Handling customer demand and optimizing the inventory for perfect in-stock and turnover rate is a process involving many both simple and more unpredictable steps. There are several parameters to take into account which makes it difficult to get an overview, which in turn results in missed goals and poor inventory turnover.

This makes it a very suitable area for applied AI, not least because an algorithm is always objective, unlike humans who tend to underestimate or overestimate figures.


GKN, a world leading aerospace supplier, had vast resources tied up in inventory and wanted to improve their inventory forecasting.

Data sources

In this project we have used historical data of inventory levels, program and production data as well as quality data, and customer forecasts.


Hierarchical neural networks, Bayesian neural networks, and gradient boosted trees.


Tenfifty’s algorithm showed significantly improved forecasts in comparison to earlier used models. The result and algoritms were used by the customer as performance scores for forecasts over different time horizons, and served as base for a new and extended project.



Anders Bjurström