Partial in,
partial out

Generating vast amounts of data is only half the challenge. The real problem is what happens to that data once it exists.

Aviation is an industry built on split-second coordination between at least six major stakeholder groups: airlines, airports, air traffic control, ground handling, maintenance teams, and regulators. Each operates its own systems, on its own update cycles, often with its own data formats. Schedules update in one system but lag in another. A gate change is communicated to passengers, but is not necessarily synced with the baggage handling system. A delay in one airport's database takes hours to propagate to partner airlines.

The result is structural fragmentation, and it shows. According to IATA, about two-thirds of airlines struggle with operational silos, and nearly half of all flight delays (47%) stem from poor coordination between functions like ground handling, maintenance, and flight ops.

These are data integration failures. And they don't stop at the hangar door.

The same fragmentation that plagues aviation's operational backend also runs through its commercial front end, where travellers search, compare, and book flights. Different systems, different formats, different update cycles. The pattern repeats, just with different players.

In a 2025 global survey by Sabre, 91% of travel agencies reported using four or more different booking systems. The three dominant distribution models (GDS, NDC, and direct airline APIs) each operate with their own data formats, pricing structures, and content frameworks. Even within the NDC standard, different airlines interpret and implement it inconsistently, making seamless integration an ongoing challenge.

Here is where the stakes have risen sharply in just the past two years.

The commercial front end of aviation is no longer a collection of search boxes and filter menus. It has become a conversational interface.

AI is now a core part of how millions of people decide where to go, when to fly, and how much to pay.

This shift has profound implications for the underlying data layer.

AI agents planning (and soon booking) trips need access to comprehensive, real-time, standardised data across multiple domains such as schedules, availability, connection rules, and pricing. Pricing in particular is mission-critical. According to our own research into the current state of airline pricing, up to 25% of all airline fares were dynamically priced at the end of 2024, with most airlines stating the goal of eliminating static pricing entirely over time. In a world where fares are generated dynamically in response to demand signals, competitive moves, and individual traveller profiles, an AI agent working with stale or partial pricing data is fundamentally broken.

This is a gap the current AI agent debate in the airline context still largely misses. Much of the conversation focuses on model capabilities, reasoning, and user experience. Far less attention is paid to whether the data these agents are drawing on is comprehensive enough to support the decisions they are making on behalf of travellers. Dynamic pricing makes this gap very consequential: the fare an AI agent sees at 10:00 may be obsolete by 11:00.

This brings us back to the core thesis of this analysis: in aviation, having accurate data from one source isn't enough. You need accurate data from many sources, integrated coherently, updated in real time, and accessible in formats that AI systems can actually consume.

This is precisely why the breadth of an aviation data provider's offering matters as much as the depth of any single dataset. The fewer the gaps, the less an AI system has to guess, aka hallucinate.

Already today, incomplete data integration causes painful, entirely avoidable failures.

Consider a real-world example from Lufthansa Group that illustrates this with striking clarity.

Before every flight, cabin crews receive a hold item list, which is a document detailing known defects on the aircraft. At Lufthansa, this list could run to forty pages of legacy telex shorthand, which is often delivered just a few minutes before boarding. No time to read it all, let alone cross-reference entries.

So the passenger in 7D sits down, discovers a dead screen, and presses the call button. The crew member learns about the defect from the passenger, not the other way around. That reversal, where the customer knows more than the crew does, is when the service recovery window closes. A glance at the booking data would have shown a top-tier frequent flyer. A proactive gesture before pushback would have cost almost nothing.

Now, the irony is: the data existed. All of it: the maintenance record, the partial repair status, the seat assignment, the loyalty tier. It just lived in separate systems and never reached the right person in a usable form.

Fortunately, Lufthansa partnered with zeroG to build an AI-driven solution that surfaces every known cabin defect on the crew's tablets as a visual seat map, links each affected seat to the passenger sitting in it, and puts compensation options one tap away. No forty-page printout. No buried contradictions.

And here's the telling detail: the hardest part wasn't building the AI model. It was curating the data, reconciling entries from maintenance systems riddled with contradictions, and building logic to resolve conflicting records across fragmented sources.

This is the data quantity problem in miniature.

Scale this pattern across an airline's entire operation (think crew scheduling, turnaround coordination, disruption management, passenger rebooking) and the cost of incomplete data integration becomes enormous.

This pattern is not unique to the airline industry.

Across sectors, the companies that have gained the most from data-driven intelligence are those that have unified fragmented sources.

Target is an example from the retail space that perfectly exemplifies this, and Starbucks took the concept even further. Here we explore what each brand achieved and why it matters.

In both cases, the lesson is the same.

• Weather data alone doesn't drive a promotion.
• A single purchase record alone doesn't predict a pregnancy.
• Loyalty data alone doesn't tell you when a customer might prefer an iced frappuccino over their usual flat white.

But unified, those signals become intelligence that no single source could deliver.

Aviation faces exactly this challenge, only with higher stakes, tighter margins, and zero tolerance for error.

The pattern across every example in this article is the same: the data often exists, but it doesn't connect.

Stitching together aviation's fragmented data landscape is extraordinarily difficult.

• Legacy infrastructure makes integration slow.
• Competitive sensitivities make data sharing complicated.
• And the sheer volume makes manual reconciliation impossible.

This is precisely the environment where AI is supposed to help.

But AI can't integrate what it can't access.

To ensure that AI can deliver on its promise of transforming the airline business, comprehensive, well-connected data is just as essential as accurate data. Data breadth alone isn't enough. What matters is what happens when those datasets are connected – when schedules, real-time flight status, connection intelligence, demand signals, and competitive pricing stop living in silos and start informing each other.

That's where data becomes intelligence.

This is the role OAG was built for. We provide aviation's broadest spectrum of datasets, and the intelligence layer that connects them. By turning fragmented inputs into coherent, AI-ready intelligence, OAG enables airlines, airports, and travel-tech partners to build systems that see the full picture and, most importantly, make decisions accordingly.

Fewer blind spots. Fewer gaps between what an AI system knows and what it needs to know. Because in aviation, what your AI doesn't know can ground your operation.