LATAM: flight fuel decisions, predictive maintenance and an internal AI platform
LATAM has documented machine-learning work on dispatch fuel, fleet health and employee tools. The distinctive question is how quickly a model reaches service, and which decisions remain with operational staff.
The operating picture
The work behind an airline's operating data
PT-MXN LATAM Airlines Brasil Airbus A321-200, Guarulhos International Airport. Photograph: Wilfredor / CC BY-SA 4.0 (cropped to fit). Illustrates the airline, not an AI deployment.
Carrying additional holding fuel adds weight, but dispatchers must protect the flight against uncertainty. LATAM's Cosmos programme includes a model used on flights in Colombia and Brazil to support that trade-off.
Predictive maintenance presents a separate problem, where an alert needs an engineer's assessment before an aircraft is withdrawn or a component replaced. The airline's Amelia employee platform provides broader staff access to generative AI, but general availability is not evidence of improved dispatch or maintenance outcomes.
Selected implementations
The work behind the evidence
01 / Deployed
How much holding fuel should dispatch carry?
The operational problem
Excess fuel adds weight; too little buffer reduces operational resilience. Dispatch must weigh both.
What the evidence reports
A ZenML summary of LATAM's Cosmos presentation describes a deployed machine-learning fuel-holding model for Colombian and Brazilian flights, with dispatchers retaining final authority. The secondary account attributes an estimated four kilotonnes of annual CO₂ savings.
Operational significance
The model addresses a recognisable flight-planning decision rather than a generic AI transformation.
Evidence still needed
The emissions figure and model-deployment speed come from secondary reporting of a company presentation, not an independently inspected flight-data analysis.
A source can verify an announcement without verifying its projected outcome.
02 / Deployed
A warning before a component fails
The operational problem
Maintenance teams need reliable advance notice of faults without creating unnecessary inspections or removals.
What the evidence reports
Lufthansa Technik announced in April 2025 that LATAM adopted AVIATAR Predictive Health Analytics and electronic technical logs under an agreement involving more than 300 aircraft.
Operational significance
Fleet-wide predictive work can be evaluated through actual faults, intervention accuracy and aircraft availability.
Evidence still needed
The announced agreement does not confirm full fleet rollout or a quantified reduction in technical delays.
A source can verify an announcement without verifying its projected outcome.
The wider register
Other tracked activity
Other records are shown separately, including non-AI technology and programmes at different stages of deployment. A register entry does not establish a measured AI benefit.
No other publicly listed records.
Editorial assessment
LATAM's most consequential applications are tied to decisions already measured in airline operations: flight fuel and maintenance interventions.
The broader internal platform is a different category. The next editorial update should establish use and outcomes, without counting access to AI as proof of operational improvement.
Live data from the Aerostack register. Status reflects published evidence only. See Our approach for how records are graded.
Aerostack register · 3 tracked deployments
Aerostack tracks AI deployments across airports and airlines, with the published evidence behind each record. Photography is real aviation imagery. Record thumbnails may show the named airport or airline; other photographs do not.
