AI in airline operations: American's contrail cut and Southwest's 11 minutes

Will Drewes

Founder, Fern Strategy · LinkedIn · 3 min read

Four numbers from airline operations

Airlines have talked about AI for years. The last 18 months produced some numbers, across flight operations, customer service and disruption management.

American Airlines: contrails down 62% on 2,400 flights

American ran AI-generated pre-departure altitude adjustments across 2,400 transatlantic flights and reported a 62% reduction in contrail formation for a 0.3% fuel penalty.

Contrails are a meaningful share of aviation's climate impact, so 62% at that scale is a real operational result. The mechanism matters as much. The model produces an altitude recommendation before departure and a pilot flies it. The analysis is automated and the control surface stays human, which is where regulators and insurers want the line drawn. The 0.3% fuel cost is a useful calibration for anyone pricing environmental compliance in flight ops.

Southwest: 11 minutes per service case

Southwest reported 11 minutes saved per customer-service case and a 36% increase in employee productivity after putting generative AI tools into service operations, plus fewer contact-center calls per trip and higher day-of-travel app use in 2024 versus 2023.

Eleven minutes a case compounds fast across thousands of daily interactions. A 36% productivity figure changes headcount conversations. Southwest hasn't said whether it meant fewer staff or redeployed staff, only that the shift was large enough to put in front of investors.

Disruption management

Irregular operations are where airlines lose the most goodwill and spend the most on recovery. One carrier, unnamed in a consultant's case study, applied AI to disruption workflows and reported a 40% reduction in live call volume, a 15-point increase in customer satisfaction and 35% faster issue resolution.

Those three moving together is the tell. Call deflection on its own can mean customers gave up. Deflection with faster resolution and higher satisfaction means problems were getting solved. Unnamed carrier, so treat it as directional.

Delay analysis

An airline in a beta program used an AI assistant for delay-impact analysis and benchmarking and reported a 93% time saving on one analysis. Delay analysis sits where crew scheduling, gate management and downstream connections meet, and compressing it from hours to minutes changes what's possible during a live disruption. One case, one unnamed airline, so illustrative. The direction matches what operations teams say anecdotally.

What this means if you run airline operations

The outcomes that hold up are narrow, bounded tasks with stable rules and messy inputs: an altitude adjustment, a case summary, a delay calculation against known constraints. In none of them does the model make the final call on something with safety or liability consequences.

Aviation's governance question is sharper than most industries'. FAA oversight, union agreements and liability exposure mean the audit trail on any agent touching flight operations has to be complete and legible at the time, with no reconstructing it afterward. The American case works partly because a pilot still executes the recommendation. Take that checkpoint out and the regulatory and insurance picture changes.

If you're deciding where to start, Southwest's numbers point at the contact center as the lowest-friction entry. American's show flight ops is viable with the human checkpoint designed in from day one.