Moving from Demand Planning to Automated Demand-Sensing

In a recent post on the importance of inventory management, we noted how new technologies are changing how demand forecasting is done. Long-standing demand-planning methods are gradually giving way to automated “demand-sensing” solutions that blend use of real-time data on market events that influence demand patterns and big data analytics with artificial intelligence (AI) and machine learning techniques.

Broadly speaking, the traditional demand-planning process extracts historical shipment and sales data to predict future product demand. In addition, known variables are sometimes factored into these projections. For example, it’s no secret that mitten sales increase in the late fall and during the winter whereas lawn mower sales pick up in the spring—that is, seasonality and promotions need to be accounted for.

Although it’s helpful, basing future demand predictions on past experiences (and common sense) has obvious limitations. This approach cannot identify changing market conditions in real-time or quickly evolve production and delivery across the supply chain pipeline in response to those changes.

The challenge for human forecasters is that demand trends are often hidden in voluminous raw data coming from a wide variety of sources. It’s simply impossible for even highly trained forecasters to sift through this data deluge to identify meaningful patterns that may have a bearing on future demand fluctuations.

That’s where AI-enabled big data analytics comes into play. Unlike human brains, AI and machine learning don’t become overwhelmed with too much data—in fact, the more data, the better for the accuracy and insights delivered by these technologies. Importantly, that data can come from established as well as formerly untapped sources.

Point-of-sale terminals, which can provide real-time pictures of what’s happening on the product front lines—whether in retail stores or online sales channel—are among the best sources for demand-sensing systems. But actual sales data can be supplemented with everything from weather forecasts to social media posts. If social media posts—pro or con—about a product go viral, for instance, it’s a good bet that demand will fluctuate as a result.

Massive amounts of diverse data are the fuel for AI and machine learning, but sophisticated mathematical algorithms are their engines. And, as with data, the more algorithmic models, the merrier when it comes to demand-sensing.

How so? AI and machine learning often use a predictive modeling technique called classification or decision trees. In the past, forecasters might run just a single decision tree model that they deemed the most likely to produce good results.

Nowadays, thanks to advances in compute power as well as modeling techniques, forecasters can run dozens or hundreds of slightly different decision tree models, each learning from previous models as well as from observations of actual sales fluctuations. The end result is greatly improved accuracy in demand forecasting.

Getting a better handle on future demand is of little value, of course, if your supply chain is unable to quickly adapt and respond to the anticipated fluctuations. That’s why having a well-integrated and synchronized supply chain is so critical. Demand-sensing can point companies in the right direction, but it takes end-to-end supply chain communications, visibility, and processes to actually arrive at the desired destination.

Learn how GEP can help you optimize your end-to-end supply chain at www.gep.com.