Interpretable AI Boosts Airline Revenue with Optimal Action Trees

A new research introduces Counterfactual Optimal Action Trees (COAT), an AI framework designed to generate clear, actionable business policies from existing observational data. This system integrates advanced counterfactual analysis, which predicts outcomes under different scenarios, with large-scale optimization techniques. It uses a method called column generation to translate these causal predictions into practical, transparent decisions that adhere to specific business and regulatory constraints. The COAT framework was put to the test in a real-world application: optimizing ancillary pricing for a major global airline. This sector is known for its complex business rules and limited opportunities for experimental testing. Over a 17-week pilot, COAT successfully increased upsell revenue per booking by 6.9%. This significant improvement led the airline to project an incremental annual premium seat revenue of $50-$150 million across its eligible domestic markets. The success of this pilot has resulted in the scaled adoption of COAT and is now influencing broader AI-driven decision-making initiatives within the organization, demonstrating its practical value in complex commercial environments.