Using Data Science and Engineering to Build a Price Elasticity Estimator at Lufthansa Group
This case study is featured in ISG’s “Digital Excellence: 25 Winning Partnerships”, an e-book that profiles the best examples of digital transformation success 2019
Air transportation is a capital-intensive industry that often results in net profit margins between 3 to 5 percent. Even during periods of low fuel prices, the industry has not been able to stretch its margins beyond these levels for several years. Airlines face constant competitive and operational challenges such as keeping fares low to match competitors’ pricing, and optimizing passenger load factors for maximizing route profitability. At the same time, carriers have relatively little control over fixed costs such as airport rentals, taxes, aviation fees and others.
Lufthansa, a leading European carrier, has been aiming to compute the optimal seat pricing that would ensure higher bookings and conversion rates. Traditionally, most forecasting systems that compute seat pricing are based on booking class structures and often rely on small datasets. Pricing is driven by data and rules, and Lufthansa spent nearly three years in perfecting a data science model to determine the optimal price elasticity curve. The algorithm used three years of historical fare data by class, seasonal demand patterns and booking history data to predict the optimal price for a given seat. Lufthansa’s execution time to accurately recommend and release the optimal seat pricing based on its analytical models for a scheduled flight was more than 48 hours.
Under these operating conditions, Lufthansa partnered with Mindtree to engineer the Price Elasticity Estimator (PELE) model, the next-generation price recommendation engine. The PELE solution computes and provides the price elasticity estimates required for demand forecasters to compute the fares. The business objective for PELE was to develop a uniform platform for the Lufthansa Group airlines (Lufthansa, Swiss and Austrian Airlines) based on a completely new and pioneering approach. The solution was expected to enable the airlines to conform to the directives provided by the IATA NDC (New Distribution Capability) standard.
The PELE algorithm has a unique and innovative design that ensures parallel execution for handling more than 5 million records on a Hadoop cluster. There have been frequent instances where the quantum of data accelerates to 20 million records during processing for specific flight routes. The highlight of this innovative design is its ability to handle the complexity of processing three years of historical booking data for every route and to estimate the pricing forecast over a long booking demand window up to a year ahead. The solution was delivered in four major releases over 18 months. The PELE solution was implemented using Microsoft R Open, Java, Spark on Scala and a Hadoop cluster on the Microsoft Azure cloud. Mindtree also developed a customized database to store the PELE-computed values that are consumed by business users every day. A user interface displays the demand curves and fare values for each flight route.
Mindtree also automated the unit testing process, implemented a data comparator for easy evaluation and deployed DevOps-based continuous integration/ continuous delivery (CI/CD) automation to increase the delivery team’s productivity. Artificial intelligence and machine learning algorithms were used to improve the analytical model performance and accuracy. By using these algorithms, Mindtree was able to reduce the execution time by 98 percent – from 48 hours to 45 minutes. The model is currently being used on more than 300 flight routes with plans to cover 17,000 routes across the group airlines. There has been a more than 5 percent upside in profits in markets where PELE was implemented, which has enabled Lufthansa to cement its position as one of the most dominant carriers in Europe.