Comparison of Airline Co-Branded Credit Card Programs via Frequent Flyer Money Saver Analysis for Full-Service U.S. Carriers
BORA SUAVI UNSAL - Ph.D. Student in Embry Riddle Aeronautical University
To overcome the problems associated with earning and redeeming frequent flyer miles on flights via airline co-branded credit cards, a practical system is necessary to fulfill passenger expectations. In the first part of this research, a quantitative model called the frequent flyer money saver (FFMS) analysis is used to compare the official credit cards offered by the loyalty programs of leading carriers operating in the United States. In the second part of the research, Structural Equation Model (SEM) is used to determine what factors affect the FFMS ratio based on each airline route characteristics and operational outcomes. After ranking the airlines according FFMS, SEM modelling will help us to understand what are the reasons behind the different FFMS results. Therefore, this dissertation can also help airline top management to analyse and redesign their airline co-branded credit card program specifications in order to provide more value to their customers.
Revenue management forecasting in times of change – Addressing the Need for Speed
Thomas Fiig and Mike Wittman - Amadeus
The COVID-19 pandemic is a unique challenge for revenue management systems (RMS). Months of cancelled flights have left holes in RMS’s historical database, while bookings, cancellations, and willingness-to-pay have drastically changed. In this environment, many airlines have frozen their forecasts and relied on manual interventions to steer flights. This highlights the need for a new forecasting concept that can rapidly adapt based on as little as a few months of data.
We introduce a new methodology to actively adjust demand forecasts by monitoring forecast error on live flights. The key concept is to separate forecast components into two categories, depending on if they are likely to be resilient or volatile during the recovery period. Resilient components are estimated from historical data, while volatile components are adjusted using live sales data. Using both simulated and actual data, we demonstrate how our method can significantly reduce forecast error.
Conservative Selective Redistribution of Airport Delays
Max Z. Li - Massachusetts Institute of Technology
The network-level redistribution of airport delays reflects an aggregation of microscopic, tactical actions in response to airport capacity constraints. We investigate designing delay redistribution control policies under delay-conserving constraints, reflecting the fact that incurred delays cannot be removed in the absence of mechanisms such as flight cancellations. We demonstrate our framework on historical hourly sequences of US air transportation network disruptions, comparing the optimal selective redistribution policies against actual operations. We also provide a ranking of least- to most-costly delay-absorbing airports. Our control policies could be implemented as constraints in a standard air traffic flow management problem (TFMP), encouraging solutions to the TFMP that conform to delay redistribution requirements.
Cargo space contracting in the presence of price competing carriers under demand uncertainty
Benny Mantin - University of Luxembourg
How shall airlines compete in the spot markets for demand from Logistics Service Providers (LSPs) given their respective capacities and the realized demand? We find that the optimal prices may follow mix pricing strategies unless realized demand is either sufficiently high or sufficiently low which command pure pricing strategies akin to monopoly and purely competitive prices, respectively. Often, airlines contract directly with LSPs ahead of the spot market to sell some capacity in advance. This can secure early sales but may be at a lower price. Such contracts remove airlines’ supply from the spot market as well as LSPs’ demand. What quantity shall be contracted and at what price? We study optimal contracts between an airline and an LSP in such competitive environments. We model this negotiation as a bilateral Nash bargaining process. We characterize the contract terms based on demand characteristics and the margins that LSPs charge to their customers.
Improving Humanitarian Flight Scheduling
Yoram Mekking - Aviation Decision Sciences / Delft University of Technology
This study aims at improving the efficiency and effectiveness of flight routing and scheduling in a humanitarian setting by creating a linear programming model. Building on previous work, a new airport-based, formulation of the vehicle routing problem is presented. This model also incorporated the monthly minimum guaranteed flight hours per aircraft. Applied to a test case for the United Nations Humanitarian Air Service, the model results are shown significant benefits. When considering day-to-day optimization, the model realized cost savings of 4.6% till 10.5% with respect to the human flight planners. When considering the minimum guaranteed hours, the model obtained solutions that were 4% cheaper compared to the daily optimization mode and 1.6% compared to the human flight planner. Furthermore, analyses were performed that offer insight in the effect of the contract structure on the operational costs.
When and How Airline Bookings Will Return to Cruising Altitude
Ross Winegar - PROS
Airline passengers are starting to return, and their booking behavior has completely changed. We investigate what this new normal is going to look like and how airlines can prepare. PROS has access to a unique dataset of airline bookings attained from our customer base of dozens of carriers across the world. This has allowed us to create a global network of passenger booking flows that we have paired with other external data sources such as epidemic rates and government movement controls. We’ve built ML models to identify signals airlines can look for indicating that passengers are soon to return, and the booking behavior of those returning passengers. We forecast returning passenger mixes under multiple recovery scenarios utilizing regularizations for feature selection and importance. These predictions can be used to improve configurations of the RM forecaster and reduce the reliance upon historical bookings.
An Empirical Investigation of the Revenue Management Practices of U.S. Airlines around Sporting Mega-Events
Yuqi Peng - University of South Carolina
Using air travel ticketing data for the travel dates around the 2015 Super Bowl in Phoenix, AZ, we exploit the exogenous shock created by the conference championship results and design a quasi-natural experiment to identify market-specific airfare and demand patterns. Our findings help quantify the impact on the resolution in uncertainty by measuring the changes in airfare and demand in the airline markets of finalist teams and teams lost in the conference championship round before and after the conference championship games. This empirical evidence provides important insights to airlines.
DeepLinking of Multi-Modal Transport Platforms for Unified Travel Solution
Keerthi Kumar N - Employee
Pandemic has disrupted air travel & has caused major descend of airline business & many economies. LCCs are facing severe business sustenance challenges & combined modes of transportation (Air, Rail, Bus, Cab) is all the more important for last leg of connections. Interlining Air-travel with Rail & Cabs turns travel economics to be much cheaper by offering an integrated solution & would fosters revival of travel industry.
The paper unveils deep linking of Air, Rail & Road transport with non-disruptive, light-weight open integration of travel options from different content providers. The literature covers about unified travel offer that makes door-to-door travel comprehensive. Though major GDS providers play the role of big content aggregators for Air & Hotel, no GDS provider owns all types of data; unification of various modes of transport as alternates can open opportunities for new business partnership ventures & revenue model that are covered in the paper.
Air Travel Demand Stimulation: Learned Lessons from the Pre-pandemic Time
Ahmed Abdelghany - Embry-Riddle Aeronautical University
Stimulation of air travel demand is one of the crucial tasks that airlines are dealing with to alleviate the negative impact of the pandemic of the COVID-19 on air travel. This research investigates the phenomenon of stimulation of air travel demand in the different airport-pairs. Data from the pre-pandemic period is assembled to identify historical scenarios where airlines were trying to stimulate air travel demand. The research quantifies the impact of several variables on demand stimulation including market size, pricing, distance, geographical location, pre-stimulation demand levels, provided capacity, time of year. The first phase of the research focuses on first-time-served markets.
Spare Aircraft Placement Optimization
Nilay Noyan Bulbul - AMAZON
We study the problem of determining the locations of spare aircraft along with their reserved crew time window schedules to support the disruption recovery management of a cargo airline network. We have developed an optimization model, which provides a flexible and computationally efficient approach to ensure a certain level of network coverage while considering the multiple key factors including the total cost and the different types of flight coverage. A benchmarking analysis showed the value of the proposed optimization model compared to a heuristic approach.
Column generation based heuristics for solving aircraft recovery problem with NOTAM, maintenance and aircraft-airport compatibility constraints.
Sourabh Kumar Choudhary - Indian Institute of Technology, Kharagpur
Flight delays and cancellations due to disruptive scenarios such as unfavourable weather, aircraft mechanical failures, unplanned aircraft maintenance, etc remain common in the airline industry. Apart from the direct cost spent as compensation for the passengers for delayed and cancelled flights, the airline companies also incur indirect costs such as the loss of reputation. Column generation based heuristics have recently been used to come up with a recovered schedule. However, to the best of our knowledge, a comprehensive approach consisting of additional aspects such as NOTAM(Notice to Airmen) and aircraft/airport compatibility has to still be addressed. These issues are addressed here. Preliminary results indicate a significant improvement over solutions obtained through intuition.
Optimal Revenue-Equivalence Baggage Pricing for Airlines
Prabhupad Bharadwaj - Indian Institute of Technology Madras
Decoupling baggage fee from regular airfare has helped airlines generate additional revenue and compensate for increasing operating costs. However, there is no consensus on optimizing baggage prices in the industry, as bags are priced based on weight or number of pieces, without considering baggage profiles. This research studies the problem with an objective to maximize the revenue from passenger baggage while generating recommended prices using the baggage profile for each extra unit as an output, depending on the route. We adopt a non-linear optimization approach and impose a revenue equivalence constraint between weight-based and piece-based pricing. Results depict that prices obtained from our model are generally aligned with current industry pricing policies with a better pricing approach for units beyond the third bag for each passenger. This method simplifies the complex baggage pricing structure and proposes an optimal pricing model at the route level.
Network design and optimisation for distribution of Covid-19 vaccines
Wim Vanroose - U. Antwerpen / Motulus
We plan an optimal periodic 7-day route network to distribute covid-19
vaccines. The model has several factory locations with daily
production capacities. The number of vaccines each destination airport
requires depends on the population size or the throughput of their
vaccination lines. We then optimize a 7-day periodic flight
schedule with minimal airplane and crew cost. It is important to plan
these routes to hourly details. Some flights only take a few hours
while long-haul can take much longer. The crew motion includes
details of the rest periods and the transfers between
destinations. Modelling crew and aircraft motion to an hourly level
allows the fine-tuning of costs and the creation of synergies and
economies of scale. We illustrate the results to a region with 150
airports and a population of 750 million. The same tool can also
optimize season schedules for a holiday carrier where the yields
depends on the number of holiday combinations.
Short-term fare estimation alongside market sizes
Deepak Sunil - Employee - Amadeus Labs
Accurate market size forecasts are important for airlines to take the right decisions during network/fleet planning. This market size data, combined with a prediction of accurate fare values at the Itinerary level, would be an invaluable tool to help with this planning. However, the volatility of fares for a given market is a challenge for any prediction model. We present a forecasting approach for fares that aims to provide an accurate range of closest fares for a given search-departure week combination, up to 5 months in the future, using historical data (with Fares). We also highlight the challenges and benchmarks used to evaluate the accuracy of such a solution, alongside estimated market sizes.
Centralized Inter-Code-sharing mechanism to circumvent the shut-down of Airlines in pandemic
GAURAV KUMAR - Indian Institute of Management Kashipur, India
In pandemic, Airlines have been in crisis because of two reasons (i) Travel-restrictions imposed by respective federal-govt. (ii) Consternation of Covid-19 led insufficient demand that could not exceed the minimum operational cost causing complete shut-down of services. Minimization of cost that could satiate demand, is endogenous & hence our research work is to obviate it. We propose formation of a Centralized Airline Crisis Control Administration (CACCA) which will play a central role in crisis of airlines. CACCA is activated upon signal of crisis to administer operations of all airlines under coalition. Key functions of CACCA are distribution of capacities, facilities & revenues under inter-code sharing mechanism (Netessine, 2005) to circumvent shut-down of airlines due to pandemic. OR mathematical model is developed particularly using nucleolus (Kimms, 2012) & Shapley value (Shapley, 1953) of cooperative game theory to make the framework executable.
Improving Operations with SImulation
Bronwyn Jackson - The Boeing Company - DS&AE
When evaluating a wide range of options available to make improvements in efficiency and availability, it can be difficult to understand and quantify the full impact in the operating environment and to know which option is best. Simulations, paired with reliable data sources and digital twins, allow experimentation and analysis of proposed scenarios in a controlled, low risk environment before roll-out. This presentation demonstrates an integrated approach to “model based sustainment”, that combines engineering, data, operational models and the application of simulation techniques to evaluate and improve operations.
What to Do When There Is No History: Predicting Demand Recovery Under Covid-19 Pandemic
Cindy Yao, Gurpreet Kaur Cheema , Olivier G. Leblanc and Yiqin Song - Air Canada | Revenue Management & AI/OR Center of Excellence
The impact of the Covid-19 pandemic on air travel is unprecedented. Traditional demand forecasting models relying on historical patterns could not quickly adapt to this unique situation. We present how a cross-department taskforce was created with the objective of providing guidance on demand recovery, based on non-traditional inputs. From the initial heuristic modelling to monitoring platforms to data science models, the innovative approaches developed here have helped reestablish necessary inputs to Revenue Management and Network Planning systems and steered the commercial teams.
Airline Delay Prediction: Embeddings for Tabular Datasets
Steve Wilson - Cirium
The presentation demonstrates the construction of embedding layers as a means for capturing salient features present in aircraft movements and airline schedules, then applies them to improve an airline delay prediction task. The talk details the intuition, successes across industry, and practicality of constructing, visualizing, and using embedding layers. From an introduction to the word2vec algorithm in natural language processing, we explore the use of transfer learning for tabular data sets through the construction of reusable embeddings and look at comparable approaches for tabular datasets to improve accuracy and reduce training time.
Missed Opportunities During Ground Delay Programs (GDP) and An Approach to Improve Throughput
Ilhan Ince - PASSUR Aerospace
Airport arrival rates (AARs) are predicted throughout an operational day to inform air traffic management decisions on how much traffic an airport can handle as operational conditions dictate. It is presumed that an unnecessarily low rate artificially reduces airport throughput below capacity, leading to an increase in avoidable delays.
We present the methodology by which an “under-delivery” event can be identified and measured. Highlight how determination can be made that a called AAR is too low relative to the expected demand levels. Identify and quantify the delay impact of periods of potential AAR underestimation resulting in unnecessary unrecoverable delay from GDPs. Explanation of how a higher and more appropriate rate was determined and present an approach for forecasting airport capacity under various conditions.
Online gradient-descent algorithms for computing capacity allocation and bid prices with stochastic demand
Shashi Mittal - Amazon.com
The problem of computing the bid price for flight legs is well known in the airline industry. When we assume that the origin-destination demand is stochastic with a known distribution, then this becomes a non-linear but convex optimization problem. Traditionally sub-gradient type algorithms have been used for finding the optimal bid price in this setting. We have developed a new algorithm for solving this problem, Online Gradient Descent, which is inspired from similar algorithms used for solving optimization problems arising in Machine Learning literature. The Online Gradient Descent algorithm has better convergence properties than the sub-gradient descent algorithms, and in our computational experiments it reduced the run-time for computing the optimal bid prices by 43%. Besides computing bid prices, we also show how this algorithm can also be used for capacity allocation in the Amazon Air network for different origin-destination pairs.
Selling your last seat cheaply or expensively? A single dimensional DP approach for RM
Daniel Hopman - VU University, Amsterdam
In this paper, an optimisation method is introduced that accounts for cancellations. We do so by estimating the opportunity cost of a booking between the time of booking and the expected time of cancellation. The probabilities that are required are found using the framework of our earlier work. The dynamic programming formulation we introduce remains single-dimensional, which is important for this algorithm to be implemented in practice. The formulation involves an estimate of the value of the state of the system at the time of cancellation (which is in the future), which is found by making a choice of novel heuristics which we introduce. The fare that is used to determine whether a product is available for sale, is adjusted by the risk the airline faces. Through simulation studies, we show increases in revenues against a traditional dynamic programming formulation that does not explicitly models cancellations.
Large-scale Vibration Monitoring of Aircraft Engines from Operational Data using Self-organized Models
Florent Forest - Safran Aircraft Engines/Université Sorbonne Paris Nord
Vibration analysis is an important component of aircraft engin health monitoring. Vibrations, mainly caused by unbalance, misalignment, or damaged bearings, put engine parts under dynamic structural stress. Thus, monitoring the vibratory behavior of engines is essential to detect anomalies and trends, avoid faults and improve availability. This work presents a methodology for large-scale vibration monitoring of operating civil aircraft engines, based on unsupervised learning algorithms and a flight recorder database. Firstly, we present a pipeline for massive
extraction of vibration signatures from raw flight data. Then, signatures are classified and visualized using interpretable self-organized clustering algorithms, yielding a visual cartography of vibration profiles. The approach is global, end-to-end and scalable, which is yet uncommon in our industry, and has been tested on real flight data.