A couple of weeks ago, I discussed the issue of our obsession with forecasts and forecasting accuracy in one of the Episodes of “Think About It.” The link to that episode is below.

https://vimeo.com/manage/videos/848923933

I mentioned in a section of the video that there can easily be instances where an improved forecast accuracy will not translate into gains. Sometimes, they may actually end up generating an adverse impact. I was approached by a few folks to expand on this. In this article, I will walk through an example to explain that forecast accuracy does not always translate into a positive impact.

Let us consider a hypothetical CPG company. It has a large portfolio of SKUs and has had the capability to forecast at the SKU level for a while with a fair level of accuracy. Recently, they were persuaded to implement an “advanced” form of forecasting.

When the initial pilot results arrived, there was excitement all around. A significant level of improvement in forecasting accuracy was observed. These forecasts were used for planning purposes across functions. However, after a year, the company found that profits had taken a hit after implementing the new forecasting method, despite the significant improvement in accuracy.

Every organization that offers a large portfolio of products will have a set of products that constitute a small percentage of the total portfolio but contribute significantly to revenue and margins (80/20 principle). The company found that while the SKU level forecast had increased significantly for many low-margin products (leading to an overall significant improvement in forecasting accuracy), the accuracy for high-margin products had suffered. This was due to the demand pattern of this high-margin product.

Of course, there are ways to tweak the algorithm and forecasting approach to handle such scenarios. You can bake in an “error-optimization” element in the algorithm that quantifies the impact of errors for certain categories. But as the example illustrates, the vanilla reliance on forecasting accuracy does not cut it.