouts land and how fast.
- Are AI features included, or an add-on? Cloudbeds states Signals is included; many pages say nothing.
- Can the AI send without approval? Insist on a draft-first mode you control.
- What data does it use, and does it train anyone's model? Hotels hold passport details, card tokens and contact data.
- How do I get my data out? Ask for a full export format and whether it is free. Hotel Management Software for Small Hotels: A Pre-Switch Checklist goes further on switching.
Three places to keep AI out of the stack
First, inventory. A chatbot that quotes room availability or rates it cannot verify will eventually promise a room you do not have. Let any assistant answer policy questions, and send availability and rate questions to the booking engine, which reads the PMS.
Second, money decisions. Refunds, cancellation waivers and compensation involve judgement and sometimes rules in the booking contract. An assistant can draft the explanation after a person decides.
Third, photos. Google's hotel listing policy and Expedia's partner content policy both bar AI editing that distorts rooms, views or condition, and Expedia asks partners to disclose ongoing construction. Straightening converging vertical lines is one thing; adding a sea view is another. Is It OK to Edit Hotel Photos With AI? Fair Edits vs Misleading draws the line with examples.
Hotel software questions owners ask before they buy
Do I need a separate channel manager if my PMS has one built in?
No. Many PMS products include their own, and a bundled one removes a connection that can fail. Check which booking sites it reaches, who maintains each connection, what alerts you get when one drops, and whether the price changes when you add channels. A separate channel manager only earns its place if your PMS lacks the channels you sell on.
How can I tell if my systems have drifted out of sync?
Once a week, pick three room types and compare the next fourteen days in your PMS calendar with each booking site's extranet. Look at availability and rate, not just bookings. Also read the sync log or error alerts after any change to room types, rate plans or logins, because that is when mappings break.
Is revenue management software worth it for a small hotel?
It depends on how much your demand swings by date. If weekends and events sell out while midweek stays empty, a tool that forecasts and suggests rates can pay for itself. If demand is flat, a weekly manual review is enough. Start in recommend-only mode and compare its suggestions with your own before you let anything apply changes.
Can ChatGPT replace hotel software?
No. A general assistant cannot hold live availability, take payments or sync rates, and a wrong answer there means an overbooking or a lost room-night. It is useful on the edges: analysing an export with guest names removed, drafting replies and room descriptions, and checking figures. Keep it away from anything that must be exact.
What should I ask about AI features before paying?
Ask whether the feature is included in your plan or sold as an add-on, what data it can see, whether it sends to guests without staff approval, how to switch it off, and whether guest data trains anyone's model. Then ask for a live demonstration of the exact task on your own sample data, not a slide.
Further reads
- Channel Manager for a Bed and Breakfast: Do You Need One? — A smaller property's version of the channel-manager question.
- Is AI Dynamic Pricing Worth It for a Small Hotel? — Whether AI pricing is worth the fee at small scale.
- Hotel AI Guest Messaging Tools Compared for Independent Properties — Guest messaging tools compared for independent properties.
- AI Agents for Hotels: What They Can Do Today and Where They Fail — What hotel AI agents can do today and where they fail.
- Where Should a Small Hotel Start With AI? — The first AI jobs a small hotel should pick.
- Hotel Rate Shopping: A Weekly Competitor Rate Sheet Built With AI — A weekly competitor rate sheet to judge pricing suggestions.
- Accommodation Booking Software for Hostels: Systems Compared — Booking software for hostels compared on bed-level inventory, channel sync, payments and AI messaging, with a bed-counting worked example.
- Free Hostel Management Software: What You Give Up — Free hostel software works for one room and little else. Here is what the free plans leave out and the signs it is time to pay.
- Hotel Housekeeping Schedules: Allocate Rooms by Arrival Time With AI — A 32-room hotel's full-house Tuesday planned with AI: arrival priorities, minutes per room type, checked sheets for each attendant and a mid-day re-plan.
- Hotel Maintenance Software or AI? Log, Prioritise and Chase Requests — Decide between maintenance software and AI with a two-week test: one request form, urgency rules, a weekly chase list and a room-out-of-order tracker.
- Smart Room Automation Through AI in Hotels: What a Small Property Can Add — What a small hotel can add to its rooms first: thermostat setback linked to the booking system, window and door sensors, and clear guest consent.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Cloudbeds, Mews, SiteMinder and Sirvoy product and pricing pages (October 2026); Little Hotelier pricing page as checked on 29 October 2026; Cloudbeds AI page; Hostaway and Guesty help pages on automated replies; Google and Expedia partner content policies on edited images.