This week I interviewed Robert Byrne, Founder of Terra Technology, on the results of their fourth benchmarking study on forecasting excellence.  For those that do not follow this work, let me give a preamble. The work done by Terra Technology, in my opinion, is one of two accurate sources of benchmark data on forecasting in the industry. The other is Chainalytics demand benchmarking.
There are many forecast benchmark studies in the industry, but most have a tragic flaw. The issue with most forecasting benchmarking is that the data is self-reported.  Demand planning processes lack standardization and self-reported data is suspect.

Background on the Study

Eleven multinational consumer products companies participated in the study. They are large and significant, representing a total of $230 billion in annual sales.
Complexity Escalates. Companies want to grow. Success in new product launch and trade promotions is critical to accomplish this goal. However, the increase in new products and trade promotions makes the task of forecasting tougher than it was four years ago. In the study, new products represent 17% of total cases shipped. New product shipments increased 10% over the last three years.  This was coupled by an increase in seasonality and promoted items. Traditional forecasting processes support the forecasting of turn volume, or baseline products, well, but are not well-suited for new, seasonal and promoted products. New products and promoted items had 4-5X the bias of turn volume. Products in the long tail of the supply chain have an average error of 70% MAPE and a 15% bias.
Figure 1.

• Tougher in Europe. Both bias and error are higher in European supply chains. The average MAPE for North America was 36% while the European average MAPE was 45%. The average bias of European forecasts versus North America had 2% more bias.
• Process Excellence Helps. Forecast Value Add (FVA) analysis has increased in popularity. The use of this continuous improvement process had a significant major impact on the bias and error of leaders.  The average MAPE for top performers in the study is 46% and the use of FVA and other techniques reduced bias from 7% to 2%.
• Technologies Need to Change. In addition, the use of statistics to replace rules-based consumption (often termed “demand sensing”) reduced demand error of the forecast at the warehouse level by 33% as shown in figure 1.

My Take:

If growth is important to your business, you cannot manage demand planning processes like you did ten years ago.  My recommendation is to:

• Use FVA: Aggressively implement Forecast Value Add (FVA) processes.
• Focus Outside-in. Get serious about demand modeling. Many of the forecasting systems in the market just do not have the depth to do the type of modeling that is required in the face of this complexity. Reimplement traditional forecasting systems to model “what is to be sold” using attribute-based modeling. Aggressively integrate multiple demand streams (downstream data, warehouse withdrawal data, and market intelligence).
• Flexibly Manage Attributes to Help Modeling. Manage history based on market attributes and aggressively move from “SKU-based modeling” to an “attribute-based model” based on attribute-based views of history. Synchronize and harmonize downstream data using an attribute-based model.
• Build Global Excellence. Carefully define the role of the region and the role of the global team in the reduction of bias and error. Actively use FVA to improve and align global modeling.
• Implement Demand Sensing. Companies that have successfully implemented demand sensing to improve the forecast at the warehouse DC have reduced inventory on the balance sheet by 10% within two years. I do not see the same results from the multitier inventory projects.

I look forward to getting your thoughts.