This insight summarises How airlines are solving disruption in the AI era, by Matthew Walker, head of research and marketing at Plan3, published on PhocusWire on 13 October 2026.
Ready to act, within limits. At Grounded 2026 in Reykjavík, leaders from more than 35 airlines representing over 40% of global passenger volumes were asked how much disruption decision-making they would hand to AI if the technology worked tomorrow. 74% would let AI take action of some kind:
- 39% would let AI act within predefined rules.
- 28% would let it act autonomously and escalate exceptions.
- 26% would restrict it to recommendations.
- Just 7% would consider full autonomy for most passengers.
The barriers are internal. 41% named systems and data as the main obstacle, 27% organisational readiness and 18% trust and governance. Only 10% chose the business case and 4% regulation.
Speed is the expectation gap. Asked where passenger expectations most outrun what airlines deliver, 41% chose speed of information, ahead of personal relevance (27%) and quality of options (25%). Once technology can understand a problem in seconds, passengers will wonder why resolving it still takes hours.
The 1% problem. During the 2025 Thanksgiving period, 31 million US passengers flew with a 99% completion rate, which still left 310,000 people disrupted. At scale, edge cases are populations. Errors also differ in weight: a poor hotel choice is not the same as a journey the traveler cannot legally complete or a separated family. An automated system can repeat one flawed decision thousands of times before anyone notices, so accuracy alone says little about risk once AI is allowed to act.
Trust compounds. Walker expects AI to advise, then recommend, then act within boundaries, extending one use case at a time. People do not disappear: they set policy, supervise systems and step in where accountability matters.
Why this matters for hotels. Hotels face the same choices as they automate guest recovery, rebooking and service. Decide which decisions AI may take, which failures are tolerable and where a person stays in the loop, then expand autonomy as each use case earns trust.