Supply planning has not changed in 30 years and is still largely the same today in ERPs, despite ever more complex and more global supply chains. At first glance, supply planning works and there is not a lot of optimization potential. In his research at ETH Zurich, Simon has tried a novel approach using machine learning to plan material purchases for a pharmaceutical company. The results have been astounding, savings representing around 10% of the whole purchasing volume. Ever since then we have been convinced that there is a lot of potential dormant in this process.
The largest shift in supply planning for the last 30 years
A machine that learns to optimise hard trade-offs
As in many places in real life, decisions in supply planning are trade-offs. An example is how to square benefitting from quantity discounts while keeping inventories low and customer service high. The proprietary GenLots algorithm has an incentive structure modelled on the actual total cost of supply planning for a specific material.
By trying out many solutions out of a pool of possible solutions which is almost as big as the number of atoms in the universe, our algorithm learns what strategies lead to the optimal supply plan, squaring all trade-offs. Not only allows this a cost optimization with astonishing results, but it is also extendible to include other factors such as CO2 emissions in its model of the world.