EN
Back to the archive

The encyclopedia · R&D & Science · Operational decision · 2020–2023

Bombardier forecast its lumpy parts with flight data, up 7% accuracy

A tree-based ML model splitting demand size from interval lifted spare-parts forecast accuracy ~7% on >$1B of Bombardier aftermarket.

Bombardier

the move

Business-aircraft spare parts arrive in bursts, so standard forecasts reliably get them wrong and cause shortages or overstock.

Bombardier and IVADO Labs built a pipeline mixing tree-based machine learning with time-series models, estimating demand size and inter-demand interval separately.

Flight data and many other features fed the models, and an ensemble combined their outputs for different demand-pattern groups.

why it works

  • Splitting size from interval tackles the root of lumpiness.
  • Flight data gives the model causally relevant signals.
  • Ensembling improves robustness across part categories.
  • Forecast accuracy rose ~7% and unbiased forecast ~5%.
the payoffSplit lumpy demand into size and gap, then flight dataclever

what transfers

For lumpy demand, predict the size and the gap separately; the interval is often more forecastable than the amount.

what came after

Validation showed an improvement in forecast accuracy of about 7% and in unbiased forecast of 5%; the system was deployed and used to forecast Bombardier aftermarket demand of more than 1 billion Canadian dollars on a regular basis.

references

spotted an error? The archive wants to know.

same kind of clever