11 Measuring forecasting accuracy on a product portfolio

In chapters 9 and 10, we discussed multiple forecasting KPIs trying to assess the quality of a single product’s forecast (or a single time series). In practice, demand planners deal with hundreds, if not thousands, of products. Unfortunately, when computing a global accuracy metric for various products, you compare pears and apples because products are not all equally important (we will see in table 11.1 an example with nails, hammers, and anvils). Many companies understand this and look at accuracy by product categories rather than for all their products globally at once. Their implicit assumption is that looking at forecasting accuracy globally might be mixing up pears and apples, whereas analyzing products by categories will reduce the heterogeneity.
Are we bound to analyze forecast error by product family, or is there a smarter way to deal with broad product mix?
In this chapter, we will discuss value-weighted KPIs, allowing you to compare pears and apples when forecasting—and focus on the products that matter the most.
11.1 Forecasting metrics and product portfolios
Imagine that you are responsible for forecasting three products: nails, hammers, and anvils. As shown in table 11.1, the absolute forecast error is more significant on the nails (500 pieces) than on the hammers (50 pieces) and the anvils (2). This effect is even more important for the squared error.