The supply chain industry has been adopting machine learning and applying it to demand planning for quite some time now. We’ve seen its success and therefore it follows that attention has now turned to using machine learning to streamline and support supply, or production, planning. However, whilst in demand planning machine learning has been relatively straight-forward, applying machine learning to supply planning is a different kettle of fish and rife with difficulties and therefore requires a slightly different approach.
ARC Advisory Group November 2017 study, demonstrates why supply planning is of paramount importance to the full scope of supply chain planning. They state: “Supply chain planning (SCP) solutions use complex algorithms, optimization techniques, and heuristics to solve supply chain problems that occur in the planning horizon (one week to one month for operational planning, one month to several months for tactical planning, and one year to multiple years for strategic planning). While there are a variety of payback categories from the use of supply chain planning solutions, the ability to reduce inventories and effectively balance demand and supply are the central value propositions.”
Machine learning can be viewed as a means of continuous improvement, learning from multiple situations over time to refine strategy and decision-making. At its heart, it is about improving forecasting and continuing to refine this. However, the central parameters which are critical with supply planning are hugely variable. They impact immensely on the amount of inventory a business must hold, for example. It needs to be a closed loop system with continual checks in place. The vast complexity of the variables involved, which are themselves constantly changing, means that it’s an incredibly difficult system to manage. The machine learning itself needs to be capable of updating the parameters as needed.
This is a real difficulty for supply planning compared to demand planning. In demand planning the machine learning can relatively simply go back and forth refining and monitoring its accuracy and using this to predict demand. As accuracy improves over time it becomes more and more successful. What’s more, manual inputting of data ensures much of it is ‘clean’ from the start, and is often from a single or limited number of sources.
In supply planning, on the other hand, the data which is used for machine learning is coming from a range of different sources and systems. If those systems aren’t compatible with one another, or the individuals entering data into the different systems don’t have complete investment in the outputs, then the data becomes muddled and far from clean. The data management required becomes an unwieldy beast.
For supply planning using machine learning to be intelligent and useful it must have access to clean and accurate data. This could be done through a middle platform which itself analyses the different inputs and then changes parameters as required. Systems such as this do exist and are important for machine learning in supply planning.