Resources · Glossary
A working glossary of travel AI
Plain definitions of the terms that come up when travel operations meet AI, from allotments and bedbanks to grounding, evals and the Model Context Protocol. Where a page explains a term in depth, the term links to it.
Terms, A to Z
- Accessibility constraint
A routing rule that reflects a traveler's access needs, such as step-free paths, maximum walking distance, rest intervals or seating, treated as a hard limit rather than a note. Good itineraries verify these details with suppliers instead of assuming them.
Where this shows up- Agent-safe checkout
A booking flow designed for AI agents acting on a traveler's behalf: quotes that expire, short holds, spending limits, explicit traveler confirmation, and idempotent requests so a retry never books twice.
Where this shows up- Agentic commerce
Buying and selling in which an AI agent searches, compares and books on a person's behalf. For travel suppliers, it means making inventory, policies and checkout usable by software, not only by people on a website.
Where this shows up- AI agent
Software that uses a language model to pursue a goal across several steps, calling tools such as search, availability checks or booking along the way. In travel, agents plan trips, compare options and increasingly transact.
Where this shows up- AI search visibility
How often, and how accurately, AI assistants mention and cite a business when travelers ask relevant questions. Best measured with a fixed set of test prompts, run on a schedule, so changes are comparable over time.
Where this shows up- Allotment
A block of rooms, seats or tour places that a supplier sets aside for a seller at agreed rates until a release date. Unsold places go back to the supplier after release.
Where this shows up- API key
A secret string that identifies the application making a request. Scope each key to one environment and the smallest set of permissions it needs, and rotate it on a schedule.
Where this shows up- Attribution
Recording which channel, campaign or AI agent led to a booking or a citation. For agent traffic, attribution travels with each tool call and booking, so you can see which assistants actually send business.
Where this shows up
- Bedbank
A wholesaler that contracts hotel rooms in bulk and resells them to travel sellers through APIs, usually at net rates. Bedbank content often needs mapping before it can sit alongside direct contracts.
Where this shows up- Booking reconciliation
Matching what was booked, confirmed, invoiced and paid across suppliers and payment providers, and flagging the differences. Often the last manual step in booking operations, and the one that hides the most hours.
Where this shows up
- Cancellation risk score
A per-reservation estimate of how likely a booking is to cancel before arrival, produced by a model trained on your own history. Useful when it shows its reasons and leads to a specific, considerate action.
Where this shows up- Channel manager
Software that distributes a property's rates and availability to multiple sales channels and pulls bookings back, keeping inventory in sync so the same room isn't sold twice.
Where this shows up- Constraint solver
An algorithm that searches for a plan satisfying hard rules, such as opening windows, capacity and travel time, while optimizing soft goals like pacing, preferences and cost. The engine underneath any itinerary that actually works on the day.
Where this shows up- Content mapping
Linking records that describe the same hotel, tour or venue across different suppliers and feeds, so that one real-world thing has one identity in your systems.
Where this shows up
- Data contract
An agreed definition of what a data feed contains: fields, formats, freshness, owner, and what happens when it breaks. Integrations with data contracts fail loudly instead of quietly.
Where this shows up- Data minimization
Collecting and keeping only the traveler data a task actually needs, for only as long as it needs it. Especially important for preference data that can reveal health, family circumstances or religion.
Where this shows up- Demand forecasting
Predicting future bookings or visits by date, product and segment from history, booking pace and outside signals such as events and weather.
Where this shows up- Disruption playbook
A prepared response to a known kind of disruption, such as weather, a closure, a delay or a supplier no-show, with triggers, pre-approved alternatives, message templates and an approval path.
Where this shows up- DMC (destination management company)
A company with local expertise that designs and operates ground arrangements, including transfers, guides, activities and venues, for tour operators, agencies and corporate groups.
Where this shows up- DMO (destination organization)
A destination marketing or management organization: the public or non-profit body that promotes a place and, increasingly, manages visitor flows, seasonality and relations with residents.
Where this shows up- Draft, don't decide
Our working principle for AI in travel operations. Software prepares proposals, re-plans and messages; a person approves anything that affects a traveler's money, safety or access.
Where this shows up- Dynamic packaging
Assembling a trip from separately priced components, such as stays, activities and transfers, at the moment of request, with one total price. Packages built this way follow live availability rather than last year's brochure.
Where this shows up
- Energy budget
A daily allowance of physical and mental effort for a traveler, spent on walking, climbing, driving and time on the move, and restored by rest. Routing that respects it produces days people can actually enjoy.
Where this shows up- Evals
Short for evaluations: repeatable test sets that score an AI feature's outputs for accuracy, safety and usefulness before release and continuously afterward. An itinerary generator without evals is an untested product.
Where this shows up
- First-party data
Information a business collects directly from its own travelers and operations, with their knowledge, such as preferences, bookings and feedback, as opposed to data bought from or inferred by third parties.
Where this shows up
- GDS (global distribution system)
A large network that aggregates airline, hotel, car and rail inventory and lets travel sellers search and book it. GDS feeds bring scale, along with their own content quirks.
Where this shows up- GEO (generative engine optimization)
Work that makes a brand more likely to be cited accurately in AI-generated answers: answerable content, structured data, consistent facts across the web, and live operational information.
Where this shows up- Grounding
Constraining an AI system's answers to verified, current sources, such as your inventory, opening windows, closures and policies, and declining to answer beyond them. The main defense against confident wrong answers.
Where this shows up
- Hallucination
A fluent, confident output from a language model that its sources don't support, like a café that closed two years ago or a trail that doesn't exist. Reduced by grounding and caught by evals.
Where this shows up- Human-in-the-loop
A workflow in which a person reviews or approves an AI system's output before it takes effect. In travel, the loop belongs wherever money, safety, accessibility or a supplier relationship is at stake.
Where this shows up
- Idempotency key
A unique value sent with a write request so that retrying the same request, after a timeout for example, never creates a second booking or charge.
Where this shows up- Intent capture
Turning a traveler's words ("somewhere with lakes, nothing scrambly") into structured preferences and constraints an itinerary engine can use, while keeping the original wording for the advisor.
Where this shows up
- Live re-planning
Revising a trip in progress when conditions change, by re-solving the remaining days around the new facts rather than cancelling and rebooking piece by piece.
Where this shows up- llms.txt
A proposed convention: a plain Markdown file at the root of a website that points AI systems to the pages and facts that matter most. Useful alongside well-structured pages, not instead of them.
Where this shows up- Low-carbon routing
Planning routes and modes that reduce a trip's emissions, such as rail over short flights, fewer backtracks and longer stays, without making the trip worse.
Where this shows up
- MCP (Model Context Protocol)
An open protocol for connecting AI assistants to tools and data. A supplier's MCP server lets compatible assistants search its catalog, check availability and, where the supplier allows it, book.
Where this shows up- Model drift
The gradual loss of accuracy as the world a model learned from changes: new booking patterns, new channels, a different season. Caught by monitoring and corrected by retraining or recalibration.
Where this shows up
- Opening window
The hours during which a place can actually be visited, including last entry, seasonal closures and booking-only slots. One of the hardest constraints in an itinerary, and the easiest to get wrong.
Where this shows up
- Pacing
How effort and rest are distributed across a day and a trip. Two itineraries with identical stops can feel completely different depending on pacing.
Where this shows up- PMS (property management system)
The core operating system of a hotel or lodge: reservations, room inventory, rates, check-in and billing. Usually the system of record for what is really available.
Where this shows up- Preference profile
A structured, consented record of what a traveler likes and needs, such as pace, interests, dietary needs and mobility, that personalization draws on. Trust depends on travelers being able to see and edit it.
Where this shows up- Price sensitivity
How strongly demand for a product responds to a change in price, which varies by segment, lead time and season. Knowing it tells you when a package will work better than a discount.
Where this shows up- Prompt injection
An attack in which text hidden in content, messages or tool results tries to make an AI system ignore its instructions, such as a review that tells an assistant to issue a refund. Mitigated by permission boundaries and by treating outside text as data, never as instructions.
Where this shows up
- RAG (retrieval-augmented generation)
A pattern in which an AI system first retrieves relevant documents or records and then writes its answer from them. A common way to ground a concierge in current, operator-specific information.
Where this shows up- Rate limit
A cap on how many API requests a client may make in a given window. Well-behaved clients read the limit headers and back off when asked.
Where this shows up- Release date
The date by which unsold allotment must be handed back to the supplier. Miss it, and you may pay for places you never sold.
Where this shows up
- Shoulder season
The periods between a destination's peak and off-peak seasons, when weather is often good, prices are lower and capacity is available. The weeks where demand shaping pays off most.
Where this shows up- Structured data
Machine-readable markup, usually Schema.org vocabulary, that describes a page's facts, such as a tour's duration, a lodge's address or an event's dates, so search engines and AI systems read them reliably.
Where this shows up- Supplier confirmation
A supplier's acknowledgement that a requested service is booked, arriving by API, portal, email or occasionally phone. Collecting, parsing and filing confirmations is one of the largest hidden costs in tour operations.
Where this shows up
- Tool call
A structured request an AI model makes to an outside function, such as checking availability or creating a quote, instead of answering from memory. The basic unit of work between an agent and your systems.
Where this shows up- Travel time matrix
A table of travel times between every pair of stops, by mode and time of day. Route optimization reads from it constantly, so its accuracy sets the ceiling on itinerary quality.
Where this shows up
- Visitor dispersal
Spreading visitors across more places and times so popular sites aren't overwhelmed, by suggesting comparable alternatives and quieter hours rather than telling people where not to go.
Where this shows up- Voucher
The document or code a traveler presents to a supplier as proof of a prepaid service, with the booking reference, inclusions and meeting instructions.
Where this shows up
- Weather exposure
How vulnerable a stop or route segment is to weather, like an exposed ridge in monsoon season or an open boat in high wind. Routing can schedule exposed segments for the hours when forecasts are kindest.
Where this shows up- Webhook
An HTTP request a platform sends to your system when something happens, such as a booking confirmed or a route re-planned, so you don't have to keep polling for changes.
Where this shows up
Related
Keep exploring
- BlogField NotesPractical writing on agentic travel, itinerary science, predictive analytics and booking operations.
- ResourcesAgent-ready checklistA 49-item checklist for travel suppliers preparing for AI agents: content, policies, live availability, safe checkout, permissions, monitoring and legal terms.
- PlatformPlatformSix modules for proposals, routing, booking, prediction, in-trip help and AI-agent checkout, on one shared data layer. Start with one, add the next.
- ResourcesDevelopersREST API, webhooks and MCP access to itineraries, routes, quotes, bookings and signals. OAuth 2.0, environment-scoped keys, sandbox access by request.
Hearing a term we haven't covered?
Tell us which word keeps coming up in your vendor calls. We'll add it here, or explain it properly on a call.