The Potential of Machine Learning Models: Forecasting Amid Unprecedented Uncertainty

Good demand planning and forecasting is all about accuracy. But how do you nail demand planning amid the unprecedented uncertainty of a COVID-19 world? Using the power of machine learning models, Project ADAM is exploring how to improve forecasting for spare parts demand around the world. By incorporating data from a variety of sources, we believe we can create machine learning models that help us to learn from what is happening today, to make better predictions for tomorrow.

Our work on this project actually began 10 months ago, when air traffic was sky high. For airlines to meet demand and minimize costly aircraft downtime, they must have ways to ensure the right spare parts are in the right place at the right time. Together with Airbus’ subsidiary Satair, which provides aftermarket services such as parts and maintenance, we began to look at how we could harness the growing amount of high-quality data on Airbus’ Skywise platform and use data mining and machine learning models to improve spare parts forecasting accuracy.

No more business as normal

Then COVID-19 struck. The pandemic has seriously disrupted global supply chain management, and IATA’s most recent COVID-19 impact assessment indicates that as of early April, worldwide flights were down almost 80 percent. Right now, business-as-normal scenarios are out the window. Which leaves us with a new set of challenges, including how to adapt our exploratory work with Satair to adjust demand planning for spare parts amid drastic, uncharted changes in customer behavior. What alterations need to be made to warehouse inventories? What will the impact on sales be?

In a COVID-world, the traditional method of using a 12-month rolling forecast presents two main challenges:

  • The present: Historical data from 2019 is totally different to the current context and will be much too high
  • The future: In 12 months, with a recovery underway, 2021 consumption levels will be higher than the same period in 2020

Learning from China as recovery kicks in

At Acubed, we pride ourselves in moving fast and pivoting when we need to. Our team is now creating and testing new dynamic models, feeding them not only historical data but also information on current events, such as government travel regulations and fleet utilization, as well as estimates on how the air travel industry will recover in the months and years ahead so we can recalibrate our demand planning and forecasting.

We’re paying particularly close attention to China, which is ahead of the rest of the world in its recovery phase. By using dynamic modeling to understand what happens to demand there, we could be able to predict how demand will shift across the rest of the world, as recoveries kick in over the coming months.

But these are still early days. As things progress, we’ll have more data, more evidence and a greater understanding of what’s happening. This will enable us to build richer, more accurate machine learning models to make better forecasting decisions and simulate different scenarios as we continue to adapt to this ever-changing crisis.

This is just one of the many ways Project ADAM is using its software expertise and toolsets for the benefit of the aviation industry. Visit the project page to find out more about how ADAM is accelerating aerospace design and manufacturing through digital technologies, and watch this space for more updates.