Finetuning Your Forecasting System

I’ve been involved in demand forecasting for decades in a variety of circumstances, including as a business forecaster, consultant, and analyst.Throughout the years, demand planning vendors have told me how their statistical algorithms could forecast better than anyone else’s and what a competitive advantage they were. As a skeptic, I’ve always advised them that as long as the methods they used detected recurring patterns in historical demand data, then that sufficed, since you could only extract so much information from history. I further advised them that they should sell their products on other features—like collaboration that incorporates market intelligence— that offer improved forecast accuracy when applied to a statistically based baseline forecast.

Terra Technology, a startup staffed by pedigree, quantitative–savvy employees from best–of–breed supply chain vendors including Numetrix/J.D.Edwards, i2 Technologies, and Manugistics, recently pitched what appeared at first to be the same old story, but this time, my response was different. Terra’s statistical methods to improve the accuracy of a forecast seem to have potential. Its approach, based on consulting experiences and incorporated in its High Definition Demand software products, uses detailed data—such as non–aggregated individual orders, cannibalizations, and substitutes—to improve an existing forecast that supports the approach of getting the "best estimate of demand, given available information." For example,Terra’s method would detect when a customer places a large forward–buy order, signaling a decrease in its demand as it is unlikely to order for a while, in contrast to most forecasting approaches that would either wrongly disregard them or take them as an increase in future demand, rather than a decrease.

Terra’s algorithms project more granular daily and weekly variations in demand, aimed at improving existing weekly or monthly forecasts. Its methods have been successfully piloted at Campbell Soup, where forecast error rates have improved significantly. Campbell is awaiting software with which to supplement its existing Distribution Requirements Planning (DRP) and demand planning systems to improve transportation equipment planning, to be followed by distribution and production planning. Other high–profile Consumer Product companies,such as Georgia–Pacific, are planning to pilot Terra’s approach.

Generally, I believe there is real promise in Terraÿs methods,but more proofs of concept are needed. Meanwhile,companies—especially those in the Consumer Products industry—looking to improve daily and weekly short–term forecasts without replacing their existing systems,might benefit by getting in on the ground floor of what could turn out to be one of the most important developments to come along in operational forecasting in a long time.

—Larry Lapide, Vice President of Supply Chain