|

Cases and Engagements

Get an Inside look at Economics with the experts.
Managing Director
Education

Ph.D. in Economics, Northeastern University

B.A. in Economics, University of California at Los Angeles

Econ One, January 2018 – Present

UCLA Department of Sociology, 2015 – Presen

Independent Economic Consultant, 2015 – 2018

EY (Formerly Ernst & Young), Transfer Pricing, 2013 – 2014, Advisory Services, 2014 – 2015

UCLA for Int’l Science, Technology, and Cultural Policy, 2005 – 2017

Northeastern University, 2007 – 2009

Share this Article
May 1, 2024

Decision Trees for a Major Airline

In 2023, Econ One was retained by a major player in the US airline industry to accurately predict the residual points left by the customers of its loyalty program after redemption for regulatory compliance and financial accounting. The client operates a comprehensive loyalty program aimed at rewarding frequent flyers and incentivizing customer retention. The client faced difficulties in accurately forecasting the residual points remaining in customers’ accounts post-redemption. This lack of precision posed challenges in financial planning, regulatory compliance, and accounting for potential liabilities associated with mile redemptions. To tackle this challenge, Dr Amarita Natt (who has extensive experience in modelling outstanding points using sophisticated machine learning techniques and algorithms) and her team was retained by the client.

Dr Natt and her team proposed leveraging machine learning techniques, specifically decision trees, to develop a predictive model. Decision trees offer interpretability and can handle both numerical and categorical data, making them suitable for this problem. Using historical transaction data from the client’s loyalty program, the team trained a decision tree model to predict the optimal residual points left by customers after redemption based on their earning and redemption patterns. Cross-validation and hyperparameter optimization were used to improve model’s predictive accuracy. Variations on the depth of the decision trees as well as the choice of explanatory variables or intermediate decision points (nodes) were tried for the purpose of testing residual sensitivity to these choices. Model with highest accuracy was chosen for interpretation of results on residuals after rigorous validation checks to assess model’s robustness and generalization ability across different customers. The deployed decision tree model demonstrated significant improvements in predicting residual points left by customers post-redemption. By accurately forecasting these residual points, the airline client could enhance regulatory compliance, streamline financial planning, and mitigate risks associated with mile redemptions. Moreover, the insights derived from the model empowered the client to optimize their loyalty program offerings and tailor personalized incentives to drive customer engagement and loyalty.

Industries: Airlines and Aviation

Latest Related Resources and Insights