Computer Says No: Why Airlines Won't Take Your Upgrade Money

From the Travel Tech Podcast newsletter for episode 22, Computer Says No: Why Airlines Won't Take Your Upgrade Money.

A friend of Ann Cederhall's wanted to pay out of her own pocket to upgrade from economy to business. She opened the airline app at the airport. The system said no. Not "unavailable." Not "sold out." Just no, with no explanation. Ann knew exactly why: the booking lived in a travel seller's PNR, the upgrade would move it to an airline PNR, and the two systems had no way to talk to each other. The technical reason for refusing money is a decade old and still hasn't been fixed. I wanted that perspective on the show.

She started as a consultant, moved through Amex and Amadeus, and landed a role at Scandinavian Airlines that gave her a view of distribution contracts from the airline side. She then joined Lufthansa in 2014, where she and Stefan Bingemer built three Direct Connect implementations that became the first commercial NDC deployments, triggering the infamous 16 Euro GDS surcharge that sent shockwaves through the agency community. She went on to publish the State of Airline Retailing 2026 report, which surveyed what airlines are actually doing with their retailing infrastructure today. That report is what brought her to the show. You can listen on Apple Podcasts, also streaming on Spotify and YouTube.

The thread running through the whole conversation was a specific fault line: airlines have had the commercial motivation to modernize retailing for over a decade, and most of the pieces they need have existed in some form for almost as long. Yet the gap between what airlines say they want to sell and what their systems can actually deliver has barely closed.

The gap between what airlines want to sell and what their systems can deliver

That tension has a structural cause. NDC, the messaging standard designed to let airlines sell directly and flexibly, was introduced in 2012. But the standard did not include the ability to change, cancel, or refund a booking made through it until version 24.1, released in 2024. That is 12 years of a distribution standard that let airlines take a booking and then left them unable to service it. Ann cited the Delta COO making the same point directly: you cannot implement something you cannot service. The consequence is that approximately 70% of airlines currently on NDC are still running version 17.1, the version before servicing existed, because upgrading is a major project and skipping straight to 24.1 is the only version worth the effort.

The data from Ann's 2026 survey adds another layer. Nearly 70% of airlines have no order management system, no centralized orchestration layer that tracks what has been sold, to whom, under what conditions, and what can be done with it post-booking. Amazon has had one since day one. Airlines are instead coupling third-party providers (Booking.com for hotels, external eSIM vendors, local ground services) without any internal system to verify that what they sold actually exists at the time of use. Ann waited eight months for a refund on a fast track service that was not open at 5am when she needed it. She eventually filed a chargeback. The airline never acknowledged the inventory failure. AI is entering this environment with real promise in specific areas: revenue management agents that can synthesize competitor cancellation rates, demand signals, and pricing data overnight are a genuine step forward, and Ann highlighted Fetcher and Telos as examples of new thinking in that space. But, as she put it directly, no agentic layer fixes a shopping engine that hasn't been modernized in 30 years and still generates 20 million fares to surface a handful of relevant results. Interlining still runs on teletype messaging. Proration disputes between interline partners generate between $650 million and $3 billion in annual revenue leakage globally, and many are settled by one party simply giving up because the cost of the dispute exceeds the amount at stake.

Ann Cederhall, co-founder of LeapShift and author of the State of Airline Retailing 2026 report, covers NDC's servicing gap and version fragmentation, the order management absence at most airlines, upgrade economics and the travel wallet effect, what AI can realistically do in revenue management and what it cannot, and the mechanics of interline revenue leakage.

Key Takeaways

• NDC without servicing is a booking tool, not a retailing tool: The standard was commercially limited from launch because airlines could sell through it but couldn't change or refund those bookings, a constraint that lasted until 2024 and still affects 70% of current implementations.
• Order management is the missing infrastructure layer: Without a system controlling inventory, post-booking actions, and ancillary fulfillment, airlines are outsourcing accountability to third-party vendors with no visibility into whether the product was delivered.
• Upgrade revenue requires proximity to departure: Willingness to pay rises sharply at the airport, but the PNR ownership problem means most airlines cannot process an upgrade when the booking lives in an agency system, and the messaging to the customer reflects exactly that failure.
• AI accelerates revenue management but does not replace the data architecture under it: Overnight agents harvesting competitor signals and demand patterns are a meaningful improvement, but they run on top of proration and inventory systems that have not changed in decades.
• Revenue leakage is the long tail of problems no one got around to fixing: Interline settlement disputes, uncollected taxes, and missing ancillary proration cost the industry up to $3 billion annually, not because the problem is unsolvable, but because resolution has historically cost more in staff time than the disputed amounts.

🔧 Three to Build

Ann Cederhall's case is simple: airlines have built sophisticated ways to sell and almost nothing to service, settle, or recover revenue after the fact.

Build a post-booking transaction processor that handles refunds and exchanges without human queues.

NDC spent twelve years as a booking only pipe. There was no standard mechanism to cancel, change, or refund an order until the 24.1 spec landed in 2024. That gap forced servicing back onto legacy PSS queues or manual call centre work, killing any cost case for modern retailing. An RPA layer sitting between your order management surface and the PSS can process the high volume, rules driven cases automatically while routing edge cases to agents.

Catch: the bottleneck is rarely the automation logic. It is the carrier granting your system the credential scope to write back to the reservation.

→ Automate post-booking transactions

Build a revenue leakage detector that audits booking channel transactions against settled amounts.

Interline alone is estimated to leak between $650 million and $3 billion annually across the industry, and most carriers cannot tell you their own share of that number. A data pipeline joining booking records, fare filed data, and settlement reports can surface the discrepancy before it becomes a reconciliation problem. Running it continuously, rather than as a quarterly audit, means you catch systematic mis-pricings while the bookings are still recent enough to dispute.

Catch: the hardest data to get is not your own. Interline partner feeds are slow, inconsistent, and often batch only, which limits how real time your detection can actually be.

→ Identify and address revenue leakage

Build a centralised inventory view that aggregates sold seats and available capacity across all channels in one place.

The episode cites a survey finding that nearly 70 percent of airlines have no order management or orchestration system. Without one, an upgrade offer made through one channel has no reliable way to check what a different channel already sold. An AI powered allocation layer can only optimise what it can see, so the prerequisite is a single read surface across PSS, NDC connections, and direct channels before any demand balancing logic runs.

Catch: airlines often discover mid build that their PSS exposes availability in pull mode only, which means your "real time" view is actually a cache, and your allocation decisions are always slightly stale.

→ Automate inventory management and seat allocation

Alex Brooker, Host of the Travel Tech Podcast, Founder of Airside Labs