The AI Revolution
at 10,000 Metres

Commercial aviation safety is, statistically, extraordinary. IATA’s 2024 Annual Safety Report recorded seven fatal accidents across 40.6 million flights — an accident rate of 1.13 per million flights (better than the five-year average of 1.25), and a fatality risk of just 0.06 (IATA Safety Report release). Airlines on the IOSA registry — which includes all IATA member airlines — had a rate of 0.92 accidents per million flights.

The decade-on-decade trend is straightforward: one accident per 810,000 flights on the 2020–2024 average, compared to one per 456,000 flights in 2011–2015. This is the industry AI has to work alongside — not to make flying safer than it already is, but to maintain that safety margin while everything else changes around it.

Market estimates for AI in aviation vary widely depending on what’s counted, but most analyst firms cluster around $6–7 billion in 2024 with projections in the $20–45 billion range by 2033–2035 (MarketsandMarkets · Transparency MR). The spread reflects different definitions of “aviation AI” rather than disagreement about direction.

Jet fuel is one of the largest single line items in airline operating cost — exactly how large depends on the year, the carrier, and the fuel price curve. Even small percentage savings translate into meaningful annual P&L. The major airline groups all publish digital and AI initiatives in their annual reports targeting two areas:

Route and trajectory optimisation — using real-time wind, weather, congestion and weight data to recompute the most efficient flight path in flight, not just pre-departure. The technique is well-established; the savings claimed by individual carriers are within the noise band of single-digit percentages.

Predictive maintenance — using vibration, temperature, and operational sensor streams from engines and airframes to anticipate component failure before it grounds an aircraft. This is the highest-ROI category in most majors’ digital programmes. Specific impact figures vary by carrier and are usually presented in annual reports without independent audit.

I’ve deliberately not quoted single-airline percentage figures here that I couldn’t trace to a primary disclosure. The honest summary: the savings are real and the direction is consistent, but most of the eye-catching numbers floating around the industry press are from vendor pitches, not audited carrier disclosures.

A modern airline ticket and journey involves several layers of ML even before you board. Dynamic pricing sets fares based on demand, competitor pricing, and historical buying behaviour. Seat-selection screens simultaneously balance aircraft weight distribution, revenue optimisation, and group-traveller heuristics. Boarding-group assignment increasingly uses ML to reduce turnaround time. Delay-prediction notifications in airline apps use ground-handling, ATC, weather, and inbound-aircraft data to give passengers earlier warnings than the gate agent can.

None of this is one big “AI moment.” It is twenty small optimisations stacked on top of each other, each contributing a small basis-point improvement to operations or revenue, none of them visible to the passenger as “the AI.”

The highest-impact AI in aviation is the kind nobody markets: continuous monitoring of thousands of sensors per aircraft, looking for the subtle drift patterns that precede a hard failure. ML applied to fleet-wide sensor data can detect issues that single-aircraft thresholds would miss — because the model has seen the same component degrade across hundreds of identical airframes.

The same approach is being applied to turbulence prediction (more granular than current radar), runway-condition modelling for crosswind landings, and weather-system path forecasting for both routing and crew-fatigue planning. The pattern is the same as on the operations side: small basis-point improvements stacked.

Aviation AI is not flashy. There is no autonomous-flight headline coming any time soon — the regulatory bar for that is correctly very high. What you’ll experience over the next five years instead: more accurate delay notifications, fewer flight cancellations from unexpected maintenance, slightly faster boarding and turnarounds, and ticket pricing that is more responsive to actual demand.

None of these are individually revolutionary. Taken together, they are the practical face of AI in a safety-critical industry: incremental, instrumented, regulated, and aimed at the operational margin rather than at the passenger imagination.