Airline ecommerce and AI agents: a buyer’s integration guide
Airline commerce for AI agents — also described as agentic commerce or agentic retailing — means enabling an authorized agent to shop, book, buy ancillary products, upgrade and manage trips through supported airline systems. Aviark builds airline-side software for these workflows, under the airline’s rules. A useful answer becomes a commerce outcome only when the approved action and its result can be verified.
Published by Aviark · Updated
What does an airline ecommerce platform need to support?
Airline ecommerce connects the sale of flights and optional products to payment, orders and fulfillment. A buyer needs to evaluate the whole trip: finding a fare, selecting seats or bags, completing checkout and managing the purchase afterward. The relevant capabilities may span several systems rather than one platform.
An AI assistant adds another customer interface to that commerce journey. The airline still needs to determine which offers and actions the assistant can access, who approves a purchase, where the traveler pays and how the completed order is verified. Aviark builds the airline-side software for those supported agent workflows.
- Retailing and merchandising
- Define eligible fares, bundles, seats, bags and upgrades, with current prices and conditions. Ask whether the agent can retrieve the same relevant offers and restrictions as the airline’s other supported channels.
- Booking and payment
- Create the intended order and establish who collects payment. Aviark’s approach keeps payment on the airline’s checkout; a production integration must verify the payment result and its connection to fulfillment. Sandbox payment is simulated.
- Delivery and servicing
- Confirm the ticket or order, each purchased service and later changes. Ask what happens to seats, bags and refunds when one part succeeds and another fails.
- Agent access and operations
- Scope access to the right traveler and order, bind consent to the quote, prevent duplicate execution and give unresolved cases an owner. Require evidence for the specific backend and caller.
Use the seats, bags and upgrades worksheet to evaluate ancillary fulfillment and contribution. Use the transaction guide for consent, authorization and recovery checks. These are buyer evaluation tools, not claims of complete Aviark production coverage.
Core scope and current sandbox
Shopping, booking, ancillaries, upgrades and servicing are core product scope. Disruption is one use case. A traveler request or an airline event can initiate a workflow; neither determines which workflow a customer should start with.
Swipe the table to compare scope with current sandbox support.
| Workflow | Core product scope | Current sandbox boundary |
|---|---|---|
| Shopping and booking | Find eligible offers and complete an authorized purchase. | Search, fare choice and booking run on the hosted sandbox airline against a test backend. |
| Ancillary purchases | Buy relevant seats and bags during or after booking. | A paid seat at booking is supported. Seat changes and added bags after booking run on the sandbox airline’s own ledger, not through a booking system; fulfillment against an airline backend needs implementation and validation. |
| Cabin upgrades | Quote and purchase an eligible upgrade with explicit consent. | An exposed end-to-end cabin upgrade workflow is not yet complete. |
| Servicing | Manage changes and cancellations, including purchased services. | Date changes and cancellation are supported. Preserving purchased services through a change remains incomplete. |
The demonstration uses fictional Halcyon Air, simulated identity, payment and events, and a test backend. It demonstrates part of the execution foundation, not production airline coverage. A combined flight change, preserved seats and bags, and optional upgrade remains a product target.
How AFCI fits with NDC and PSS systems
AFCI means Agent-Facing Commerce Interface. It is the open specification Aviark stewards; Aviark’s maintained implementation and operating support are separate from the specification. Interface definitions describe possible capabilities. They do not establish that a particular airline adapter implements them.
Aviark’s integration approach connects an airline-controlled agent endpoint to supported backends, including New Distribution Capability (NDC) implementations, passenger service systems (PSS), order systems and ancillary services. It complements those systems: the authoritative backend must supply the offers, eligibility, prices and order state needed for each action.
IATA’s NDC overview describes the distribution standard; its ONE Order overview explains the move toward integrated order management. For procurement, ask for the actual implemented operations, versions and fulfillment records, not only a standards label.
Existing airline vendors may overlap with Aviark’s scope, so each engagement starts by identifying a specific capability or operating gap. Discovery establishes the system version, booking authority, reachable orders and supported products, and who owns identity, payment, deployment, data flows and recovery. Agent channels need their own authentication and integration testing.
What governed execution requires
- Retrieve and quote. Read current order and service state. Check eligibility and quote the selected action, including price, conditions and expiry. An event is a prompt to check, not permission to act.
- Separate permissions. For servicing, obtain traveler consent to the exact quoted action and separate airline authorization. Airline policy may allow a routine case without an operator; it does not replace traveler consent. Material changes require fresh approval.
- Execute and reconcile. Submit the approved action through a supported adapter. Retries must avoid duplicate execution. An uncertain backend outcome requires reconciliation before another attempt.
- Verify the whole result. Read authoritative state again and identify completed, pending and failed components. A changed flight alone does not prove that purchased seats or bags were preserved.
Agent platforms and endpoint discovery
Aviark’s airline execution core is platform-neutral. ChatGPT, Claude, Gemini, Meta Muse, airline assistants and corporate agents are potential callers. The sandbox has run with Claude and Meta Muse, through connectors Aviark set up rather than listings those platforms reviewed; every other integration needs its own validation, and a working connector is not production platform support.
The endpoint publishes a signed, airline-branded server card. The open schema gives callers a shared interface, while each airline and platform connection still needs capability and authentication testing. Listings are subject to each platform’s review terms; a server card or registry entry alone does not make an agent find or connect to the airline. Discovery does not establish authority to purchase or service an order.
Data, ticketing and compliance responsibilities
The airline controls deployment, access and data policy. Authorized agents receive information needed for the traveler’s task, so data flows and platform retention must be reviewed. Aviark persists workflow and audit state; the order system remains authoritative. Airline control does not mean that no information leaves its environment.
Accreditation and ticketing depend on the issuing channel, authorization model and settlement arrangement for the selected workflow; the airline retains its policies, customer data and settlement. Payment stays on the airline’s own checkout: the agent never touches a payment credential, the traveler pays on the airline’s checkout page under its existing payment provider, 3-D Secure and PCI scope, and the agent receives a signed result and the audit trail. Aviark’s sandbox payment is simulated; production checkout and payment verification must be validated with the airline’s provider.
Aviark records the action, permission and outcome for review. Compliance depends on the workflow, jurisdiction and production implementation. The sandbox does not demonstrate automatic refund settlement or regulatory compliance. Agree who owns exceptions and reconciliation before rollout.
Open AFCI, maintained Aviark implementation
AFCI is an open specification: the text is licensed under CC BY 4.0 and the schemas under Apache-2.0. Any airline or vendor can implement it. Aviark sells its maintained implementation and operating support; an open interface does not make the commercial software or implementation engagement free.
Every call that creates, pays for or services an order travels between the agent and the airline’s own endpoint; no other party, Aviark included, stands in that path or is paid per booking on it. The airline retains its data and policies. Agreements define data and configuration export, software licensing and transition support. Aviark retains its reusable product and adapter IP.
Request the AFCI specificationChoose a bounded first workflow
Start with a customer problem, a real caller, accessible orders and a measurable gap. The first workflow may be a purchase, upgrade or change. Agree the supported population, exception path, acceptance criteria and operating owner before rollout.
Evaluate fulfilled purchases and incremental contribution against a baseline. For servicing, examine verified resolution, preserved entitlements, repeat contacts and total operating cost. These are evaluation measures, not established Aviark revenue uplift or savings claims.
Aviark starts with a two-week discovery: time with your sales, servicing and technology teams, then an audit of your offer, order and servicing systems. You get project proposals, sized and priced. Scope implementation separately from annual workflow coverage, capacity and support, with added scope priced explicitly. Validate backend behavior, failure handling and operational readiness before agreeing a limited production rollout.
Discuss your commerce workflowPut the workflow to the test
Use these resources to examine execution controls and qualify a first commercial workflow. Each distinguishes evaluation requirements from current Aviark evidence.
How an airline AI agent completes a transaction
System responsibilities, consent, execution and recovery, with dated synthetic test findings.
Read the transaction guideSeats, bags and upgrades readiness
A practical checklist and scorecard for fulfillment, contribution and operating cost.
Use the readiness worksheet