Unite and Conquer the Demand Plan

By Darryl Landvater

In recent months, many analysts and business administrators have noticed a change in how companies are forecasting demand for their products and necessary supplies. For many, this has led to a question of whether distribution resource planning (DRP) is an outdated or near obsolete form of predicting how much product will be needed and when.

It is a bit premature to be asking that question. The "death" of DRP, as some have seen it, isn't occurring, even with current advances in both computer hardware and software. However, DRP is evolving, and new approaches are taking hold, creating substantial improvements in demand planning accuracy. There are some lessons to be learned from current implementations that will help companies bring demand planning into the 21st century. The impact on financial performance is huge, so it's worth taking a closer look.

The Data Problem

Going back to the 1970s, suppliers and retailers have calculated demand for products based on boxes leaving the warehouse — or in other words, the people managing supply chains typically figured out how much inventory they would need based on how much of that inventory was shipped. But anticipating demand at the distribution center turned out to be less than ideal. For example, if a company's stores had more inventory in a current year than they did the previous year, the demand on that company's DC would be significantly less, even though the product may have been selling similarly in stores. Or, the demand for the product may have increased in stores, but that increase might not have become visible at manufacturing facilities for weeks — one of the many characteristics of what's referred to as the "bullwhip effect." So, for many years, people have realized that it would be better to forecast demand at the store level, and then calculate demand back through the supply chain.

Unfortunately, this vision simply wasn't technically feasible — until recently. Two main problems traditionally kept forecasters out of the store. The first is the incredibly high volume of data involved with planning products at the point of sale. When a business forecasts at the warehouse level, that business is dealing with potentially several thousand SKUs — pallets of product moving in and out. At the store level, however, especially with larger, global chain stores, those businesses need to consider those SKUs multiplied by the number of stores. This means the system needs to deal with tens or hundreds of millions of store-item combinations over the entire organization. In the past, the software used to plan distribution centers was not able to handle the store-level data volumes with then-current computer hardware.

A second problem has been the functional differences between retailers and suppliers, which operate their DRP systems at the store level and at the DC-level, respectively. Slow-moving items are a good example: There are relatively few slow-moving items at the DC level, so those items can essentially be ignored while they're in the warehouse. However, at the store level, there are a significant percentage of these items, and so they cannot be ignored. Instead, there must be logic built right into the system to handle them appropriately.