Reduce delays caused by luggage offloading for a large hub-airline operating international and domestic flights.
On international flights, the airline had a large share of passengers at its hub who were connecting via different origin airports. Its operating hub allowed a minimum connection time of 45-60 minutes on average for all flight connections. Due to numerous factors, a substantial number of passengers missed their connecting flights in the past. As per international baggage acceptance guidelines, every airline needs to ensure that the baggage loaded inside the aircraft has a passenger accompanying it. So while the baggage may have already reached the aircraft, the passenger may still be underway to make their connection. If they then did miss their connection, that bag would have to be located and removed from the plane, resulting in a delay. This hurt overall passenger experience and increased delay costs.
In 2018, the airline had 13.000 flights which were affected by missed-connection passengers and baggage offloading due to which the total delay for the year ended up being 110000 minutes (about 2 and a half months). This resulted in estimated delay costs of €5.5 Million for the airline that year.
The airline wanted to build a prediction system that would tell them which passengers were likely to miss their connecting flight. This would allow them to use the baggage loading permissions in the existing departure control system, ensuring that their baggage does not get loaded until there is a confirmation that the passenger has boarded.
zeroG was tasked with leveraging existing data, while ensuring data compliance and anonymity, to reliably predict which passengers will miss their connecting flights.
We joined forces with the airline and airport’s IT and ground operations team and undertook the challenge to build the system. There were two underlying challenges for us:
1) Identifying the correct data sources and making them purpose-ready
2) Developing a reliable predictive analytics pipeline to integrate and analyse operational data.
The team first identified all relevant data features crucial to predict passenger connectivity e.g., arrival time, arrival area, travel group size, departure area, buffer transfer time, passport control flag, security check waiting time, booking class, wheelchair information, bus boarding etc. All these data belonged to different operating systems owned by departmental silos. We first consolidated the data from various sources and made it purpose-ready for the analytical use-case. As a next step we developed, trained, tested, evaluated, and scaled a prediction model that would tackle the challenge. In parallel, to make the predictive model run reliably, it needed to process all relevant data in real-time and be integrated within both a reporting tool and the actual baggage permission systems of the airline’s departure control system. We implemented a robust data architecture using Hadoop, RapidMiner & REST APIs in Linux environment that would feed prediction scores to the Amadeus interface and visualize daily reports on Tableau, adhering to existing airline’s tools and architecture.
The project exceeded the client’s expectations , both in terms of business impact and product delivery. After one month of utilizing the system at its hub, the airline saw a reduction of 39% in flights delayed by baggage offloading.
This accounted to savings of more than €280.000 after one month of operations. The airline now aims to roll out this application in a cloud native platform across all its major hubs in Europe and has forecasted the annual cost savings of up to €3M per hub.