How a new approach to demand planning can redefine success 

Devising a KPI to capture the accuracy of a demand plan is a knotty problem, even though it might seem intuitively simple. After all, there is a forecasted value for the demand for a given period (usually monthly) and the actual sales, which are well-known after the fact. Compare them, and there you go.

However, building out a KPI definition quickly leads to several complications. First, you need to decide which demand plan to use. Using the most recent will usually lead to the highest accuracy since predictions are easier for the short term than for the long term. But for supply chains with high latency marked by long lead times, the most recent demand plan will not accurately reflect the risks to the business posed by poor demand plans.

There are also aggregation issues, covering how to capture the accuracy of a group of Stock Keeping Units (SKUs) – unique identifiers used to track individual products rather than just one SKU. The approach taken to resolve the aggregation question represents a bifurcation point that lumps demand planning KPIs into two broad categories.

In the first category, each SKU is measured independently and then aggregated by some sort of average. The averaging method can come in different flavors, such as weighted/not weighted, or the units may be in volume or sales revenue. This category of demand planning KPI is referred to as demand planning accuracy.

The second group of KPIs falls under the rubric of forecast bias. Over the past 20 years, several organizations have considered forecast bias as an important measure. In simple terms, forecast bias is the tendency to systematically over or under-forecast, leading to higher inventory or worse service levels. There is therefore an inherent tendency to over-forecast – you are unlikely to be fired for too much inventory, but to deliver poor service on a major product launch will leave you searching for a new job. 

Sales teams, particularly in tough economic times, also tend to introduce positive forecast bias: it’s far easier to beg for forgiveness for not meeting lofty ambitions than to ask permission to fail, so sales will continue to attempt to hit an unrealistic target.

It should also be noted that bias is not always driven by human intervention; in times of prodigious growth, statistical forecasts based on historic orders will tend to forecast low. During rapid slowdowns, statistical methods tend to forecast high.

However, just like measuring forecast accuracy, measuring forecast bias is fraught with difficulty. At an aggregate or financial planning level, it is easy to understand and grasp. However, to get an accurate picture, we need to look at systematic bias at an SKU or forecast object level.

The overall result is invariably a KPI that is difficult for those outside of demand planning to understand. This is a serious problem. Gartner, the supply chain advisory firm, believes that the quality of the demand plan is the most important supply chain KPI. But a KPI that is not easily embraced or appreciated undermines the ability of the supply chain to communicate the performance – and importance – of the demand plan to their colleagues in other functions in the company. This is especially relevant when more and more companies aim for touchless flows, including demand forecasts, but come from a tradition of frequent forecast overwrites.