Suppliers’ Dilemma: Top-down Versus Bottom-up; New Study Examines the Value of Point-of-Sale Data in Forecasting

A new study by a University of Arkansas logistics researcher confirms that relying on retail point-of-sale data can increase the accuracy of predictions and reduce forecasting error. But contrary to recent findings, the new study also revealed that in specific situations point-of-sale data might not be as accurate as simple order data from client stores.

The so-called “bottom-up” approach to forecasting demand for goods relies on point-of-sale data, or raw sales information, which retailers share with each other and manufacturers. This approach allows manufacturers to plan production based on overall consumer demand.

In contrast, “top-down” forecasting refers to a forecasting approach in which manufacturers do not have access to point-of-sale data and therefore must depend on order data from client stores and distribution centers. In these cases, the manufacturer must create a single forecast for a retail company’s total demand and then disaggregate that forecast for each distribution center or store.

Manufacturers and industry analysts assume that point-of-sale data consistently leads to greater accuracy, but the new study found that simple order data may be more useful for forecasting demand at the account level, which includes individual retail stores and distribution centers. This top-down approach is also more useful, the researchers found, when manufacturers are trying to accurately predict long-term issues such as production and capacity planning.

“Conventional wisdom holds that suppliers can exploit point-of-sale information to improve forecasting performance and supply-chain efficiency,” said Matt Waller, logistics professor in the Sam M. Walton College of Business. “While this is true for the most part, it doesn’t tell the whole story. In most cases, order forecasts based on point-of-sale data exhibit lower forecast errors than those based on order data, but there are specific conditions when a top-down approach based on order data can achieve more accurate demand forecasts.”

Waller and Brent Williams, assistant professor at Auburn University, empirically tested claims about the performance of top-down versus bottom-up forecasting. They then investigated whether a given supplier’s demand forecast, when based on shared, point-of-sale data, might be more accurate than forecasts based on order data. Overall, the researchers found that sharing the right data in appropriate contexts leads to greater accuracy when forecasting demand in the retail supply chain. In other words, the choice of a method – top-down or bottom-up forecasting – depended on the availability of shared, point-of-sale data.

“We find that the superiority of the top-down or bottom-up forecasting as the more accurate method depends on whether shared, point-of-sale data are used,” Waller said.

Firms benefit from a top-down approach to demand forecasting when they do not have access to point-of-sale data and must rely on order data for long-term planning for production. Furthermore, in this same context, a top-down approach should be used for short-term planning and shipping forecasts to distribution centers. When available, point-of-sale data can increase forecast accuracy and improve performance of short-term issues, such as inventory and transportation planning, the researchers found.

The study also gives retailers new insights. For example, large retailers share their point-of-sale data with suppliers generally because they have the technology and resources to do so, but this type of sharing may be even more beneficial for small retailers.

The researchers’ findings were published in the Journal of Business Logistics.

Waller holds the Garrison Endowed Chair in Supply Chain Management.

CONTACTS:
Matt Waller, professor, department of marketing and logistics
Sam M. Walton College of Business
479-575-8741, mwaller@walton.uark.edu
Matt McGowan, science and research communications officer
University Relations
479-575-4246, dmcgowa@uark.edu