What is a reasonable forecast accuracy for my business?” I get asked that question all the time. People generally understand that forecasts cannot be 100% accurate, but they’re seeking that elusive magic number that represents a good forecast—is it 90%, 85%, 80%? The truth is that a single number cannot describe the quality of the forecast for every situation. The example that I like to use is predicting the number of days in a given week versus predicting the total rainfall in a given week.
I think we can all agree that predicting the number of days in a week with a forecast accuracy of anything less than 100% would be indicative of a poor forecasting model; on the other hand, the prediction of rainfall would have a much lower acceptable forecast accuracy. The difficulty associated with the prediction is related to the variability of the time series.
In retail or consumer packaged goods, variability increases as we move further down the product hierarchy.Take a category like women’s tops—we know that consumers will purchase women’s tops, but having 98% accuracy for that forecast doesn’t really help us. There are trends to consider: Short sleeves, long sleeves or sleeveless? Pink or yellow? Prints or solids? One designer versus another? What is the geographic location of the consumer? Variability make new product introductions more difficult to forecast.
When we consider forecast accuracy over time, we need to look at all aspects of variability and account for: 1) The number of product introductions and 2) The variability of sales. This variability may be driven by multiple factors, including:
- Assortment changes
- Promotion strategy
- Competition
Due to this variability, forecasting teams may struggle with accuracy degradation over the course of one year, and then see a marked increase in accuracy the following year with very little effort.
I’ve seen this happen over and over again: When you change your strategy, introduce new products and/or the competition changes, you add variability to demand forecasting—and variability negatively impacts accuracy.
Without understanding the conditions driving forecast accuracy, you may affect the quality of the forecast—and that means you may introduce interventions that aren’t beneficial.