Airline chatbots for disruptions, rebooking, and baggage

An airline's contact volume is mostly flat, then a storm rolls through a hub and it is not. Cancellations cascade, every affected passenger reaches for the app or the phone at once, and a queue that handled 2,000 conversations a day is suddenly looking at tens of thousands in an afternoon. The disruption is the test. Everything an airline chatbot does on a normal Tuesday matters far less than what it does during irregular operations.

This guide is about that test. It covers what airline chatbots can safely automate across the three highest-pressure areas (disruptions, rebooking, and baggage), where a human still has to step in, and how to build for the volume spike instead of the average day.

TL;DR

• The hard problem is the surge during irregular operations (IRROPS), when a hub goes down and contact volume multiplies. On a normal day, volume is the easy part.
• Rebooking is the highest-value automation: passengers want to self-serve a new flight instantly, and airlines have shown it works at scale.
• Baggage queries are high-volume and largely answerable from tracking data, which makes them a strong automation target.
• Keep complex itineraries, disability assistance, distressed passengers, and compensation disputes with trained agents.
• Buy and price for the spike, because a chatbot that is affordable at 3,000 daily conversations can break your budget at 40,000.

The disruption is the whole game

Airlines do not get judged on the smooth days. They get judged on the bad ones, when a winter storm or an air-traffic-control issue cancels a wave of flights and every passenger needs help at the same moment.

That surge is brutal for traditional support. Phone lines jam, hold times stretch past an hour, and agents burn out working through a queue that keeps growing faster than they can clear it. The passengers are stressed, often stranded, and the experience defines how they feel about the airline long after the weather clears.

This is exactly where automation earns its keep, because the surge is made largely of the same few requests repeated at scale: what happened to my flight, what are my options, rebook me, where is my bag. An airline chatbot that can field those instantly, for everyone at once, turns the worst day of the operation into something survivable. The constraint is that it has to hold up under load, which is a real engineering and commercial question covered below.

Rebooking: the automation that proves the case

Rebooking during a disruption is the clearest win, and it is no longer theoretical. American Airlines built a generative-AI rebooking tool that lets disrupted passengers self-serve a new flight, and aviation analyst OAG reported in August 2025 that it had already helped more than 200,000 travelers during severe East Coast storms.

What makes rebooking a good automation target is that it is high-stakes for the passenger but well-structured underneath. The system needs to detect the disruption, pull the right fare and routing options, apply the correct waiver or policy, present the passenger personalized alternatives, and execute the change in the passenger service system. Each of those steps is defined. There is a right answer, and the AI's job is to find it and act on it, which is very different from improvising.

The passenger experience is the other half. Instead of refreshing the app and waiting on hold, a passenger gets alternative flights within seconds and picks one. That speed is worth a lot during a disruption, when the difference between rebooking in thirty seconds and ninety minutes can be whether they make it home that night. The same conversational pattern shows up beyond airlines, in AI booking assistants for travel that handle changes, cancellations, and refunds across hotels and trips.

Where rebooking automation should hand off is the messy itinerary: multi-carrier tickets, interline agreements, complex international routings, and any case involving a duty of care obligation like overnight accommodation. Those carry rules and judgment calls that an agent should own. A system that hands off when it is unsure, the way our Agent 5 model does, keeps the AI on the rebookings it can do correctly and routes the rest before it guesses.

Baggage: high volume, mostly answerable

Baggage is the other natural target, because the volume is high and most of it is informational. Where is my bag, when will it arrive, how do I file a claim, what is the status of my delayed bag. These are answerable from tracking and policy data.

The scale of the problem is real. In SITA's 2024 Baggage IT Insights, the industry mishandled 33.4 million bags, at a rate of 6.3 per 1,000 passengers, costing an estimated $5 billion. Delayed bags made up 74% of those mishandlings. A delayed bag is the best case for automation, because the bag exists and is moving, so the passenger mostly needs status and a delivery plan, which a chatbot connected to the baggage tracking system can provide instantly.

The encouraging trend is that the mishandling rate is falling (down from 6.9 per 1,000 in 2023) as airlines invest in tracking technology. Better tracking data is precisely what makes baggage queries automatable, because the AI can only answer "where is my bag" well if the system actually knows. The automation rides on the tracking investment.

Baggage automation should escalate the cases that involve money or loss: a lost bag, a damaged bag, a high-value claim, anything contested. Those need a human and a process. The chatbot's job is to resolve the status questions and start the claim, then route the resolution to a person.

What to keep with human agents

Automating disruptions does not mean automating everything inside them. Several categories should stay with trained agents, always.

Distressed and vulnerable passengers come first. Someone stranded overnight with a child, a missed funeral, a medical situation: these are moments that need a human, and an airline that routes them to a bot will be remembered for it. The AI should detect distress signals and escalate fast.

Disability and special assistance is the second. Accessibility needs are regulated and individual, and they require a person who can take responsibility for getting it right.

Compensation and disputes are the third. Whether a passenger is owed care or compensation often involves regulation (in the EU, for instance, established passenger-rights rules) and judgment about edge cases. An AI can explain the policy and gather the facts. A human should make the call on a contested claim.

The principle underneath all three is the same one that works across customer service: let AI handle the high-volume, well-structured requests, and protect the human moments where empathy or accountability is the actual product.

Build for the spike, and price for it

Here is the part teams underestimate. The volume profile of airline support is spiky, and both the architecture and the contract have to assume the spike.

A chatbot that handles 2,000 conversations a day comfortably can face 40,000 during a major IRROPS event. If the system degrades under that load, it fails at the exact moment it was supposed to help, which is worse than not having it. Latency, concurrency, and graceful degradation are the engineering questions to press a vendor on. Ask specifically what happens at 10x and 20x normal volume.

The commercial model matters just as much, and it is easy to miss. Per-message or per-conversation pricing turns a disruption into a runaway bill, because the spike you most need to cover is the spike that costs the most. Per-resolution pricing aligns better, since you pay for outcomes rather than for the raw flood of messages a storm produces. Our own pricing works this way, with human escalations free, so a bad-weather day does not punish you for routing distressed passengers to people. Whatever you choose, model the cost of a worst-case IRROPS week before you sign. An average week will understate what a storm actually costs you. Whether a vendor can hold up at airline scale is a fair thing to interrogate directly; pressure-test whether a startup vendor can handle enterprise-scale support before you commit.

The other architectural note is integration depth. Rebooking automation is only as good as its connection to the passenger service system and the disruption-management tools. A chatbot that can talk about rebooking but cannot execute it in the PSS just moves the work to an agent. The value lives in the workflows that actually complete the action.

Airlines sit inside the broader travel and hospitality picture, where the same split between routine automation and human moments plays out across the whole guest journey in hospitality. Airlines have a structural advantage here that most industries lack: their busiest, most painful support moments are also their most repetitive. A thousand passengers off a cancelled flight need nearly the same thing, which is the rare case where scale and similarity line up. That is the opening. An airline chatbot that resolves those at the speed of the surge, while routing the stranded family and the disability request to a person who can actually help, turns the worst day of the operation into the day the technology earns back everything it cost.