Each day, approximately 4,000 crew report for duty and operate between two and five flights. Inevitably, disruptions occur and the airline has to predict the level of disruption and allocate enough standby crew to ensure that the plane can fly.
The objective was to identify the significant factors that drive standby demand and create an improved method for crew rostering. Given all the moving parts and the sensitive nature of the data, a secure platform like SherlockML with full auditing capabilities and a centralised place to manage permissions was critical.
The analytics team worked with anonymised staff data and built a machine learning model that more accurately predicted the amount of spare staffing capacity they would need to roster a full cabin crew for flights. The dataset was too large to be handled by most computers and training the model was computationally intense. SherlockML allowed the team to scale computational resources depending on the needs of the project, and to iterate quickly.
Once the model was developed and tested, a tool was built on SherlockML using the new dyanmic model which allowed more accurate estimates of the standby staff required.
SherlockML also allowed the company to easily deploy the trained model on new data and provided the environment for effective version control.
The dynamic model reduced standby staffing levels by 7%. Consequently, the headroom was dynamically adjusted to suit the time of year, location, and other affecting factors. No longer needing as many staff on standby has resulted in more than £10 million of savings each year.
The insight on the months and locations with particularly high standby requirements have given the airline the evidence to further investigate and understand which interventions might reduce the crew standby requirement even further.