Who Is Shopping for Your Passengers?

Airlines have spent decades optimizing for a human shopper: fare design, ancillary pricing, channel strategy, and loyalty, all built around someone with a browser open and a limited attention span. That shopper is now being replaced by an AI agent working from standing instructions, a saved rule set, and a tokenized payment method, comparing carriers and booking the one that clears its criteria without a human ever opening a page. The airline often has no idea the transaction happened until the ticket is issued. This isn’t a new distribution channel to optimize for; it’s a structural shift already in motion, and it changes who, or what, the airline is actually selling to.

This isn’t the first attempt to remove the human from the loop

Google Flights ran direct booking from 2015 to 2022, no AI involved, before killing it: passengers booked the cheapest fares, skipped ancillaries, and still preferred booking direct anyway. Swifty launched in 2023 out of Lufthansa Innovation Hub with fully autonomous AI booking tested live on Swiss’s own site, before Revolut acquired it in 2025 and folded it into its fintech platform. The interface keeps getting smarter, but the structure underneath it doesn’t.

Anyone who has tried to wire an AI agent into a live airline inventory system runs into the same wall almost immediately, and it has nothing to do with the AI. The bottleneck is everything required to turn intent into a valid booking inside an airline’s own systems: availability, pricing, rules, passenger data, ticketing. Getting an agent to compare fares is trivial. Getting an airline to issue a real ticket against that comparison, with no OTA and no middle layer, is not.

That bottleneck is structural, not incidental. Ann Cederhall of LeapShift describes the gap between airlines wanting to be retailers and actually being retailers as one of operational capability: legacy infrastructure (PNRs, SSRs, EMDs, fare filing) was never built to be machine-addressable. Per PhocusWire’s coverage of LeapShift’s 2026 retailing research, close to 70% of airlines still don’t use an order management system to unify a booking, and production-level MCP servers, the infrastructure that lets an agent talk to airline systems directly, sit at just 9%. The pipes aren’t ready. That’s the real constraint, not appetite.

Where the value moves when the interface stops mattering

OTAs and travel management companies spent years competing on the comparison experience itself; the search interface was the product. Agentic shopping removes that battleground. If an agent is doing the comparing, the interface stops being the differentiator.

That repositioning is already showing up in public partnerships, not just strategy decks. Skyscanner, with 160 million monthly users and two decades of behavioral data, launched a dedicated ChatGPT app in February 2026 after already serving as a launch partner for OpenAI’s Operator agent and integrating with Microsoft Copilot Actions. Its Chief AI Officer, Piero Sierra, called it moving “beyond form-fills toward dynamic, answer-led experiences.” Sabre made the same bet from the supply side: that same month it paired its Mosaic content library, covering more than 420 airlines and two million hotels, with PayPal’s agentic commerce stack, built for AI-to-system transactions rather than a human on a screen. Neither company is trying to win by building a better search box. They’re trying to be the data an agent reaches for first.

That should worry airlines too. One that treats every booking as a fresh, anonymous transaction hasn’t built the layer needed to compete on completeness of information instead of completeness of UI.

Pricing a product for a shopper who never sees the website

This is where it gets uncomfortable for commercial teams. At UATP’s Airline Distribution 2026 conference, American’s VP of sales and United’s managing director of digital sales both pointed to a look-to-book ratio that has climbed into the thousands under NDC, with AI compounding it before agentic volume is even material. United’s own framing was blunt: the issue isn’t really AI yet. Commercial constructs were never built to price for that ratio. A planner used to pricing for human search behavior is now pricing for a shopper whose criteria were set once, upstream, and never renegotiated mid-session. Layer on what OAG’s CEO has called the industry’s real exposure, fragmented data that propagates errors fast once AI is doing the searching and booking, and a planner is optimizing bundles for a buyer the airline can’t see clearly.

The deeper issue sits underneath all of it. Every AI agent shopping on a traveler’s behalf today arrives as an anonymous, cold session. The airline has zero context, zero history, zero signal. It gets priced and served like any other unknown request: full cost, zero qualification, the same way a loyalty member who forgot to log in gets treated like a stranger. NDC solved the offer problem by handing airlines back control of the product, but it never solved for who’s on the other end of the query. That’s not a distribution problem; it’s a recognition problem.

This is a recognition problem, not a technology one

None of this is an argument for bolting a chatbot onto an existing booking flow. Every one of these three vantage points traces back to the same gap: airlines don’t have a way to recognize a shopper, human or agent, that persists across sessions, channels, or shopper type. AI doesn’t create that gap. It just moves fast enough to expose it.

This is the decision airline commercial and IT leaders are actually making right now, mostly without recognizing they’re making it. Every fare filed without a plan for machine-readable rules, every booking flow that still treats a returning traveler as a stranger, every roadmap that adds an AI interface without touching what sits underneath it: these are the decisions that will determine which carriers can still tell a real customer from a cold, anonymous request once a meaningful share of shoppers aren’t human at all.

The airlines that solve recognition first will be the ones still setting the terms of the sale. The ones that don’t will keep discovering, one full-price anonymous booking at a time, that NDC gave them the product back without ever telling them who’s buying it.