machine learning, automotive erp

Whenever I visit an automotive supplier, I often hear how difficult it is to deal with fluctuating schedules from OEMs and Tier 1 suppliers. It has been an issue in the supply base for as long as I can remember. No one has been able to come up with a good way to deal with it. However, through an Industry 4.0 pilot project, we believe that assistance is finally on its way by leveraging machine learning.

Planning Efficiently Can Be Difficult

If we look at today’s environment, planners are often challenged with ordering long lead-time parts, making it difficult to know what to order from sub suppliers. We all know the closer we get to the shipment date, the more accurate the data gets. However, imagine you have a 16-week lead time for the cloth that goes into a seat cover because your lower tier supplier has to grow cotton in order to produce it. How does a planner attempt to easily and accurately determine what they need to schedule? Many planners succumb to developing a complex manual spreadsheet to decide the quantities for planning. However, this often leads to costly over or under planning. In addition, due to the skills gap in manufacturing, many organizations find it difficult to replace a highly skilled planner with a new hire on a complex spreadsheet.

As we look to the future, it is not going to get easier for OEMs to predict demand. Therefore, dealing with fluctuating demand is only going to get worse. For instance, no one knows with precise accuracy when the demand for combustion engines will start to significantly decrease as more electric vehicles are offered to the market. In addition, no one knows if the demand for vehicles will start to decline as ride sharing gains in popularity. These are just some examples of the difficulty for OEMs and Tier 1 suppliers to plan for the foreseeable future.

Leveraging Machine Learning

We have completed three pilots with customers to understand how machine learning might help. Similar to the example above around growing cotton for cloth seat covers, QAD took two years of EDI planning data (e.g., 830, DELFOR) and shipment data from a customer in the most recent pilot to determine if we could more accurately predict what the OEM would ask the supplier to ship 16 weeks in the future by leveraging machine learning (see Figure 1 below). In this pilot, we found the OEM Plan Mean Percentage Average Error (MAPE) for EDI planning data was wrong 70 percent of the time. However, the machine learning MAPE was wrong only 13 percent of the time. One could make a good argument that by carrying safety stock you can significantly save in inventory.

machine learning, automotive erp

(Figure 1)

In all three pilots, both QAD and participating customers found that machine learning was more accurate in predicting quantity than the OEM. Machine learning offered promise to assist planners by significantly reducing the time and complexity in predicting the right quantities. In addition, with each new EDI forecast message and shipment, machine learning will adjust the predicted quantity accordingly.

A New Kind of Digital Assistant

There is debate among our customers on whether or not to leverage the results from machine learning as a digital assistant or place the results directly into QAD automatically. If it is used as a digital assistant, the planner would be able to see the predicted quantities and determine whether or not to use it. Alternatively, the planner could elect to place the predicted quantity automatically into the Required Ship Schedule (RSS) and note that it was the result of the predicted quantity from the machine learning.  

There is much excitement and promise from these pilots because there is potential to further lower inventory costs, optimize resources, maximize capacity utilization and mitigate risk. Machine learning has the potential to free up the planner’s time by reducing manual spreadsheet efforts, and more importantly, reduce costs associated with over and under planning.

If you are interested in learning more, please view our presentation from Explore 2018, which discusses our first pilot with QAD customer, Adient.

2 COMMENTS

  1. That’s a great read, thank you for posting it! I think planning is an awesome area to implement Machine Learning. While Artificial Intelligence and Machine Learning are still very far from being completely autonomous, implementations like that could work. Optimizing processes and mitigating risks are in my opinion two of the most important benefits it could actually bring. I see unlimited potential in Machine Learning for planning, I see very bright thing ahead!

LEAVE A REPLY