Symposium Program

We analyze several popular control methodologies for network revenue management (network RM) considering their relevance to industry practice and pricing applications. Comparing advantages and disadvantages of both established and novel methods such as network decompositions (resource- or product-based), iterative proration or heuristic dynamic programming methods, we point out ways to improve their performance in practice as well as user adoption. We focus on recent results available in the literature that approach network decomposition from the product point of view and suggest ways to improve them.

Variable Opaque Products encourage shoppers to actively interact with websites in order to create personalised price-product combinations. This produces a rich data set that is well-suited as training data for machine learning models to determine willingness-to-pay and to optimise the price offered to the shopper. An overview of the process used and the models underpinning the WTP calculation is presented.

What do the crew think and want from crew planning? A short presentation drawing on my experience presenting to, talking to and interviewing pilots and cabin crew about crew planning. The presentation will explore some of the common conceptions we have as crew planners, are they true or are they not? Some of the topics explored will be: Do crew like the idea of mathematicians and optimizers planning their lives. Are we, crew planners, unrealistic in our expectations of crew? Do trade unions help or hinder? Do crew planners get an unfair reputation amongst crew? Looking forward: is the landscape changing in terms of both crew lifestyle expectations and also can AI change what can be delivered to crew. Can established AI techniques help deliver an improved experience for crew.

Fleet assignment models (FAM) are widely used to maximize an airline’s profitability by optimizing the assignment of aircraft types to an existing flight schedule. In order to increase the feasible solution space and aircraft availability, the model is typically allowed to retime flights. In a novel approach to schedule optimization, we leverage this retime functionality within the fleet assignment model to also optimize for connectivity in our hubs. We use an efficient machine learning model to teach the fleet assignment model about how demand on connecting itineraries will be affected by retimes. While our method is generalizable to O&D-based fleet assignment models, we will demonstrate how it improves the quality and profitability of the schedule when used with a leg-based fleet assignment model.

Funnel-based Learning and Optimization for Offers and Revenue (FLOOR) is a machine learning framework for optimizing offer distribution in airline bookings. It uses granular customer interaction data to predict conversion probability and allocate incentives strategically—offering discounts only to users less likely to book. Features like time spent, navigation patterns, and customer attributes train gradient boosting models to forecast booking completion. FLOOR improves conversion rates while reducing discount costs versus traditional methods. Its adaptive feedback loop captures seasonal and market shifts, enabling continuous refinement. This approach shows how behavioral analytics can drive efficient, revenue-maximizing offer strategies.

We study airline ancillary seat price optimization as a contextual multi-armed bandit, where context (flight-type/itinerary/time-to-departure) informs price selection and the policy “revenue-manages” the full seat inventory over the booking window. We model demand with a Poisson GLM and treat unknown elasticities within a Bayesian belief-MDP. On small problem instances, we compute the optimal policy by value iteration, balancing exploration and revenue exactly. To scale to large problem instances, we approximate the value function with a dual-stream deep learning network that separates arm uncertainty from contextual effects and fuses them into a single value estimate. Across realistic simulations, the approach increases revenue and reduces regret versus LinUCB/LinTS/Tree-UCB benchmarks, while preserving fast decision time. We discuss sensitivity to priors/price grids and integration with inventory and booking-window constraints.

Airlines must balance ticket sales across channels with distinct demand dynamics and price sensitivities. This study examines the revenue implications of direct bookings versus allotment sales to tour operators and travel agencies, using a unique year-long dataset from two major European carriers: a touristic airline and a network airline. These airlines operate under different market structures and employ contrasting sales mixes, enabling a comparative analysis of channel strategies. Focusing on flights to key holiday destinations, we construct flight-level performance indicators and apply descriptive statistics, booking-curve analyses, and fixed-effects regressions to quantify the effect of allotment shares on overall network revenue. The findings provide evidence on how channel composition influences revenue outcomes, offering insights for optimizing seat allocation strategies in airline revenue management.

Traffic forecast at any airport or airline is key to operations planning as well as financial planning. At Groupe ADP, our traffic prediction team predicts traffic at a rather coarse granularity, typically looking at the number of flights operated from/to a world sub region by a particular airline, over a period of time, that can be days, weeks, or months. The specific problem we have tackled is the estimation of the number of scheduled flights which, eventually turn not to be operated, whether they are removed from later schedule versions or whether they get cancelled. Conversely, flights which were not planned in early versions of the schedule appear in later versions. These aspects challenge our long term planning analysis. We will go through alternative problem approaches and prediction models, their results, remaining challenges, and lessons learned.

“Can you explain why you recommend this flight?” This presentation traces the evolution of our research on flight itinerary choice models from simple linear approaches (past) to advanced AI methods (present). As these models grow more complex, enhancing transparency and explainability becomes essential (future), empowering both airlines and travellers to trust and understand flight recommendations among other applications. Finally, we discuss a new methodology for integrating LLM capabilities with advanced choice models to generate explanations that are understandable to human users, answering the traveller’s question: 'Can you explain why you recommend this flight to me?'"

Airlines increasingly use A/B testing to evaluate new pricing models and tactical changes. These experiments generate booking data that reflect complex demand patterns. A common mistake is to treat data points as independent, when seasonality, booking curves, and weekday effects actually create correlations. Ignoring this inflates significance and leads to misleading business decisions. A/A tests show the issue: false positives rise, so effects appear even when none exist. Clustering helps but widens confidence intervals and lengthens runtimes. More efficient approaches explicitly model autocorrelation. Methods such as generalized least squares (GLS) and switchback designs reduce noise, enabling more trustworthy conclusions with fewer samples and shorter experiment durations. This talk highlights why error structure matters and how accounting for autocorrelation improves both accuracy and speed in airline A/B tests.

Airlines struggle balancing planned CASK vs. operational robustness in short-term scheduling. We present a genetic algorithm (GA) based optimizer that improves schedule quality by including trade-offs between turn times, ops spares, and crew productivity, allowing for tuning turn buffers, ops spares, and crew productivity to balance CASK impact with day-of-ops recoverability. Using efficient-frontier views and simple stress tests, we quantify OTP–resilience trade-offs without claiming a single best answer. We’ll show preliminary experiments in simultaneous network planning optimisation across ports/days and outline next steps to incorporate crew feasibility and flows. The goal is a clear method for measuring revenue and cost trade-offs and informing schedule choices.

For an airline operating in the cargo spot market, real-time pricing of new request-for-quote (RFQ) orders is essential. Requests must be processed rapidly, and additional complexity arises from their multidimensional characteristics, such as type, weight, and volume. Furthermore, delivery dates are often flexible for both new and accepted orders, as shipments may be delayed at a potential cost. In this paper, we introduce a Dynamic Programming algorithm that quickly estimates the opportunity cost of cargo and reschedules accepted orders to minimize total cost. The algorithm offers greater computational efficiency than partial network re-optimization, while its accuracy and explicit rescheduling decisions compare favourably to bid-price controls derived from relaxed problem formulations. We also provide a numerical study on simulated datasets and discuss directions for future development.

SWISS and the Lufthansa Group are developing advanced software to support our operations control centers during disruption events. Disruptions may stem from various causes, but their impact ultimately falls on our passengers, making it essential that recovery decisions prioritize restoring their journeys effectively. In this presentation, we introduce a leg-based passenger recovery approach, where valid and tailored alternative itineraries are efficiently generated from leg data and assigned to affected passengers. We outline our methodology for assembling feasible legs and assessing seat availability, leveraging passenger-specific information to build customized alternatives. We then describe our cost function to price these alternatives and the optimization process used to identify the global optimal solution. Finally, we illustrate how this product is being applied in practice and provide an outlook on upcoming developments.

Railway time-tabling and scheduling is one of the toughest Operations Research problems to solve but of tremendous utility for firms. In a setting where competition is both dense and intense such as the high-speed corridors of Spain, the importance of finidng a schedule that satisfies all constraints and also is revenue-maximizing is challenging. We describe the implementation of a revenue-optimal scheduling formulation and solution that scales well and takes into account maintenace considerations and demand profiles. We describe computational experiments as well as practical implementation details.

KLM Royal Dutch Airlines tackled critical Embraer 195-E2 engine challenges, where reduced maintenance intervals threatened fleet availability, increasing parked aircraft and in-service activations. A rapid, data-driven "E2 Availability Optimiser" Proof of Concept (PoC) was developed in one week. This model maps aircraft and engine states into an engine state network, minimizing unavailability and activation costs. Initially, the model had long runtimes (>4 hours) and low optimality (>20%). Performance was drastically improved using "virtual engines" and network simplification, reducing runtime to under 20 minutes with an optimality gap below 20%. The user-accessible tool optimizes engine assignments, resulting in 2-6 fewer parked aircraft and over 25% fewer engine changes. While overall aircraft availability slightly decreased, network stability improved. Future plans include tool maintenance, continuous decision support, and new innovation initiatives.

This study proposes a two-stage stochastic optimization model to minimize total operational costs and improve OTP in the presence of uncertainties. First-stage decisions optimize pre-hub departure adjustments: aircraft swaps, departure delays, and cruise speeds controlled via cost index (CI) adjustments. Second-stage decisions, applied in the subsequent hub cycle, re-optimize CI and swaps to mitigate realized disruptions.

Achieving a shift from air to rail is key to decarbonising transport, requiring models that capture multimodal behaviour and network performance under different schedules, policies and disruptions. This talk presents the multimodal performance framework and the strategic and tactical evaluators from SESAR’s MultiModX project. The strategic evaluator generates itineraries from flight schedules, rail timetables and policies, computes network-wide indicators, and evaluates passenger impacts of replanned networks during disruptions. The tactical evaluator assesses the realised network with passenger-centric metrics. Applications to Spain include long-term policies (integrated ticketing, CO2 taxation, flight bans), short-term disruptions (industrial action, cancellations) and mechanisms to support multimodality (airport fast track). Results show rail can absorb displaced demand, strengthening resilience and demonstrating the complementary role of air and rail.