Airlines Can Now Ask AI Why Flights Run Late

AvioBook, an airline software company owned by the French tech group Thales, has built an AI assistant on top of Amazon's Bedrock AgentCore — a service for creating AI "agents," meaning programs that can look things up and take steps on their own. The tool, called Connected Analytics, lets airline managers and dispatchers ask questions in everyday language about why a flight was late, instead of digging through records by hand.

The problem will feel familiar to anyone who has tracked a delayed flight. Airlines coordinate each turnaround — the window between a plane landing and leaving again — in a live chat room per flight. That leaves a rich trail of timestamps, alerts and messages, but it is really an archive: useful only if someone has the patience to scroll back through weeks of history. Worse, the official "delay codes" are usually filled in by one rushed person under time pressure. A code can be filled in correctly by the rules and still hide what actually went wrong.

That matters because the gate is where airlines quietly win or lose. Time lost at the gate costs roughly $20 a minute through fuel, crew pay, gate fees and missed connections. For a mid-size carrier running 200 flights a day, shaving just two minutes off the average turnaround is worth about $240,000 a month. Two AI agents — one tuned for managers, one for dispatchers — do the digging, pull the relevant flight events, check them against what was logged in the chat room, and return a direct answer with its evidence.

The catch: this is still a proof of concept, not something you can buy today. Airline operations are messy, and the company itself notes that crew accounts and paper records may not hold up under scrutiny — an AI reading them inherits those gaps. On the plus side, the chat data stays inside each airline's own systems, under the airline's control, which addresses privacy worries. But a confident answer is not automatically a correct one.