Now in its third year, the study identifies forecasting trends based on activity from about one third of the North American CPG market, and provides insight into the composition of retail sales for the prior year. The benchmark includes nine multinational manufacturers that have 90,000 items stocked in 475 locations, totaling four billion physical cases and more than $100bn in annual sales.
"Overall shipments remained flat and manufacturers relied more on promotions and product innovation to drive sales," says Robert Byrne, president and CEO of Terra Technology. "These marketing activities change consumer behavior and increase supply chain complexity. As a result, overall demand planning forecast error rose slightly to 53 percent."
Accurate forecasts are important because they form the basis of all manufacturing, procurement and inventory planning decisions, and ultimately drive overall supply chain costs. For companies under pressure to improve cash flow by cutting inventory and expenses, improving forecast accuracy is the key to safely lowering stocks without risking revenue or customer service. Furthermore, the gains in supply chain efficiency help lower operating costs and offset the margin shortfalls from an increased reliance on promotions.
The benchmark study shows that demand sensing performs throughout the turbulence caused by volatile markets, reducing forecast error from 50 percent to only 30 percent. Demand sensing is shown to be effective in improving forecast accuracy across many business activities, including new products, promotions and regular sales. It also dramatically reduces the instances of extreme error, where sales exceed or fall short of forecasts by two times or more. These extreme error events are the most disruptive and costly for supply chains.
Other key findings include:
• Promotional volume increased eight percent in 2011; only 11 percent of items were not promoted during the year compared to 23 percent in 2010.
• Manufacturers remain optimistic, consistently over-forecasting, especially for promotions and new products, where bias is three or more times higher.
• New products represent one-third of all items; half of all items have less than two-years history.
• Forecast error for new products over their first year is 74 percent, up slightly from 70 percent in 2010.
• Over one third of all forecasted volume is subject to extreme forecast error.
• Top-selling products responsible for 20 percent of the volume represent only two percent of items, whereas the slowest moving products responsible for the same 20 percent of the volume represent 80 percent of items; these slow-moving items have a high forecast error of 73 percent.
The benchmark results demonstrate the value of using big data and dynamic mathematics to create forecasts that adapt to changing conditions. With half of the items having less than two years of sales history, traditional seasonal forecasting models lack the basic information to make meaningful forecasts, especially in volatile markets. Demand sensing takes a different approach, drawing on information from multiple sources in manufacturer and retailer systems, to accurately predict future demand in a way that reflects current market realities. This is a major departure from the systems which the industry has relied on for decades.
