4 Presentations

• 
  03:40 PM - 04:05 PM

  Decision support system for customer requests using stochastic optimisation

  • Simon Boivin, Ray-Mont Logistics
  • Nicolas Boez, presenter, Ivado Labs
  • Jean-François Cordeau, HEC Montréal, GERAD, CIRRELT
  • Guy Desaulniers, GERAD - Polytechnique Montréal
  • Emma Frejinger, DIRO and CIRRELT

  Ray-Mont Logistics, a company specialized in international freight transportation, and Ivado Labs, an AI solution provider, collaborated to improve decision-making regarding new customer requests, especially when a risk of exceeding the capacity of Ray-Mont resources exists. The purpose of the project is to provide recommendations on how to respond to new requests considering the future utilization of Ray-Mont assets. The proposed methodology is composed of two components. First, a Monte Carlo simulation-based tool using historical data to translate customer requests with uncertain/unknown information into a projection of resources utilization. Second, a decision model that leverages this projection to provide appropriate answers to new customer requests. This model is calibrated to find the best trade-off between maximizing resource utilization versus the risk of exceeding capacity, while respecting at best the initial wish of customers.
• 
  04:05 PM - 04:30 PM

  Blackbox optimization for generic component assembly optimization

  • Sayed Ibrahim Sayed, presenter, IVADO Labs
  • Louis-Philippe Bigras, Ivado Labs
  • Jean-Daniel Dea, MDA Technology company
  • Peyman Kafaei, IVADO Labs
  • Julie Kenziel, IVADO Labs
  • Julien Laferrière, Ivado Labs
  • Charles McDonach, MDA Technology company
  • Jean-François Landry, Ivado Labs
  • Guy Desaulniers, IVADO Labs
  • Louis-Martin Rousseau, IVADO Labs

  Minimizing waste in a production line through the best utilization of components involves optimizing component selection to build an assembly in the most efficient way possible. The problem involves assigning components from a warehouse to seats or positions in an assembly to create an optimal kit with the best score. The score is generated by a simulator (black box). We propose a procedure to approximate the black box and model the approximated problem as a MILP. We then embed the MILP in a Large Neighbourhood Search (LNS) procedure to iteratively update the approximation and move towards a better solution. The LNS generates near-optimal solutions for building the optimal assembly under several configurations. Additionally, we generalize the single kit problem and adapt the solution methodology to solve the multi-kit assembly problem with the objective of optimizing the score of the worst kit.
• 
  04:30 PM - 04:55 PM

  Reserve price optimization for airlines’ ancillary products

  • Maëlle Zimmermann, presenter, Ivado Labs
  • Claudio Sole, Ivado Labs
  • Louis-Philippe Bigras, Ivado Labs
  • Nicolas Boez, Ivado Labs
  • Andrea Lodi, Ivado Labs
  • Maxime Cohen, Ivado Labs
  • Pierre-Luc Bacon, Ivado Labs
  • Emma Frejinger, Ivado Labs

  Plusgrade powers the global travel industry with its portfolio of leading ancillary revenue and merchandising solutions. They pioneered a bidding system for several ancillary products, such as upgrades into premium cabins, allowing passengers to bid an amount of their choice (above a given reserve price) to obtain a product. Optimizing the reserve price is a key lever to maximize Plusgrade's revenues. The proposed methodology is a stochastic optimization approach where scenarios are drawn from learnt distributions. While the optimization problem can simply be solved by explicit enumeration, the key challenge is to learn the distribution of the passengers' willingness to pay for upgrades. The solution leverages a non parametric empirical distribution estimator for censored data from survival analysis techniques.
• 
  04:55 PM - 05:20 PM

  Optimizing airline branded fare pricing with machine learning to maximize revenue

  • Carl Perrreault-Lafleur, presenter, Ivado Labs
  • Sajad Aliakbari Sani, Air Canada
  • Adam Bockelie, Air Canada
  • Tianjiao Liu, Air Canada
  • Alan Regis, Air Canada
  • Yury Sambale, Air Canada
  • Cindy Yao, Air Canada
  • Aldair Alvarez, Ivado Labs
  • Teodora Dan, Ivado Labs
  • Emma Frejinger, Université de Montréal
  • Andrea Lodi, CERC, Polytechnique Montréal, Montréal, Canada and Jacobs Technion-Cornell Institute, Cornell Tech and Technion - IIT, New York, USA
  • Guillaume Rabusseau, Université de Montréal

  In the airline industry, companies often choose to bundle their airfares with additional features and ancillaries, such as refundability, seat reservation, and checked baggage. These bundles, called branded fares, are built to offer a set of products matching different customer needs and preferences. Pricing each bundle relative to the others (“buy-across prices”) poses challenges and opportunities to decision-makers since this has the potential to impact the revenue of the airlines. Although the revenue management literature in the airline industry is rich, how the buy-across prices should be determined is not well studied. In this talk, we present parametric and non-parametric methods to model willingness to pay when there is low variability in the data. We use such models to optimize buy-across prices with the objective of maximizing expected revenue and deploy our models in a decision support application for pricing managers.