Data Academy · Tutorial 10 of 10 · Airlines

Governed AI in airlines

Airlines are applying AI where the operation is hardest: re-accommodating passengers when flights are cancelled, recovering crew plans after delays, and spotting aircraft defects before they ground a tail. Each of these touches safety rules, passenger rights or crew agreements, so AI only scales when rules decide, licensed people keep authority, and every change is recorded.

01 What is changing

Where AI in airlines is heading

From decision-support screens in the operations control centre to agents that draft and, within limits, execute recovery plans across PSS, crew and maintenance systems.
From separate fixes for aircraft, crew and passengers to recovery that considers all three together, because a change in one moves the others.
From reacting at the gate to contacting passengers with rebooking options and care entitlements before they reach the airport.
From isolated tools to AI embedded in redesigned roles across functions, with silos between network, operations and customer teams as the main brake.

02 Use cases

Three use cases on the governed path

Each use case runs the same path: a business question, governed context, a deterministic rule, specialist agents, a policy check, an action and a record. How the path works →

Illustrative: names and figures are invented to show the flow.

Use case 1

Cancellation re-accommodation

When a flight is cancelled, agents queue for hours rebooking passengers by hand while duty-of-care costs and complaints build. Rebooking by rule and issuing entitlements straight away means faster recovery, lower care costs and fewer claims escalated to regulators.

“NV412 is cancelled: where do the 168 passengers go, and what are we obliged to give them?”Asked by a duty manager in the operations control centre

Context: Operations control: Novair flight NV412 Frankfurt to Lisbon cancelled for an aircraft technical fault, coded so that the extraordinary-circumstances exemption is not claimed
PSS reservations: 168 passengers booked, 121 point to point and 47 connecting onward in Lisbon; 6 need mobility assistance
PSS inventory: the next two same-day departures have 74 and 63 open seats, 137 in total
Remaining 31 passengers can be protected on the first departure tomorrow morning, which has 58 open seats
Rule: Rebook in the same cabin on the earliest same-day departure, connecting passengers and passengers needing assistance first; anyone left over goes to the next morning with hotel, meal and transfer vouchers, and EU261 compensation is calculated from the flight distance band.
Decision: Rebook and issue care entitlements Severity: Medium
Agents: Re-accommodation agent Ranks alternative flights by seat availability, onward connections and cabin, keeping bookings travelling together.
Passenger rights agent Works out care and compensation entitlements under EU261 from the route, delay and cancellation reason.
Passenger messaging agent Sends each passenger their new itinerary, vouchers and a self-service link to choose a refund or a different flight.
Policy: Same-cabin rebooking and standard care vouchers are pre-authorised within the disruption policy; downgrades, rerouting on other carriers and payments above the standard band need a duty manager. Passenger data is used under GDPR and the stated purpose of the booking.
Action: New segments ticketed and reissued in the PSS, hotel and meal vouchers issued, released automatically within policy; compensation claims are prepared in the customer system for each eligible passenger.
Data: PSS reservations, inventory and ticketingOperations control flight scheduleDeparture control (DCS)Loyalty and CRMHotel and voucher suppliers

Use case 2

Crew legality recovery after a delay

A late inbound aircraft can push a crew past flight time limits hours later, and finding out at the gate means a cancellation. Spotting it early and proposing a legal reserve keeps the flight operating without planning on discretion.

“With the inbound running late, is the crew for NV88 still legal, and if not, who replaces them?”Asked by a crew controller

Context: Operations control: inbound aircraft for Novair NV88 Frankfurt to Tenerife running 2 hours 10 minutes late
Crew management system: the captain's planned flight duty period would now end at 13 hours 5 minutes against a 13-hour maximum for this report time and number of sectors
Reserve roster: two standby captains at Frankfurt, one rested, type-rated on the A320 and inside their own duty limits
Crew tracking: the first officer and cabin crew remain within limits on the revised times
Rule: If a crew member's planned flight duty period would exceed the legal maximum under EASA ORO.FTL, replace them with a legal, type-rated reserve; extension by commander's discretion is never planned in advance.
Decision: Swap in reserve captain Severity: High
Agents: Duty limit agent Recalculates every crew member's duty period on the revised times against the approved FTL scheme.
Reserve matching agent Finds reserves with the right type rating, rest and recency, and checks the knock-on effect on tomorrow's roster.
Crew notice agent Drafts the call-out and the notice to the original captain in line with the crew agreement.
Policy: Only a crew controller may change a roster; the agent proposes. The swap must respect EASA ORO.FTL, the airline's approved FTL scheme and the collective agreement on reserve call-out times.
Action: Crew swap created in the crew management system as a proposal pending crew controller approval; the roster and the captain's duty record are unchanged until it is approved.
Data: Crew management: rostering and trackingOperations control flight scheduleCrew qualifications and training recordsFTL scheme and crew agreement rulesAircraft movement messages

Use case 3

Repeat defect referral

A defect that keeps coming back on the same aircraft is a warning sign that a deferral is masking a real fault, and it usually ends in an unplanned grounding away from base. Catching it early lets engineering fix it where the parts and people are.

“Tail NV-07 keeps reporting the same bleed fault: can we keep flying it on the deferral until it is back at base?”Asked by a maintenance control duty engineer

Context: Technical log in the M&E system: bleed air fault on A320 tail NV-07 reported 3 times in the last 10 sectors under the same ATA chapter
M&E system: the open MEL deferral for the item expires in 2 days; the replacement valve is in stock at main base only
Aircraft health monitoring: duct pressure on the affected side drifting lower on each flight
Schedule: the aircraft night-stops at an outstation tonight and returns to main base in 2 days
Rule: Three or more reports of the same defect under one ATA chapter within ten sectors make it a repeat defect, which is referred to licensed engineering; agents may not raise, extend or close an MEL deferral.
Decision: Refer repeat defect to engineering Severity: High
Agents: Defect history agent Pulls the technical log entries, actions taken and parts changed for the defect into one history.
Health trend agent Summarises the sensor trend and how it compares with the fleet.
Maintenance planning agent Sets out options, such as a part flown to the outstation or a tail swap to bring the aircraft to base, with their schedule effect.
Policy: Airworthiness decisions belong to the CAMO and Part-145 certifying staff; the agent drafts only and makes no technical log or MEL entry. Records must meet Part-M and Part-145 record-keeping requirements.
Action: Engineering query raised in the M&E system and referred to a human specialist, the duty CAMO engineer, with the defect history and options attached; the deferral status is unchanged.
Data: M&E system: technical log, MEL deferrals, work packagesAircraft health monitoringParts and stores inventoryOperations control flight scheduleFleet reliability records

03 The foundation

What the agents need to understand

Core entities in the ontology

FlightAircraft (tail)PassengerBooking (PNR)TicketCrew memberDuty periodDefectMEL deferralAirport station

Systems they come from

PSS: reservations, inventory, ticketing: Bookings, seats, fares and e-tickets for every passenger.
Departure control (DCS): Check-in, boarding, load and special assistance on the day.
Operations control system: The live flight schedule, aircraft rotations and delay codes.
Crew management system: Rosters, crew tracking, qualifications and duty limits.
M&E / MRO system: Technical log, MEL deferrals, work packages and parts.
Loyalty and CRM: Passenger profiles, status, preferences and contact history.

04 Guardrails

The controls that let it scale

Airworthiness stays with licensed staff

No agent may make technical log entries or raise, extend or close MEL deferrals; the CAMO and Part-145 certifying staff decide.

Flight time limits are hard limits

Crew proposals must comply with EASA ORO.FTL or the applicable rules, and never plan on commander's discretion.

Passenger rights by rule

Care and compensation follow EU261 and equivalent regimes, so passengers in the same situation get the same outcome.

Passenger data used for its purpose

Booking and contact data are processed under GDPR, and PNR data only within the uses the law allows.

Every change recorded

Each rebooking, roster proposal and referral keeps the rule, evidence and approver, ready for authority audits and EU AI Act obligations.

05 Rollout

From the first use case to many

1
Start in the operations control centre
Begin with passenger re-accommodation, where rules are clear and the pain on disruption days is visible to everyone.
2
Model the operation once
Connect flights, tails, crew, passengers and defects into one model so a change to one shows its effect on the others.
3
Agree who decides what
Set which actions are pre-authorised, which need a duty manager or crew controller, and which stay with licensed engineers.
4
Prove it on real disruption days
Run agents alongside controllers through bad-weather and technical disruptions, and compare outcomes before widening automation.
5
Extend to crew and maintenance
Reuse the same model, policy checks and audit trail for crew recovery and repeat-defect handling, and reshape roles around them.

06 What to measure

Outcomes, not activity

On-time performancePassengers re-accommodated same dayDisruption care and compensation costCrew-caused cancellationsRepeat defect rateUnplanned groundings away from base

07 Pitfalls

What usually goes wrong

Fixing one silo at a time. Rebooking passengers without seeing the crew or aircraft plan creates new problems; recover against one shared model of the operation.
AI bolted on at the edges. Tools added without changing roles go unused on busy days; redesign controller and agent roles around the new workflow.
Incomplete maintenance records. Defect predictions fail when technical log entries are inconsistent; clean up ATA coding and log quality first.
Automating safety decisions. Pressure to keep aircraft flying can creep into automation; keep airworthiness and legality decisions with licensed people by design.

08 Diagnostics

Questions to ask your team

1
On our last bad disruption day, how many passengers were rebooked before they reached the airport, and who decided how?
2
Can our recovery tools see aircraft, crew and passenger plans together, or does each team fix its own part?
3
Which actions are agents allowed to take in PSS, crew and maintenance systems, and who signed that off?
4
Can we show any authority, for any change, the rule applied, the evidence used and the person who approved it?

09 Keep going

Frequently asked

Can an agent decide whether an aircraft is fit to fly?

No. Airworthiness decisions stay with the CAMO and certifying staff. The agent gathers the defect history, sensor trends and options so the engineer decides faster and with better evidence.

Why rebook automatically but not swap crew automatically?

Same-cabin rebooking within policy is low risk and easy to reverse, so it can run within agreed limits. Crew changes touch legal duty limits and crew agreements, so a crew controller approves each one.

Where should an airline start?

In the operations control centre with passenger re-accommodation, where rules are clear and disruption days show the value plainly. The shared model of flights, crew and passengers then supports crew and maintenance use cases.