What Percentage of Complex, Multi-Leg Flight Disruptions Can AI Rebook Without a Human? | ITILITE Analysis

An ITILITE analysis of public data and independent research. Sources: US DOT and the Bureau of Transportation Statistics, IATA, GBTA, Skift and McKinsey, and peer-reviewed AI benchmarks. Last updated September 2026.

Every travel-tech vendor now promises AI that monitors your trips and rebooks you when a flight falls apart. So it is fair to ask the specific version of that promise: when the disruption is a complex, multi-leg, multi-carrier itinerary, what share of those does AI actually fix on its own, with no human involved?

The honest answer starts with a number that does not exist. No credible, independent source publishes a "percentage of complex multi-leg disruptions AI rebooks fully autonomously." Not the airlines, not the analysts, not the vendors making the claim. And that absence, once you look at why it exists, is the answer.

The short answer: AI reliably handles simple, same-carrier reroutes today. Complex multi-leg and interline disruptions still hand off to a human, because the machine lacks both the authority and the information to resolve them alone. Any vendor quoting you a clean autonomous-rebooking percentage for complex trips is quoting marketing, not a measured result.

Why the honest answer starts with a missing number

When a specific, commercially useful statistic is missing across an entire industry, that usually means it is hard to measure or unflattering to report. Both are true here. What exists instead splits into four groups, and none of them is the headline number.

Vendor claims describe capability without auditing it: assistants that "monitor and rebook automatically," with no published autonomous success rate and nothing isolated to complex cases. Airline systems that genuinely auto-rebook do so on simple, same-carrier misconnects. Independent academic benchmarks that measure the closest analog, autonomously building a feasible multi-constraint trip, come back very low. And industry research shows travel teams want automated rebooking but report that AI has barely touched the hard cases. Put together, they point one direction.

How often complex disruptions actually happen

Disruption is frequent and getting worse. About 25% of US domestic flights failed to arrive on time in the second quarter of 2025, the worst performance since 2014. Across full-year 2024, roughly 20% of flights were delayed and about 1.4% cancelled, on more than seven million scheduled flights.

There is a telling gap in that data. The Bureau of Transportation Statistics does not publish missed-connection rates, the disruptions that turn a clean itinerary into a complex rebooking problem. The scale of the hardest cases is not even measured publicly, which is part of why the autonomy question has no clean answer. Against a $1.71 trillion global business travel market in 2026, that unmeasured middle is expensive.

Where automation works today: simple, same-carrier reroutes

Automated rebooking is real, but narrow. American Airlines runs an automated reaccommodation system that rebooks passengers it predicts will misconnect, using its own available inventory. It works on a single carrier's own network, and even there it publishes no success rate. In 2026 it drew a backlash for acting without human oversight, rebooking travelers who could still have made their original flight.

That is the current frontier of autonomous rebooking: one airline, one network, one simple failure mode, and it still generates complaints when the human is removed. The corporate travel version, an AI-driven travel program, handles the same easy cases well, such as a delayed domestic leg on a single carrier with obvious alternatives. The difficulty begins the moment the itinerary stops being simple.

Why complex multi-leg is the wall

A machine can rebook autonomously only where it has the authority and the inventory to act. On a multi-carrier or interline itinerary, that authority is fragmented. Cross-carrier reaccommodation depends on bilateral interline agreements, ticket reissue and endorsement, and back-end settlement between airlines, a multi-party process governed by IATA resolutions and still immature under modern airline retailing.

It gets harder. The old "Rule 240" that once forced airlines to rebook you on a competitor is effectively defunct since deregulation, replaced by a voluntary, carrier-by-carrier patchwork. Only some US carriers commit to rebooking you on another airline at all.

Layer on fare rules, ancillaries already paid for, corporate policy exceptions, and for international trips visa and entry constraints, and the complex case is one no autonomous agent can close on its own. It has neither the cross-carrier authority nor the full picture. That is the structural reason the number is missing, and it is why these trips, the kind covered in the multi-city business trip planning guide, still land on a human's desk.

What independent benchmarks actually measure

Since no one measures autonomous rebooking directly, the closest honest proxy is how well AI builds a feasible trip under many constraints, which is the same skill autonomous rebooking would need. Peer-reviewed benchmarks are blunt about it.

Two caveats keep this honest. These benchmarks test general-purpose models building trips, not the constrained, API-backed tools a production travel platform uses, so a narrow real-world system can do better on a narrow task. And they measure planning, not disruption rebooking specifically, so treat them as a ceiling, not a direct readout. Even read generously, the signal is the same: autonomous multi-constraint trip-building tops out somewhere between roughly 1 and 22%, which is nowhere near the reliability a traveler stranded overnight needs.

What travel teams want versus what AI delivers

Demand is not the constraint. Adoption and trust in the hard cases are. Industry research shows travel teams want automated disruption handling but have not seen it work on the complex cases yet.

The demand-versus-impact gap is stark: 89% of travel buyers want automated rebooking, yet 58% say AI has changed little in their program so far. The consensus across analysts is the same practical split: automate the clear wins now, and keep humans on the complex decisions, because "when things go poorly, humans are needed". Agentic AI is coming to execution, but the human touch is still called essential for the hard cases.

The honest answer, and what it means for buyers

So, the percentage. There is no defensible single number, and anyone who gives you one for complex multi-leg disruptions is selling. The honest, evidence-based answer is a shape, not a statistic: high autonomy on simple same-carrier reroutes, falling toward zero as an itinerary adds carriers, interline segments, fare complexity, and international constraints. The independent ceiling on autonomous multi-constraint trip-building sits in the low tens of percent at best, and disruption rebooking is harder than planning because it happens live, under pressure, with fragmented authority.

For a buyer evaluating an "AI rebooks your trips" pitch, the useful question is not "what percentage." It is "what happens on the complex case, and who is accountable when the agent cannot act." The right model, and the one the data supports, is AI for the simple and fast, humans for the complex and consequential.

How ITILITE approaches disruptions

ITILITE builds around exactly that split, and it is worth being plain about it: ITILITE does not claim to autonomously rebook complex disruptions, because no one credibly can. AI does the parts it is good at, monitoring trips, flagging risk, and handling simple changes, while a real person owns the hard reroute.

That means the AI layer surfaces alerts and answers questions (its conversational assistant handles travel-data queries rather than pretending to be an autonomous agent), and when a complex disruption hits, the traveler reaches a human fast: ITILITE runs live, human-powered support with in-house travel specialists, not a bot that quietly rebooks you off a flight you could still make. The role of finance and AI in travel is to automate the routine and free the humans for the exceptions, and the whole model runs on one corporate travel booking platform.

The bottom line

The question "what percentage of complex multi-leg disruptions can AI rebook without a human" has no honest numeric answer in 2026, and the reason is the answer. The authority is fragmented across carriers, the data on the hardest disruptions is not even published, and the best independent measure of autonomous trip-building is a low-single to low-double digit success rate. AI has earned the simple reroutes. The complex ones still belong to people, augmented by AI rather than replaced by it. Buyers who internalize that will evaluate travel AI on how well it hands off, not on a percentage no one can back up.

Methodology and sources

This is an ITILITE analysis of public data and independent research, not a first-party study. ITILITE does not publish an autonomous-rebooking success rate, and this piece does not estimate one; where a clean statistic does not exist, that is stated rather than filled with a number. Disruption rates are from the US Bureau of Transportation Statistics and the Department of Transportation. Interline and reaccommodation mechanics are from IATA and the DOT dashboard. Autonomy figures are peer-reviewed benchmarks (TravelPlanner, ICML 2024; TripTailor, 2025) and are labeled as planning proxies, not direct measures of rebooking. Adoption and demand figures are from GBTA, Skift, McKinsey, and Phocuswright. Competitor AI-resolution claims were reviewed and excluded as marketing.