White paper · Travel and hospitality
AI agents inside
the airline and the hotel
What they can read, what they can change, and who decides
In brief
- Airlines, airports and hotels have put AI assistants to work, and most began with the customer. One airline says its assistant has handled “nearly 4 million customer queries, 97% of them with full automation”. One hotel software company says its assistant “answers 80% of routine questions instantly”.
- The record is mixed. A tribunal ordered an airline to pay a passenger after its website chatbot misinformed him about a fare. Another airline had to tell senators that its pricing tool was “a decision-support tool”. The suppliers are moving in different directions: one says its product “can recommend an action for human approval or execute it automatically”, and a published hotel template sends compensation offers to guests. None of the vendor pages we read offers on-premises or customer-cloud hosting.
- This paper is about the company’s own work: the reading, checking and drafting done by staff in guest service, reservations, claims, sales, revenue, operations, loyalty and finance. It is not about anything that answers a guest unseen, sets a price or directs staff. It sets out what an agent can read, how it reaches a travel company’s systems, what it may change without a person, and what the rules already say about a company’s own words, passenger rights, prices, reviews and workers. It ends with seven questions to put to any supplier, and with a place to start.
- The position throughout is simple. An agent prepares. A named person decides. Nothing reaches a guest, a passenger or a partner unseen, and no agent changes a booking, issues a refund or sets a price. The agent runs on infrastructure the company controls.
Where airlines, hotels and travel companies are today
Airlines and airports began with the customer. A Microsoft account of Air India’s assistant says: “To date, AI.g has successfully handled nearly 4 million customer queries, 97% of them with full automation.” When it cannot help, “it automatically escalates to contact center staff” [1]. Salesforce said in 2025: “Heathrow will soon be deploying Agentforce” [2].
Hotels began there too. Mews said of its guest messaging: “Mews AI, trained on the property’s own data, answers 80% of routine questions instantly” [3].
Two warnings, both from airlines. In February 2024 a tribunal in British Columbia “ordered Air Canada to pay Jake Moffatt $812” after the airline’s website chatbot gave him wrong information about bereavement fares [5]. The tribunal’s reasoning, as a lawyer’s note reproduces it: “It makes no difference whether the information comes from a static page or a chatbot” [4]. In 2025 three United States senators questioned Delta Air Lines about AI in its pricing. Delta replied: “There is no fare product Delta has ever used, is testing or plans to use that targets customers with individualized prices based on personal data” [6]. No agent in this paper answers a passenger unseen or sets a fare.
The suppliers disagree on how far an agent acts. Accelya says its disruption product “can recommend an action for human approval or execute it automatically”, depending on policies the airline sets [7]. Amadeus says of its own tests: “Trials show an airline AI agent can identify booking, understand verbal change request, propose new options, articulate the fare differential and initiate payment” [8].
Where they run is rarely said. Accelya’s product is “Built on Amazon Web Services (AWS) using Amazon Bedrock AgentCore” [7]. Cloudbeds says it “runs on Amazon Web Services (AWS)” [9]. None of the vendor pages we read offers on-premises or customer-cloud hosting. So the buyer has to ask where it runs, which tools it may call, and with whose approval.
Two kinds of work in one company. The published examples answer the guest and the passenger, and both warnings come from there. Behind them sits the work this paper is about: the claim checked, the contract summarised, the handover note written. A travel company holds both kinds, and needs one set of rules for both.
What the published accounts leave out. Of the supplier pages we read, one says the company can require a person’s approval before an action. None describes who approves a change to a booking, a refund or a rate.
The work nobody has packaged
We listed the jobs done around a journey and a stay and looked for each in the template libraries and prebuilt agents of the agent and workflow platforms. They cover guest questions, complaints and rebooking. They thin out wherever a named person checks, approves or pays.
Where templates or prebuilt agents exist
- A guest’s enquiry classified and answered by email
- Guest feedback scored, with a compensation offer sent
- Routine guest questions answered from the property’s own data
- A passenger’s questions answered on the airline’s own channels
- Wayfinding and airport information for passengers
- Disrupted passengers rebooked and told
- A flight change taken by voice, with payment started
- A ticket raised in the property system from a complaint
Where none was found
- A compensation claim checked against the passenger rights rules
- A refund request checked against the fare rules before the refunds team decides
- A disruption notice drafted with the rights it must state
- A group booking or a rooming list checked against the contract
- A partner contract summarised with each rate, allotment and release date
- A commission statement checked against the bookings
- A night audit’s exceptions listed for the auditor
- A ground handler’s licences and training records checked against the requirements
Two findings matter more than the list. First, both hotel templates we opened send to guests. One “Sends recovery offers to unhappy guests”, with compensation the model proposes, and lists an “Optional manager approval step for high-value recovery offers”. The other says it “Automatically replies if the guest prefers email contact”, and posts a summary to the team afterwards [10].
Second, both are built from hosted services: a hosted form, a hosted model, a hosted mailbox, a hosted spreadsheet. That settles where guests’ data goes before anyone has asked [10].
The gaps sit where the reading is slow and the decision already belongs to a named member of staff. That makes them the safer place to begin.
Three doors, one gate
An agent is only as useful as what it can reach. In a travel company that means three kinds of thing.
Documents. Fare rules, conditions of carriage, cancellation policies, brand standards, operating procedures, partner contracts. These are indexed once as knowledge sources, with each reader’s permissions kept, so an agent answers with the clause it relied on and says when the documents are silent.
Systems of record. Reservations, guest profiles, room status, flights, work orders, loyalty accounts. Most are reached through an interface the company already runs. The company gives the platform the description of that interface, chooses which operations become tools, and decides which of them need approval.
Booking and operations data for analysis. Pickup, pace, delays, claims, feedback and audit data, held in the data warehouse or a reporting database, and read only. Payment systems and departure control are not one of the doors.
All three pass one gate: every tool call is checked against the company’s policy, which can allow it, record it, hold it for a person, or block it. Where the company’s interface checks the caller, the agent acts as the person using it and cannot claim another.
Six ways into a travel company’s systems
| Route | What it reaches | What to know | |
|---|---|---|---|
| 01 | A file the user uploads | A guest’s letter, a claim, a partner contract, a banquet order | Nothing is connected. The safest start |
| 02 | A knowledge source | Fare rules, conditions of carriage, policies, brand standards, procedures | Permissions are kept per reader |
| 03 | An imported description of an interface | Reservations, guest profiles, room status, flights, work orders | You choose the operations. Changes wait for approval |
| 04 | A reporting database | Pickup and pace, delays, claims, feedback, night audit data | Read only |
| 05 | A server from the tool catalogue, run in your cluster | Databases, web search, document conversion | It runs inside; check what each tool does |
| 06 | Public sources | Passenger rights rules, regulators’ guidance, published rates | Reads only, from the publisher |
What is published. We read 102 public interface descriptions from three publishers: a hotel property management platform, a travel distribution developer programme and a hotel platform. The importer refused none. Together they carry 3,942 operations, of which 1,656 read and 2,286 are sent as changes. The hotel property management platform alone carries 3,614, and 2,112 of those are sent as changes. The distribution programme’s 54 are mostly searches: 40 read [11].
What that exercise taught us. The open interfaces are built to do things, not only to read. Nearly three in every five operations are sent as changes. So choose the operations at import, start with the ones that read, and add nothing that books, cancels, refunds or changes a rate until the approval step has been seen working.
What is not published. Passenger service, departure control, airport operations and revenue systems, and most channel and booking platforms, publish no description a file address reaches. Where nothing is published, the company supplies its own description, or the agent reads from the data warehouse.
Cards and the gate. The safe design keeps the agent on the office side. It reads no card numbers, and it denies boarding to no one.
What an agent may do
One rule runs through every agent in this paper: it prepares, and a named person decides.
- It reads what the person using it could read, and no more.
- It drafts the reply to a guest, the check of a claim, the summary of a contract, the handover note, with each statement tied to its source.
- It lists what is missing, what does not match, what falls due and when.
- It stops before anything that changes a record or reaches a guest, a passenger, a partner or an authority. A tool that changes something waits for approval.
- It decides nothing about a person. It does not refuse a passenger or a guest, does not score a colleague, and does not infer anyone’s emotions.
Four lines are drawn tighter than the rest.
What the company tells a guest or a passenger. An agent may draft the reply to a question from the booking, the policies and the conditions, with the clause it relied on. It sends nothing unseen: a service agent reads and sends every reply. The company is held to what it says, so a person says it.
Bookings, refunds and compensation. An agent may set out the options for a change under the fare or rate rules, and check a refund request or a compensation claim against the rules and the operational record. It confirms, changes and cancels no booking, and it issues no refund and no compensation. The reservations agent makes the change. The refunds team and the claims handler decide and pay.
Prices and rates. An agent may prepare the pickup and pace report, or list where published rates differ across the channels. It sets no rate and no fare, and it prices by the person for no one. The revenue manager draws the conclusions and sets every price.
Staff. An agent may draft a handover note from the log, or a training note from a procedure. It allocates shifts to no one and scores no one. The duty manager signs the note and directs the work.
What the rules already ask
This part needs a lawyer’s read before it is relied on. It reports what the texts say; it is not advice.
What a company is held to
A chatbot’s words were the airline’s words. In Moffatt v. Air Canada, a small claims tribunal in British Columbia found the airline liable for what its chatbot told a passenger. A lawyer’s note reproduces the reasoning: “It should be obvious to Air Canada that it is responsible for all the information on its website.” And: “the applicable standard of care requires a company to take reasonable care to ensure their representations are accurate and not misleading” [4]. It is one tribunal’s decision, not a rule for every market.
People must be told. The European Union’s AI Act asks that people who deal directly with an AI system “are informed that they are interacting with an AI system”, unless that is obvious [13].
Passenger and traveller rights
The duty to inform is written down. Under Regulation 261/2004 an airline must show at check-in a notice that begins: “If you are denied boarding or if your flight is cancelled or delayed for at least two hours, ask at the check-in counter or boarding gate for the text stating your rights” [12].
The amounts and the care are fixed in the text. The regulation sets compensation at “EUR 250 for all flights of 1500 kilometres or less”, with two higher bands. It gives a choice that includes “reimbursement within seven days”, and a right to “meals and refreshments in a reasonable relation to the waiting time” [12].
Decisions about a person stay with a person. Under the GDPR, a person “shall have the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her”, with stated exceptions [14]. Whether a refused refund or a refused claim falls under these words is for counsel.
Prices, fees and reviews
A personalised price must be disclosed. Directive (EU) 2019/2161 adds to what a trader must tell a consumer who buys online: “where applicable, that the price was personalised on the basis of automated decision-making” [15]. The duty sits in a directive that covers a hotel stay booked online and leaves out most passenger transport and package travel, so how far it reaches fares is for counsel.
The American dispute ended in letters, not a ruling. Delta described its technology as “a decision-support tool that simply provides informed insights for our analysts, who oversee and fine-tune the recommendations” [6]. We found no enforcement action.
Reviews must be real. The same directive adds to the list of unfair commercial practices: “Submitting or commissioning another legal or natural person to submit false consumer reviews or endorsements” [15].
Workers and data
Travel is not on the European Union’s high-risk list. Work allocation is. The AI Act lists as high-risk systems used “to allocate tasks based on individual behaviour or personal traits or characteristics or to monitor and evaluate the performance and behaviour of persons in such relationships” [13]. A rostering system would be read against this text.
The dates have moved. Regulation (EU) 2026/1744 says the date of application of the high-risk duties “is set to 2 December 2027 for AI systems classified as high-risk pursuant to Article 6(2) and Annex III, and to 2 August 2028 for AI systems classified as high-risk pursuant to Article 6(1) and Annex I” [17]. Employment sits in Annex III, so its date is 2 December 2027.
| Market | What we read | What it asks | |
|---|---|---|---|
| 01 | European Union | Regulation 261/2004, Article 14(2) | An airline that denies boarding or cancels a flight “shall provide each passenger affected with a written notice setting out the rules for compensation and assistance” [12] |
| 02 | European Union | AI Act, Article 26(7) | Before a high-risk system is used at work, “deployers who are employers shall inform workers’ representatives and the affected workers” [13] |
| 03 | European Union | GDPR, Article 32(1) | Controllers and processors “shall implement appropriate technical and organisational measures to ensure a level of security appropriate to the risk” [14] |
| 04 | United Kingdom | Competition and Markets Authority, 7 April 2025 | The new consumer law “includes an explicit ban on the posting and commissioning of fake reviews”, and “the CMA can fine them up to 10% of their global turnover” [16] |
We did not open the refund and fee rules of the United States, the package travel rules, or the rules in the Gulf, India, Singapore or Australia. A company there should ask its own counsel.
What the rules have in common
- A company answers for what it tells a guest or a passenger, whoever or whatever said it.
- A person decides a refund, a claim and anything that judges a worker.
- People are told when they deal with AI, when a price is personalised, and what their rights are.
- Guest, passenger and worker records need protecting, which is easier where the company controls the system.
A worked example: one airline and hotel group
Take a made-up travel group: one airline, 22 hotels and about 9,000 staff. Nothing here is a customer’s story, and every figure is illustrative; it is section 4 put to work.
- In the contact centre, a passenger asks whether a ticket can be changed. The agent drafts the reply from the booking and the fare rules, with the rule cited. A service agent reads it and sends it.
- On the claims desk, 340 compensation claims follow a cancelled flight. The agent sets each against the passenger rights rules and the flight record and says which condition it meets or fails. The claims handler accepts or rejects every one.
- In the groups team, the agent checks a rooming list of 180 names against the group contract and lists seven missing names. The groups desk corrects and confirms.
- For the revenue manager, the agent prepares the pickup and pace report for the next 90 days and lists the twelve dates that depart most from forecast. The revenue manager sets the rates.
- At a hotel’s front desk, the agent drafts the handover note from the night’s log: 14 late arrivals and two complaints. The outgoing duty manager checks and signs it.
- In finance, the agent sets a travel agency’s commission statement against 2,600 bookings and the contract rate and lists 41 differences. Accounts payable approves or disputes the statement.
Every one of these reads documents or records the company already has, changes no record by itself, sends nothing to a guest, a passenger or a partner, and leaves a record of who asked and what was shown.
Seven questions before an agent touches a booking
- Where does it run? On our infrastructure, in our private cloud, or in the supplier’s service? Where is the model? Can it run with no connection to the internet?
- What can it read? Does it act as the person using it, with their permissions, or as one all-seeing account? Can it reach card numbers?
- What can it change? Which tools change a booking, a guest profile, a rate, a refund or a loyalty account, and does each wait for a person?
- Can it speak for us? Show us that no reply, notice or offer reaches a guest, a passenger or a partner before a member of staff has read and sent it.
- Who decides? Show us the step where a named person approves, and what the agent does when nobody has.
- What is recorded? Who asked, which tool ran, what was decided, and for how long is that kept?
- Can we stop it? One agent, one tool, or all of them, at once, by us.
Where to start
Pick work where reading is slow, the decision stays with a person, and no guest or passenger waits on the agent.
Policy questions and answers
Colleagues ask what the fare rules, the conditions and the policies say and get the answer with the clause cited. No booking is touched.
Compensation claim check
The agent checks each claim against the rules and the operational record. The claims handler decides.
Commission statement check
The agent lists each difference between a statement and the bookings. Accounts payable approves or disputes.
One property, one station, invented data first. Then beside the manual process, measured against it.
About Katonic AI
Katonic AI builds the platform that organisations use to build, run and govern AI agents on their own infrastructure: in their own data centre, in a private or sovereign cloud, or with no connection to the internet at all. The platform is in production with governments, banks, telecom operators and utilities. Katonic is certified to ISO 27001.
Sources
Read on 22 November 2026. Figures are the publishers’ own and were not audited by us.
- Microsoft, “Air India elevates customer support while saving money with Azure AI, data, and apps”, customer story, 15 November 2024. microsoft.com
- Salesforce, “From Stress to Success: How Agentic AI Is Streamlining Vacation Planning”, 29 July 2025. salesforce.com
- Mews, “Mews launches Guest Messaging and Automations”, press release, 1 October 2026. mews.com
- Barry Sookman, “Moffatt v. Air Canada: A Misrepresentation by an AI Chatbot”, 16 February 2024: a lawyer’s note that reproduces passages of the decision, 2024 BCCRT 149. barrysookman.com
- Jason Proctor, CBC News, report of the Air Canada chatbot decision, 15 February 2024. cbc.ca
- Delta Air Lines, “Delta responds to misinformation around AI pricing”, letter of 31 July 2025 to Senators Gallego, Blumenthal and Warner, published 7 August 2025. news.delta.com
- Accelya, “Accelya launches FLX Recover to transform airline passenger recovery”, press release, 13 October 2026; and the FLX Recover product page. w3.accelya.com
- Amadeus, press release on its report on agentic AI for airlines, 4 June 2026. amadeus.com
- Cloudbeds, security page. cloudbeds.com
- n8n workflow templates 9811 and 12946 (hotel guest feedback and guest enquiries). n8n.io
- Template libraries and prebuilt agents of agent and workflow platforms, searched 22 November 2026; and 102 interface descriptions fetched from three publishers the same day and run through the platform’s importer (Katonic research).
- Regulation (EC) No 261/2004, Articles 7, 8, 9 and 14.
- Regulation (EU) 2024/1689 (AI Act), Annex III point 4(b); Articles 26(7) and 50(1).
- Regulation (EU) 2016/679 (GDPR), Articles 22(1) and 32(1).
- Directive (EU) 2019/2161: point (ea) inserted in Directive 2011/83/EU and point 23c added to Annex I of Directive 2005/29/EC.
- Competition and Markets Authority, “CMA to boost consumer and business confidence as new consumer protection regime comes into force”, press release, 7 April 2025. gov.uk
- Regulation (EU) 2026/1744, 8 July 2026, the recital on the dates of application.