Three Things the Aviation Industry Needs to Understand About AI and Connected Data

The conversation about AI and data quality is necessary, but quality alone isn't enough.

Most of the concern around the implementation of AI in aviation has focused on data quality: whether inputs are accurate, up to date, and free from bias.

That debate is worth having. But in our latest report, CEO Filip Filipov makes the case that there is a second failure mode, one that sits a layer beneath quality and is less well understood. Even clean data fails when it lives in silos.

1. Accurate data and complete data are not the same thing

A GPS that maps every road with perfect accuracy still gives you a bad route if half the streets are missing. The same logic applies to AI in aviation.

An AI travel planner recommends a connection via Istanbul with a 70-minute transfer. The schedule data is correct. But the system has no access to minimum connection time rules, no visibility of visa transit requirements, and no terminal transfer intelligence. A traditional booking system would have filtered the itinerary out. The AI presents it as optimal. The passenger finds out at the gate.

On the operations side: an AI disruption tool detects a storm and begins rerouting passengers, but lacks real-time crew availability data. Its recovery plan creates a secondary disruption that a human controller with the full picture would have caught immediately.

In both cases, the data used was clean and trustworthy, but without looping in further data to add context it was useless, or unhelpful.

What this means for you: Before deploying any AI system, map every data source it relies on and explicitly identify what it cannot see. Gaps in data coverage are design risks, not just technical debt. Building AI on partial inputs and then tuning the model is working on the wrong problem.

2. Aviation's structural fragmentation is getting more costly

Aviation generates staggering data volumes: According to Oliver Wyman, the global fleet will produce around 100 million terabytes in 2026 alone. The problem is not a lack of data, but that the data is fragmented.

Airlines, airports, air traffic control, ground handling, maintenance, and regulators each operate their own systems, on their own update cycles, in their own formats. According to IATA, around two-thirds of airlines report operational silos as a persistent challenge, and close to half of all flight delays trace back to poor coordination between functions.

AI travel agents need comprehensive, real-time data across schedules, availability, connection rules, and pricing. With up to 25% of all airline fares dynamically priced at the end of 2024, a fare an AI agent sees at 10:00 may be obsolete by 11:00.

What this means for you: As AI moves from generating recommendations to executing decisions, the cost of incomplete inputs compounds. Organisations treating data integration as a long-term IT project while prioritising AI deployment are sequencing this backwards.

3. Connected data is the difference between AI that recommends and AI that works

Lufthansa partnered with zeroG to build an AI tool that surfaces every cabin defect as a visual seat map on crew tablets, linked to the passenger in each affected seat. The hardest part was not building the model. It was reconciling data from maintenance systems riddled with contradictions across fragmented sources.

Outside aviation, Starbucks unified transaction history, weather data, event schedules, and loyalty profiles into a single intelligence layer, attributing $2.1 billion in incremental revenue to AI-driven personalisation between 2023 and 2024. No single data source delivered that outcome, the connection between them did.

What this means for you: The question to ask of any AI initiative is not "how good is the model?" It is "how complete is the picture the model is working from?" In aviation, what your AI cannot see can ground your operation.