Mohamed Moussaoui and Félicien Fichet  |  Air France-KLM

Castor is Air France's in-house scheduler optimization system designed to enhance the profitability and efficiency of flight schedules through intelligent, data-driven decision-making. The tool focuses on retiming flights to adjust flight departure and arrival times within operational constraints. Leveraging pre-forecasted revenue estimates, Castor evaluates multiple scheduling scenarios to identify configurations that maximize revenue and network connectivity while minimizing operational costs. By integrating commercial forecasts and operational feasibility into a unified optimization framework, Castor enables network planners to make high-impact decisions that improve both financial performance and schedule robustness in a competitive and dynamic market environment.

Yavuz Durdak  | Turkish Airlines / Turkish Technology

This study presents an integrated optimization framework for cargo route selection, aircraft assignment, and demand-driven path-based cargo routing in hub-and-spoke airline networks. Leveraging real-world operational data, we conduct a route and flight feasibility analysis based on multi-level profitability metrics and aircraft compatibility. A time-expanded, capacity-aware model is proposed to maximize total profit by simultaneously determining feasible routes, allocating compatible aircraft, and assigning bookings to valid transshipment paths through the hub. The framework includes three alternative formulations: (1) arc-based (flow-based), (2) time-expanded (node-link-time), and (3) path-based routing. Each approach captures different levels of temporal granularity and routing flexibility. A comprehensive preprocessing pipeline ensures realistic inputs and identifies infeasible or unprofitable routes prior to optimization. Computational experiments demonstrate the model's scalability and the trade-offs between formulation complexity and solution quality.

Arpad Toth  |  Lufthansa Systems

As part of our ongoing efforts to modernize the core logic of NetLine Plan, we are replacing our traditional discrete choice-based market share model with a machine learning-driven solution. The new approach aims to increase forecast accuracy, model adaptability, and scalability across different market contexts. This presentation will provide an overview of our modeling framework, outline the key differences between the statistical and machine learning approaches, and share early results and lessons learned from the transition.

Ron Chu, Chad Williams, Llyod Kwan, Rohit Ashok  |  American Airlines

Accurate passenger and revenue forecasting is a persistent challenge. We present Raven, a modular forecasting tool that combines machine learning with domain-informed rule systems. The system ingests internal and external data into a unified modeling framework, enabling consistent granular forecasting, from individual flight-level to network-wide itinerary and cost modeling, demonstrating 85% predictive accuracy. We introduced premium cabin forecasting, accounting for the increasing demand of premium seats by customers. We integrated functionality that leverages behavioral patterns from existing proxy markets to model future markets, with a mechanism to identify and select the optimal sponsor market for emulation. In addition, the model is being refined at the marketing airline level, enabling better distribution of passengers and revenue across operating/marketing carriers. Beyond air travel, we plan to account for the growing role of ground transportation in airline itineraries. Further enhancements include city-pair level forecasting to better capture the dynamics of multiple airports within a city, and utilizing day-of-week market size/fare data to improve temporal sensitivity.

Operating leases have become a widely adopted method for aircraft acquisition in the airline industry. However, such leases often impose operational constraints on lessees, including target hour-to-cycle ratios that reflect engine wear and maintenance requirements. Failure to satisfy these targets within designated periods may lead to significant financial penalties in the form of supplementary lease payments. This study addresses the Tail Assignment Problem under hour-to-cycle ratio constraints. We propose an exact branch-and-price algorithm to solve this problem efficiently. To enhance the algorithm's practical performance, we integrate a beam search-based method that quickly generates high-quality feasible solutions and develop a dancing links-based heuristic to provide tight upper bounds. Computational results demonstrate that our branch-and-price algorithm significantly outperforms a state-of-the-art commercial solver (CPLEX) applied to a connection network-based formulation. The proposed method successfully solves instances involving up to 60 aircraft and 450 flights to optimality.

Pranav Gupta and Carlos Jose Nohra Khouri  |  Amadeus

Connection banks are clusters of flights arriving and departing within short time frames, improving connectivity and serving key markets. These flight banks minimize layovers at hub airports, expand network reach and capture market share by enhancing service counts. Clean-sheet scheduling optimizing for banks in an integrated manner can unlock new value for airlines. We propose a local iterative search to identify a “good” set of connections using the frequency plan and subsequently use them as constraints in our clean-sheet scheduling problem where optimal connections and bank times are identified with all operational and commercial considerations. We solve the problem for a large North American carrier having nine hubs with multiple banks throughout the day. We compare the profitability and connectivity of the resultant schedule against the resultant schedule generated with their manual recommendations.

Enea Borra | Lufthansa Systems and Bernhard von Mutius | Kearney

Lufthansa Group has successfully implemented HubDesigner, Lufthansa Systems’ advanced tool for optimizing hub-and-spoke networks across multiple airlines. As the first solution to integrate origin-and-destination optimization with robustness across hubs, it has delivered significant business impact.

This session will share practical insights from applying HubDesigner at Lufthansa, SWISS, Austrian Airlines, and Brussels Airlines. We’ll highlight selected use cases such as cross-hub market steering, hub structure redesign, robust schedule planning, and connectivity optimization to show how collaborative, data-driven planning translates into measurable value. The talk offers a practical look at how to turn an ambitious optimization concept into enterprise-wide impact.

Lisa Lentati and Sergio Vivó  |  Ryanair

Building the schedule for Europe’s largest airline—with a fleet of 630 aircraft, operations at 238 airports, and more than 2,681 routes serving 190 million passengers annually—is a complex and high-stakes challenge. This presentation offers an inside look at how Ryanair tackles that challenge from both a business and operational perspective. We will discuss how the airline’s point-to-point model influences scheduling decisions, how we manage aircraft across more than 100 bases, and how crew pairings (duties) are built directly into the initial schedule rather than layered on afterward. We’ll also share some of the optimization approaches we’ve explored, including MIP-based models and local search methods, highlighting what has worked well, where we’ve faced difficulties, and what we’re still working to improve.

Elif Dogruoz and Kivanc Meroglu | Pegasus Airlines

In this presentation, we showcase how Pegasus Airlines integrates advanced optimization tools into its network planning process to drive growth and manage operational complexity. We highlight the practical use of optimization platforms such as NetLine/Plan, Fleet Assigner, and our upcoming use of NetLine/HubDesigner. These tools enable profit-driven fleet deployment, route selection, and schedule construction while accounting for real-world constraints such as airport slots, maintenance requirements, and operational limitations. Despite the power of these systems, a key challenge remains: achieving accurate and dynamic market size estimation. Traditional macro-driven approaches often fall short in fast-evolving or thin markets. We share our ongoing research into machine learning–based stimulation models and outline future directions, with the goal of developing more robust and responsive market size estimation frameworks.