Demand Planning is the top Machine Learning App for Supply Chain Planning
Gartner recently polled both users and vendors on which types of supply chain planning (SCP) applications were employing machine learning technology. In three different analysis, demand planning and demand forecasting came out on top of the list.
With years of experience now in the rear view mirror, machine learning has matured as a SCP technology. “Early adopters” have generated significant benefits and the technology is entering more mainstream adoption. ToolsGroup’s own timeline of experience is an example. Our first Gartner machine learning customer case study appeared five years ago. About two years ago we rolled out a raft of new machine learning-based products for applications such as New Product Introduction and Early Signal Analysis. And a little over a year ago we introduced a 2nd generation of machine learning solution products transitioned to the Microsoft Azure and Cortana platforms.
Now that the technology has matured there is enough experience to draw conclusions about where it is most useful. So Gartner polled both users and technology partners to see where machine learning had made the most inroads in supply chain planning (Current Use Cases for Machine Learning in Supply Chain Planning Solutions, 19 May 2018 ID, Tim Payne). They ran their analysis three different ways:
First, Gartner asked users which machine learning applications they had deployed to augment their supply chain planning, that is to improve the quality of the planning decision. In this case, demand forecasting was the most popular application. Supply planning came in second. Demand sensing and shaping came in third.
Second, Gartner asked users which machine learning applications they were using to automate their supply chain processes. In this case, demand forecasting came in first, followed by demand sensing and shaping second. Quality management came in third.
Third, Gartner looked to software vendors to see where they were utilizing artificial intelligence (especially machine learning). Demand planning and forecasting came out well ahead of all other applications, accounting for nearly half (49%) of the total use cases. This overall category included applications for demand forecasting (16%), new product forecasting (10%), promotions forecasting(9%), demand sensing (8%), and casual forecasting (e.g., external events, weather, social sentiment, etc.) (6%).
Gartner summed up user’s needs for machine learning in demand planning: “They want to get better, more accurate demand plans that do not involve absorbing masses of demand planner time to accomplish. In essence, they want more productive demand planners who can go off to participate in more of the interpersonal relationships part of their roles with their supply-side and sales and marketing colleagues.”
Gartner also suggested that demand planning was popular because there is both the will and the way. Comparing it to supply planning, they commented, “It is a bit trickier to identify appropriate use cases on the supply planning side. One of the reasons for that is any machine learning algorithm needs data — lots of it. Relevant demand planning data tends to be more abundant, at least currently, than supply planning relevant data.” They added, “As more industrial Internet of Things (IoT) rolls out across factories, more supply-side data becomes available, which should encourage more supply planning machine learning use cases.”
Finally, Gartner saw using machine learning that improves the data accuracy as an application with few current use cases, but is up and coming. They explain, “This helps to improve the quality of planning decisions by making the supply chain model a better representation of the physical supply chain.”
Gartner’s recommendation to users is straightforward: Prioritize your efforts on demand planning, the application “that has the most SCP use cases and can bring immediate value in terms of more accurate forecasts and planner productivity.”
Click below for a short podcast on machine learning applications for sensing demand: