Industries
AI for hospitality and travel
Guest messaging, reservations support and review replies grounded in your property data and PMS. Answers come from approved sources, and nothing changes a booking or a rate without confirmation.
The sector
What's different here
Hotels answer the same questions all day in several languages: check-in times, parking, breakfast, whether the room can be ready early. Most answers live in the PMS reservation and a property fact sheet, which makes guest messaging a natural first job for a model.
What the bot says is the company's responsibility. In Moffatt v. Air Canada a tribunal held the airline liable for its chatbot's wrong answer on bereavement fares and rejected the argument that the bot was a separate entity. Answers have to be grounded in approved content, and anything the content doesn't support goes to staff.
Integration is usually the hard part. Many properties run on-premise PMS versions with limited APIs, and franchise groups depend on each property keeping its own data current. We plan for both before writing any prompts.
46%
of hotel owners name data privacy and security concerns as a top barrier to AI, ahead of cost (42%) and difficulty integrating with legacy systems (40%). From Wyndham's owner trends report, via press coverage.[1]
41%
of hoteliers in a Mews survey of 500+ properties have no formal AI policy, while 98% said they use AI somewhere in operations. A vendor survey, and 'use' is broadly defined.[2]
Workflows
Where AI does useful work
Each of these runs inside systems the sector already uses. Steps that carry real risk wait for a person.
Guest messaging and pre-arrival
Messages from web chat, WhatsApp and SMS are answered from the reservation and the property knowledge base.
- 1A message arrives on web chat, WhatsApp or SMS
- 2Detect the intent and the language
- 3Read the guest's reservation from the PMS
- 4Answer from the property knowledge base: hours, policies, amenities
- 5Offer early check-in or an upgrade through the upsell engine
- 6Hand off to the front desk with the conversation attached
Front desk staff handle complaints, special requests and anything the knowledge base doesn't cover.
PMS (Oracle OPERA Cloud, Mews) · Guest messaging (Canary, HiJiffy, Duve) · WhatsApp Business API · Upsell engine
Reservations support
Availability and rate questions are answered from the booking engine, with payment kept outside the model.
- 1Check availability and rates in the CRS or booking engine
- 2Quote the rate with its cancellation terms
- 3Collect the booking or modification details
- 4Hand payment to a tokenised payment page
- 5Confirm only after the CRS returns a booking reference
Reservations staff confirm modifications and group or non-standard bookings until write accuracy is proven.
CRS (Sabre SynXis) · Booking engine · Payment gateway with tokenisation · Channel manager (SiteMinder)
Review responses
Reviews are collected, replies are drafted in the brand's voice, and operational issues go to the team that can fix them.
- 1Aggregate reviews from the main platforms
- 2Score sentiment and tag topics such as cleanliness or noise
- 3Draft a reply in the brand voice
- 4Route maintenance and housekeeping issues to those teams
- 5A manager edits and posts the reply
The general manager or guest relations lead approves every public reply.
Review platforms · Reputation management tool · Housekeeping and maintenance system · CRM and loyalty
Revenue management explanations
Pickup and pace are summarised and RMS recommendations explained, so the revenue manager decides faster.
- 1Pull pickup, pace and the RMS recommendations
- 2Summarise changes since the last review
- 3Explain each rate recommendation in plain language
- 4The revenue manager accepts, changes or rejects rates in the RMS
The revenue manager approves every rate change. The model never writes to the RMS.
RMS (IDeaS, Duetto) · PMS · Channel manager (D-EDGE) · Data warehouse
Back-office documents
F&B supplier invoices, group and MICE RFPs and staff SOP questions, handled as drafts for review.
- 1Extract F&B supplier invoices and match them to orders
- 2Draft responses to group and MICE RFPs from approved content
- 3Answer staff questions from the SOP library
- 4Route exceptions to finance or sales
Finance approves invoice exceptions, and the sales manager approves every RFP response.
Accounting or ERP · Sales and catering system · SOP library · Email
Systems and rules
What the design has to respect
The systems we expect to connect to, and the regulation that decides what an AI system may do with the data in them.
- PMS (Oracle OPERA Cloud, Mews, Cloudbeds, Protel)
- CRS (Sabre SynXis, Amadeus)
- Channel managers (SiteMinder, D-EDGE)
- GDS (Amadeus, Sabre, Travelport)
- RMS (IDeaS, Duetto)
- Guest messaging (Canary, HiJiffy, Duve)
- CRM and loyalty
- POS
- PSS and departure control (airlines)
- Moffatt v. Air Canada, 2024 BCCRT 149
- The tribunal held the airline liable for negligent misrepresentation by its chatbot and rejected the idea that the bot was a separate entity. We treat every bot answer as a company statement.
- PCI DSS
- Applies to card-not-present bookings. Card data goes to a tokenised payment page and never into the model's context.
- GDPR and local privacy law
- Guest profiles and passport data need a lawful basis, retention limits and care over which model provider processes them.
- EU AI Act, Art. 50
- Guests must be told they are talking to an AI. Under the Digital Omnibus schedule these transparency obligations apply from 2 Aug 2026.
- Travel and consumer law
- Package travel rules, fare transparency and refund rules, such as the US DOT airline refund rules, decide what an assistant may promise.
Risks
What goes wrong, and how we design for it
- Wrong policy, fare or refund information
- Answers come only from approved, versioned content with the source logged. If the content doesn't cover the question, the assistant hands off.
- Overbooking or rate errors from write actions
- The assistant starts read-only. Write actions come later, behind a confirmation step and a check against the CRS response.
- Uneven quality across languages
- Each language you serve gets its own evaluation set, reviewed by native-speaking staff before it goes live.
- Fragile legacy PMS integrations
- We test against your PMS version early, cache reservation reads and fall back to staff hand-off when the PMS is unavailable.
- Inconsistent data across franchised properties
- Each property's fact sheet has an owner and a review date, and answers citing content past its review date are flagged.
Where to start
Guest messaging with read-only reservation lookup
Value is measurable in share of messages handled, response time and upsell revenue. Staying read-only and out of payment flows limits what can go wrong while accuracy is measured, and the Air Canada ruling argues for answers that are strictly grounded and cited.
- Build a property knowledge base with an owner and review date for each fact sheet
- Connect read-only reservation lookup from the PMS (Oracle OPERA Cloud or Mews)
- Answer on web chat and WhatsApp, with AI disclosure and a clear route to the front desk
- Keep payments and booking changes out of scope until accuracy is proven
- Measure share of messages handled, response time, hand-off rate and upsell revenue
- Review a weekly sample of answers with the front office manager
In the sector
Publicly reported examples
What others in the industry have announced. These are not our clients.
Hilton AI Planner[3]
Hilton launched the Hilton AI Planner in beta on hilton.com, a conversational tool for choosing destinations, comparing properties and exploring amenities. It went to a small share of traffic first.
Air Canada chatbot ruling[4]
A tribunal held Air Canada liable after its chatbot wrongly said bereavement fares could be claimed retroactively. A public cautionary case on liability for what a bot tells customers.
Use cases
Related use cases
- Customer support automationAn agent answers routine questions from your own policies and account data, takes a small set of allowed actions, and hands everything else to a person with the case already summarised.
- Company knowledge searchStaff ask a question in plain language and get an answer drawn from your own documents, with citations, limited to what each person is already allowed to see.
- Back-office workflowsReconciliations, approvals and onboarding are prepared by agents with the evidence attached, then routed to the right approver. Nothing is posted or paid without a person signing off.
Sources
- 1.Journal des Palaces, Hotel owners at an AI crossroads: Wyndham owner trends report (2026)
- 2.Hotel Business, Mews survey finds 98% of hoteliers use AI (2026)
- 3.Hotel Dive, Hilton launches generative AI agent (Mar 2026)
- 4.Civil Resolution Tribunal, Moffatt v. Air Canada, 2024 BCCRT 149 (Feb 2024)
Figures are as reported by each source on its publication date. Examples are public industry cases, not AI Sewa clients.
Next step