Machine Learning for Inventory Master Data Settings

Where logistics meets data, interesting questions arise. In our Data2Move Research Stories, you’ll find out how students have answered them. This time: Joep Atol’s quest in enabling the transition to a global data-driven supply chain at Philips.

High-tech manufacturers are innovating and implemening new technologies to conquer their challenges. One particularly hard challenge is the transition to a global data-driven supply chain. Innovations in machine learning could offer a solution. Atol’s thesis delves into what is necessary to make this transition possible.

 

Where it Started

Philips is a global manufacturer of healthcare devices and other appliances. A global initiative within Philips has been set up to improve material management in pre-production inventory. Philips wants to investigate how data-driven tools like machine learning can help to capture supply and demand dynamics better than the tools currently available.

Thus, Joep Atol’s goal was to explore and build a machine learning model that can be used to generate recommendations for inventory replenishment.

 

Findings

Atol used simulation techniques to advise replenishment master data settings based on actual data from a production site of Philips. Using this data, he developed a tool based on neural networks which provides recommendations for inventory replenishment settings.

The simulations show promising initial results. In 92% of the cases, the new tool outperformed current recommendation models. Moreover, with it came an estimated cost reduction of around 30%. Finally, by slightly adjusting the goal of the model, a 10% increase in service level can be realized compared to the current model used by Philips.

Drawing from these results, Atol states that the newly derived model significantly outperforms the models that are currently employed, hinting at the relatively unexplored power of these new data-driven approaches.


Conclusion and next steps

Using the results and implications derived from his research, Atol concludes:

“With careful model building, high-quality data, and smart implementation, the next step towards global data-driven supply chains can be taken using data-driven techniques like machine learning.”

Given these promising initial simulation results, an important next step is to test and validate the machine learning model for inventory settings in the real-world environment, i.e. in Philip’s production facilities.