An airline virtual agent that rebooks, refunds and answers passengers in the chat, inside the limits your people sign
Disrupted passengers finished in the chat, not the phone queue, inside limits people set.
See one case, screen by screen ↓An airline virtual agent is conversational AI for airlines that talks with passengers in the app, on the web, by text and email and does the work: it rebooks them, reissues the ticket, refunds what is owed and answers trip questions. Airline Virtual Agent does this with eight AI agents inside limits your people write and sign; a second agent checks every ticket or money action before it runs, and anything outside the limits goes to a person with a ready-to-approve brief.
How to choose AI for airlines: a buyer's guide to agentic AI solutions →
Updated
When a flight cancels, every passenger on it needs something within the hour: a new flight that actually works, a seat, the refund they are owed, a meal voucher, a hotel. The chatbot on the app can answer a question about bag fees and deep-link to “manage trip”, but it cannot finish the job, so on an irregular operations (IROPS) day the contacts go to the phones and the airport counter. A care agent works across reservations, ticketing, the seat map, the refund screen, a voucher tool and the CRM, and the passenger tells the story again at every transfer.
When a bot does act, the airline is bound by what it says. A 2024 tribunal decision held an airline to what its chatbot told a passenger, and rejected the argument that the chatbot was a separate legal entity. The DOT automatic refund rule, 14 CFR Part 260, requires the refund right to be stated before any travel credit is offered, and returns seat and bag fees for services not provided. Answers copied into a bot's FAQ drift when a fee page changes, and a translated reply can say “credit” where the passenger reads “refund”.
Estimated minutes for one passenger on a canceled flight, through the phone queue and in the chat. The chat column follows the sample conversation in the working solution.
Eight specialist agents talk, rebook, refund and check each other; people set the limits and decide the rest.
Thursday, October 15, 2026, Chicago O'Hare. Corvane Airways CV 734 to Orlando, an A321neo with 186 passengers booked for 5:10 pm, is canceled at 3:52 pm for a maintenance problem. Here is what the airline virtual agent and the people beside it do next, screen by screen, in the working solution.
The operations feed reports CV 734 canceled — maintenance, controllable. At 3:58 pm the Trip Agent opens a conversation with all 186 passengers, 121 in the app, 52 by text and 13 by email, from first-message template CX-CNL-07 v12, approved by Dana Whitfield. By 4:14 pm, 141 have finished in the chat — 124 rebooked, 96 on CV 736 tonight and 28 on CV 1102 tomorrow, and 17 refunded, $5,862.40 — 31 are with a person and 14 have not answered yet.
“I’m Corvane’s virtual assistant, an AI. I’m sorry: CV 734 to Orlando at 5:10 pm is canceled. You can have a full refund to the card you paid with, or I can move you to another flight at no cost. Reply here — no need to call.”
Elena paid $214.60 with her Visa ending 2218, plus $39.00 for extra-legroom seat 12A. Her conversation opens with the AI disclosure and a full refund of $214.60 to that Visa, or another flight at no cost. She answers in her own words, and the Rebooking Planner reads “Orlando tonight”: one Corvane flight, CV 736 at 9:40 pm, landing 1:22 am, with aisle seat 14C held for 10 minutes. CV 1102 tomorrow at 7:05 am and the refund stay on screen as options. Her trip card sits at the top of the chat, and she sees it too.
The partner flight at 8:15 pm is not offered — partner airlines need a person (L-RB2).
“Yes 736 please. 14C is fine.” Before the ticket moves, the Policy Guard checks the move on its own: same cabin, inside 48 hours, no fare difference — inside L-RB1, 99%. The Rebooking Planner reissues ticket 771 2391467318 for CV 736, seat 14C, and sends the boarding pass. The Fare and Refund Agent returns the $39.00 seat fee to her Visa (RF-26-80117, by Oct 26) and issues a $15 meal voucher, MV-734-0412, for a wait over 3 hours after a controllable cancellation. Each line on her trip card cites the rule that allowed it.
“Your extra-legroom seat isn’t available on that flight, so $39.00 goes back to your Visa.”
“I land at 1:22 and my rental car desk closes at midnight. Can you cover a hotel near the Orlando airport?” A hotel at the destination is not in the Customer Service Plan, and limit L-CR4 says a person decides. The Trip Agent tells Elena a colleague is coming and stays in the conversation. The Hand-off Writer writes the brief — her words, what was done, why it stopped — and proposes the MCO airport hotel, 1 night, partner rate $129.00, reason GW-02. People approved 58 of 63 like this last week.
“I’ve passed them everything, so you won’t need to repeat yourself. They usually reply within 10 minutes, and I’ll stay right here.”
Dana runs the digital care floor at the ORD hub. Today: 18,420 conversations, 412 live and 3,104 in 11 other languages; 80.0% finished in the chat, 0 outside limits, and the Policy Guard stopped 41 actions; 38 waiting for a person, the oldest 6 minutes against a 10-minute first-reply target. Elena's hotel is first in “Needs you”, 2 minutes of 10, ahead of a hand-off that mentions a DOT complaint and a limit change ready to sign.
Quality Reviewer: “People approved 58 of 63 destination-hotel requests last week — a limit is proposed and rehearsed.”
Before she decides, Dana asks the agents inside Elena's conversation, where they answer from that conversation only. The Rebooking Planner answers with its sources. She could also ask what Elena paid and what she has got back, have the agents offer CV 1102 instead, or add a note for the gate — every message the agents draft waits for her approval.
“The 8:15 pm partner flight connects in Atlanta and lands at 12:58 am — only 24 minutes earlier than CV 736 — and partner airlines need a person under L-RB2.”
Dana keeps reason code GW-02 — controllable, arrival after midnight — and approves the reply the Hand-off Writer drafted. Voucher HV-26-3318 goes on Elena's trip card, with the shuttle running from the terminal until 2:30 am. Elena thanks her in the chat and rates the conversation 5 of 5. She never called.
“Approved by Dana Whitfield · GW-02 · MCO airport hotel, 1 night, $129.00 · voucher HV-26-3318 · Elena rated 5 of 5.”
Fifteen limits say what the agent may do on its own, signed by Dana Whitfield and Celia Hartman. L-CR4, hotel at the destination, is never on its own: a care agent decides. In the last 7 days people approved 58 of 63 destination-hotel hand-offs after a controllable cancellation landing after midnight, so the Quality Reviewer proposes a change: up to $140 on its own when arrival is after 12:30 am and the cause is controllable.
The rehearsal replays the 2,140 conversations from Oct 8–14 with a meal, hotel or transport request; the Trip Agent and Policy Guard decide each again under the proposed limit, blind to what happened. Result: 58 would finish in the chat, 5 still go to a person, 63 of 63 match people's decisions, and the cost is $7,482 for the week. Dana approves; Celia Hartman, Director of Digital Care, countersigns.
“Nothing changes for passengers until both have signed.”
The audit trail holds every agent and person action with its sources — tamper-evident and kept 7 years. If Elena's trip ever becomes a DOT complaint, Nadia exports her evidence pack: the full conversation, every trip-card change, each action with the sources the agents read, the Policy Guard checks, and Dana's decision with its reason code, sealed with the audit hash as a 9-page PDF.
“Every agent and person action with what it read · tamper-evident · kept 7 years.”
The same disruption day, seen by five of the people at Corvane Airways who carry it — what the work looked like, and what working with the agents looks like now.
Eight specialist agents talk, rebook, refund and check each other; people set the limits and decide the rest.
Talks with passengers in their language in the app, on the web, by text and email; understands what they need, offers options and acts inside the limits. When a request is outside them, it tells the passenger who will help and when, and stays in the conversation.
Finds Corvane flights, seats and connections that meet the passenger's stated need and the fare rules — arrive tonight, an aisle, travel together — holds the seat for 10 minutes and reissues the ticket after the check.
Prices refunds of the unused fare with all taxes and fees, returns fees for services not provided, and issues vouchers inside the care limits, under Part 260 and the fare rules. Payments are staged to the original form of payment.
Decides again, from the sources and blind to the Trip Agent, whether every action that moves a ticket or money is inside a limit and the rules — Parts 259, 260 and 382 — before it runs.
Writes the brief a person can approve in one step: the passenger's words, what was done, the limit that stopped the agent, one proposed action with its cost and reason code. Picks the queue and sets the first-reply clock by tier.
Back-translates legally meaningful words — refund, travel credit, voucher, compensation — in every reply not in English, compares them with the protected-terms glossary and flags a mismatch.
Scores finished conversations for policy accuracy, citation, tone and missed hand-offs. When people approve the same hand-off again and again, it proposes a limit with the evidence.
Checks the published policy pages the agent answers from every hour; when a page changes, finds every answer since the change that quoted the old text and drafts the corrected answer.
Sign the limits, decide hand-offs. The agents propose; a named person decides.
Every working screen has a line to ask the agents about what is on it. Answers name the agent and cite their sources, and an instruction becomes a proposal you approve.
Why did Elena Brooks go to a person?
She asked for a hotel in Orlando. A hotel at the destination is outside the limits — L-CR4 says a person decides. The Hand-off Writer proposed the MCO airport hotel at $129; people approved 58 of 63 like this last week. I opened her conversation — the decision is yours.
What did Elena pay and what has she got back?
Elena paid $214.60 for the ticket and $39.00 for seat 12A. Back so far: the $39.00 seat fee to her Visa ending 2218 and a $15 meal voucher. The ticket itself was used for CV 736 — no fare difference.
Did we offer a credit before the refund right today?
No. Every one of today’s 2,410 disruption offers stated the refund right first. 212 passengers asked for a travel credit after hearing it (L-RF3). Two Spanish replies used “crédito” alone — a wording issue, now with the glossary owner.
Raise the destination-hotel limit to $140 and test it
The proposal is ready: up to $140 when arrival is after 12:30 am and the cause is controllable. I’m replaying last week’s 2,140 conversations under it now — you and Celia Hartman sign before anything changes.
13 screens from the working solution, on its sample data. Pick one to see it large.
Everything the agent may do on its own is a limit in plain words, with what happens above it. Changes are proposals, rehearsed and signed by two people; every one is on the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| Rebook when Corvane cancels or significantly changes the flight (L-RB1) | On its own | Any Corvane flight within 48 hours, same cabin, no fare difference; other airlines or cabins → a person |
| Book a partner airline (L-RB2) | Never on its own | A care agent decides — interline cost to the airline |
| Refund to the original form of payment when Part 260 applies (L-RF1) | Up to $2,500 a ticket | Over $2,500 → a person |
| Offer a travel credit instead of a refund (L-RF3) | On its own | Only after stating the refund right, and only if the passenger asks — never as the first offer |
| Return a paid seat fee when the seat is not provided (L-SB2) | On its own, any amount | Part 260 fees for services not provided |
| Meal voucher, controllable wait of 3 hours or more (L-CR1) | $15 a passenger | Customer Service Plan §4; more → a person |
| Hotel and transport kept overnight at the departure airport (L-CR2) | Up to $160 a night | Partner hotels, controllable; more → a person |
| Policy Guard confidence line | 85% | Below it, or any disagreement, the action goes to a person |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (No Jitter and CX Today — $2.70–5.60 per contact-center call; 49 U.S.C. 46301 as amended by the FAA Reauthorization Act of 2024 — up to $75,000 per violation; 14 CFR Part 260, DOT automatic refund rule, in effect October 28, 2024; 2024 BCCRT 149 on an airline's chatbot statements). Corvane Airways, its flights and the people named on this page are fictional — characters and sample data in the working solution.
An airline virtual agent is conversational AI for airlines that talks with passengers in the app, on the web, by text and email and does the work: it rebooks them, reissues the ticket, refunds what is owed and answers trip questions. Airline Virtual Agent does this with eight AI agents inside limits your people write and sign; a second agent checks every ticket or money action before it runs, and anything outside the limits goes to a person with a ready-to-approve brief.
Yes, inside the limits. On the sample day the agents rebooked 3,184 passengers, made 1,206 refunds for $498,712.40, changed 1,468 seats, added 787 bags and issued 1,930 meal vouchers. A partner airline, a destination hotel or a refund over $2,500 goes to a person.
A 2024 tribunal decision held an airline to what its chatbot said. That is why every statement and trip-card line cites its source, why the agents cannot promise what a limit does not allow, and why the Policy Guard checks each action against the limits and the regulation before it runs.
In the working solution, the first message on a cancellation states the refund right before any alternative, and a travel credit is offered only after that, and only if the passenger asks. Refunds go to the original form of payment, as 14 CFR Part 260 describes, and paid seat and bag fees for services not provided come back on their own. Your compliance team still owns how the rule applies to your tickets.
Every virtual agent to live agent handoff carries a brief: the Hand-off Writer gives the care agent the passenger's words, what was done, the limit that stopped the agent and one proposed action with its cost and reason code. The person approves, edits or rejects with a reason code; the virtual agent tells the passenger who is coming and when, and the passenger never repeats a word.
Yes. The first line of every conversation says it is the airline's virtual assistant, an AI, and that the passenger can ask for a person at any time. Card numbers are never typed in the chat; payments go through a secure payment link.
Yes. The airline AI agent acts only inside limits that are plain sentences signed by people, and some requests always go to a person — minors, accessibility, groups, agency tickets, bereavement, DOT complaints. A limit change is rehearsed on last week's real conversations, then signed by a supervisor and the Director of Digital Care before it goes live.
The Agentic Solution Engine builds and deploys this AI customer service agent for airlines from your requirements and documents — your policy pages, Customer Service Plan, limits and a sample of past conversations — and it goes live once every quality gate has passed. In the working solution, the setup ends with a rehearsal on 30 days of past conversations before the agent talks to a passenger.
Every disrupted passenger rebooked and told before reaching the counter.
1-min demoAirline Refund AutomationEvery refund decided inside the DOT clock, each with its rule and its proof.
1-min demoEU261 Claims ManagementEach claim paid or refused on the flight’s own record, with evidence a court accepts.
1-min demoWe'll run Airline Virtual Agent on your own policy pages, limits and a sample of last month's disrupted conversations.