What problem does it solve?
Planning a trip means comparing many options. Traditional online travel search, as Priceline describes it, requires navigating filters, tabs and separate browser windows. Search boxes work when the traveller already knows the destination and dates; they fail at the questions people actually have ("where is warm in February with a direct flight", "which of these hotels is quiet and walkable", "can we bring the dog"). On the service side, the same pre and post booking questions come back again and again: when Booking.com widened access to its Booking Assistant service chatbot in 2017, it listed payment, transportation, arrival and departure times, date changes, cancellations, parking, extra beds, pet policies and WiFi among the most frequently asked topics.
A concierge that only chats about destinations adds little. The value comes when the assistant is grounded in live prices and availability, in the operator's own property content and policies, and in the traveller's booking, so it can recommend, answer precisely, book or change, and hand the conversation to a human agent or the property when needed. Several companies on this page, among them Priceline, Trip.com and Holland America Line, run one assistant for both planning and service.
How does it work?
- Understand the trip. The assistant asks for or infers the essentials (who travels, dates or flexibility, budget, what matters) in the traveller's own words and language.
- Search live inventory. It queries the booking engine or supplier APIs for flights, hotels, rentals, cruises and activities with current prices and availability, never prices from memory.
- Compare and explain. It shortlists options and explains the tradeoffs from property content, reviews and policies, with links to each listing.
- Book or hand off. It builds the basket and passes the traveller to checkout, or books within set limits, and hands group, complex or high value requests to a human travel agent.
- Support the trip. After booking, the same assistant answers questions about the reservation, makes changes and cancellations within the policy, passes requests to the property and hands complaints and exceptions to a person with the context.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Early adopters
- Channels
- Mobile app, Web chat, WhatsApp, Phone and voice
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Automation rate | Too few to pool | 30% | 1 | 1 organization |
| Containment rate | Too few to pool | about 45% | 1 | 1 organization |
| Cost reduction | Too few to pool | about 16% | 1 | 1 organization |
| Time saved per task | Too few to pool | about 10 minutes | 1 | 1 organization |
Value drivers: Revenue growth, Customer experience, Lower cost to serve, Inclusion and access.
Indicative value
An online travel company or hotel group with 2 million bookings a year
USD 480,000 to USD 2.9 million
Human handled service contact cost avoided per year
How this is calculated
Formula: bookings * contactsPerBooking * containment * costPerContact. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Bookings per year bookings, bookings per year | 2,000,000 | 2,000,000 | The reference organization. |
| Assisted service contacts per booking contactsPerBooking, contacts per booking | 0.2 | 0.4 | Editorial assumption for pre and post booking questions and changes. Replace with your own contact rate. |
| Share of service contacts the assistant resolves containment, fraction of contacts | 0.3 | 0.45 | The high end matches the benchmark on this page (Airbnb reports nearly 45% of issues that begin with its AI assistant resolved without a human agent in Q2 2026); the low end allows for a first year. |
| Cost of a human handled contact costPerContact, USD per contact | 4 | 8 | Editorial assumption for a blended chat and phone contact. Replace with your own fully loaded cost. |
What it leaves out: Service cost only. It leaves out the revenue effect of better conversion (the most important and least published benefit), the cost of running the AI and the integrations, and any change in cancellations or complaints.
Who already uses it?
5 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Priceline
United States · Travel and hospitality · 2026
Priceline put its AI assistant Penny in front of customers first at checkout and in customer care. In June 2026 it announced a fully agentic version that takes a trip idea, compares hotels, flights and rental cars on a live map with real time inventory and deals, and books without leaving the conversation. Penny runs as more than ten specialized agents on Priceline's own AI stack, with Claude models for conversational reasoning and planning, and Google Cloud and OpenAI supporting search and voice capabilities.
- Time saved per task: about 10 minutes, per trip, Penny users versus customers who called support
"Priceline has estimated that travelers who used Penny saved an average of nearly ten minutes per trip compared with those who called customer support."
Claimed by: organization
Airbnb
United States · Travel and hospitality · 2025
Airbnb runs an AI assistant as the first line of customer support for guests and hosts. It was expanded to all US users in 2025 and then rolled out to more countries and languages, reaching more than 50 languages by mid 2026. Airbnb reports the share of issues resolved without a human agent in each quarterly letter and links part of the fall in support cost per booking to the assistant; it plans an AI voice assistant.
- Containment rate: about 45%, Q2 2026, issues that begin with the AI assistant
"Nearly 45 percent of issues that begin with our AI assistant are now resolved without a human agent, up from Q1, while delivering much faster resolution times."
Claimed by: organization - Cost reduction: about 16%, Q2 2026 year on year, customer support cost per booking
"In Q2, our customer support related cost per booking declined approximately 16 percent year-over-year, driven in part by improvements to our AI assistant."
Claimed by: organization
Booking.com
Netherlands · Travel and hospitality · 2025
Booking.com has moved from a standalone AI Trip Planner to AI features embedded across the booking journey. AI Trip Support is a first point of contact around the clock that answers questions about a property (for example parking) and helps travellers change reservations, with a handover to a human for complex cases. AI Voice Support lets travellers manage or cancel a booking by phone in their own words, and connects to a human agent with the context when needed. An earlier generation, the Booking Assistant chatbot, answered stay related questions from 2017.
- Automation rate: 30%, December 2017, pilot version of the earlier generation Booking Assistant, English language bookings
"The chatbot can currently respond to 30% of customers’ stay-related questions automatically in less than 5 minutes."
Claimed by: organization
Trip.com
Singapore · Travel and hospitality · 2023
TripGenie is the AI travel assistant in the Trip.com app and website. It helps travellers find inspiration, compare hotels and book hotels, flights and attractions, answers pre and post sales service questions, and during the trip offers menu help, live translation and questions about images. After three years of use, Trip.com reports that nearly 60% of TripGenie interactions are booking related, that service questions are about a quarter of interactions, and that TripGenie assisted order volume grew by around 400% year on year.
No outcome disclosed.
Holland America Line
United States · Travel and hospitality · 2024
Holland America Line built Anna, a generative AI digital concierge on its website for new and existing cruise guests and the travel advisors who book for them. The first release supports booking new cruises, adding products and services to existing bookings and general questions, and connects to the CRM and reservation data. It was rolled out in waves (contact centre agents, employees, then 5%, 50% and 100% of website visitors) and runs in the United States, with more markets and languages planned.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Live availability and pricing through the booking engine or supplier APIs
- Structured property, cabin or room content (amenities, accessibility, parking, pets) with an owner
- Booking, change and cancellation policies per rate and product
- Contact reasons and search logs to choose the first intents
Systems to integrate
- Booking engine, central reservation system or GDS and supplier APIs
- Customer account and loyalty system for authentication and personalization
- Payment service for checkout, deposits and change fees
- Property management system or messaging to pass requests to the hotel or host
- Contact centre platform for handover with the conversation context
Complexity: Medium
A planning chatbot on public content is quick to build. A concierge that quotes real prices, books and changes reservations needs live access to inventory, pricing and the booking system, authentication for post booking changes, and strict rules so it never states a price or policy it did not retrieve.
- 1
Start where the booking already exists
Post booking questions (parking, check in times, what is included, change a date) are frequent, well defined and measurable. Launch there first, prove containment and satisfaction, then move up the funnel into planning and search.
- 2
Ground every price and fact
Let the assistant quote prices, availability and policies only from live tool calls and approved content, and show where the answer came from. A fluent but invented price or policy costs more than no answer.
- 3
Keep checkout deterministic
Let the model build the basket, but run payment, terms acceptance and confirmation in a fixed flow, with the price and cancellation conditions shown exactly as the booking engine returns them.
- 4
Roll out in waves
Follow the Holland America Line pattern: internal agents first, then employees, then a small share of website visitors, widening only when resolution and satisfaction hold.
- 5
Measure revenue, not just deflection
Run the assistant against a control group and measure conversion, basket value and cancellations, not only contained conversations. Priceline reports higher conversion and fewer support contacts for Penny users, Trip.com reports growth in orders assisted by TripGenie, and Airbnb reports service effects only.
- 6
Test before travellers do
Keep a regression set of planning and service conversations per market and language, including requests for prices the rate does not allow and attempts to change someone else's booking, and run it on every change.
Guardrails
- Prices, availability and policies only from live tool results and approved content, never from the model's memory
- Authentication before showing or changing a booking; changes only through an allow list of actions with limits
- Checkout, payment and terms acceptance in a deterministic flow with card data tokenized
- Handover for complaints, accessibility needs, groups and high value or complex itineraries
- Recommendations free of undisclosed paid placement, with sponsored results labelled
KPIs to instrument
- Conversion and basket value for assistant users versus a control group
- Containment on post booking contacts, counting repeat contacts within seven days as not contained
- Share of answers with a price or policy that did not match the booking engine, on a sampled review
- Handover rate and reasons
- Customer satisfaction on assistant conversations versus human handled ones
Human in the loop
Human travel agents handle complex, group and high value trips, complaints and exceptions to policy. A content owner approves property and policy content, and a team reviews a weekly sample of conversations for wrong prices, wrong policies and unfair recommendations.
Common failure modes
- Invented prices and policies
- The assistant states a price, fee or refund rule from memory, and the company is held responsible for it, as in the Air Canada tribunal case. Retrieve every fact and refuse when retrieval finds nothing.
- A planner nobody books from
- Engagement grows but conversion does not, because the assistant is not connected to live inventory and checkout. Measure bookings, not conversations.
- Steering that breaks consumer law
- Recommendations favour higher commission options without disclosure, or hide fees until checkout. Label sponsored results and show the full price early.
- Handover without context
- The traveller has to repeat the trip details to a human agent. Pass the summary, the booking and the options already shown.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
A customer facing assistant must tell people they are interacting with AI unless that is obvious from the context (Article 50(1), applicable from 2 August 2026). Recommending and booking travel is not listed in Annex III, so it is not high risk; consumer protection law on price transparency and fair commercial practices still applies to what it says.
Rules that apply
Guidance
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). Travellers must be informed that they are interacting with an AI system unless this is obvious from the context.
- Package travel, package holidays and linked travel arrangements in the EU (Your Europe) (European Union, Europe). When an assistant combines flights, hotels and other services, the combination can become a package or linked travel arrangement with information duties and traveller rights.
- Unfair commercial practices directive (European Commission, Europe). Rules against misleading information and practices, relevant to how an assistant presents prices, fees, rankings and sponsored results.
Controls to put in place
- AI disclosure at the start of every conversation
- Price and policy statements logged with the tool result they came from
- Versioned property and policy content with an owner and review date
- Labelling of sponsored or commission based recommendations
- Change control and regression tests for every new market, language or action
When it went wrong elsewhere
- Incident 639: Air Canada Chatbot Reportedly Provides Inaccurate Bereavement Fare Information, Leading to Customer Overpayment. A Canadian small claims tribunal held Air Canada responsible in 2024 for its website chatbot's wrong statement about bereavement fare eligibility and ordered it to pay damages, rejecting the argument that the chatbot was a separate legal entity. A company can be held responsible for what its assistant says about fares and refund rules.
Frequently asked questions
- Do AI travel assistants actually increase bookings?
- Some operators report it, few publish numbers. Trip.com says TripGenie assisted order volume grew about 400% year on year, and Priceline says Penny users show higher conversion in early testing without giving a figure. Measure conversion against a control group before claiming revenue.
- How much service volume can the assistant take?
- Airbnb reports that nearly 45% of issues that begin with its AI assistant were resolved without a human agent in Q2 2026, and that support cost per booking fell about 16% year on year, partly because of the assistant. Priceline estimates Penny users saved nearly ten minutes per trip compared with calling support.
- Should the assistant book on its own?
- It can build the basket and handle simple changes within limits, but payment, terms and confirmation should run in a fixed flow with the exact price and conditions from the booking engine. Complex, group and high value trips belong with a human agent.
How to cite this page
Blits.ai AI Use Case Library, "AI travel and hotel booking concierge", last verified 26 September 2026, https://www.blits.ai/ai-use-cases/travel-and-hotel-booking-concierge. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published