The insights from the steps leading up to a statistical forecast are often as valuable as the statistical forecast itself. Learn what practical outcomes your business can get from each step/phase of statistical analysis.
In our current high-paced, highly volatile world, accurate future sales predictions are crucial to remain competitive. The more accurate the predictions, the lesser excess stock and obsoletes there will be, which is also very much desired from a sustainable point of view. Recently, EyeOn posted a blog about the Forecast Fast Scan and the Forecast Assessment. The main goal of these products is to assess and quantify the potential added value of introducing statistical forecasting to your monthly S&OP process.
When discussing the results of both assessments, we will initially focus on comparing the statistical forecast value added to the current forecast used at the customer, combined with KPIs critical to your company. While these demand planning KPIs are crucial for determining whether to incorporate statistical forecasting in your process, the insights gained from the inputs required to create a statistical forecast are often even more valuable. In essence, the Forecast Fast Scan is a simplified version of the comprehensive Forecast Assessment. However, both include the four pre-analysis activities that create the statistical forecast. These four pre-analysis activities are:
Categorization
The level for which a statistical forecast is created, i.e., a product or combination of product and region or even product, customer, and region, is known as a Demand Forecasting Unit (DFU). The higher the aggregation level, the lower the number of DFUs, and the more accurate the statistical forecast will be for each DFU. However, selecting a lower aggregation level might be relevant if that fits the next steps in your S&OP process (e.g., your supply, raw material, or capacity planning) better.
New Product Introductions (NPIs), End of Lives (EOLs), and active Demand Forecasting Units (DFUs) are distinguished from each other. NPIs are defined as DFUs introduced in the last few months (or more if desired). With limited data points, statistics cannot determine a reliable future statistical forecast. Hence the advice is to forecast these manually by the customer or to use different techniques, such as Driver Based Forecasting. EOLs are defined as DFUs that have not been sold anymore for some specified period. These will not be forecasted. The respective cut-off periods are determined with the customer, in line with its business dynamics.
A category will be determined for the remaining active DFUs, depicting the Pareto of total value (margin, revenue, or sometimes volume if there is no pricing or value information) with ABC-categorization and volatility with XYZ-categorization. The smallest set of DFUs generating 80% of total value (or other) receive A, the next 15% B, and the remaining C. Looking from another dimension, X category products are relatively stable and, therefore, the easiest to forecast using statistics, and the Z category shows the highly volatile, often intermittent, items. ABC-XYZ categorization focuses on which DFUs to spend the most time on and which DFUs the statistical forecast can ideally be left untouched.
Lastly, the categorization consultant will analyze the EyeOn outcome extensively to provide sound advice and w.r.t. inventory policies (e.g., MTO, MTS).
Outlier Cleaning
Statistical forecasting aims to find historical patterns to build future forecasts. In case of extreme outliers caused by a one-off event (such as Corona, Suez Canal, a shift in the product due to supply issues, etc.), the statistical forecast should not consider this. Therefore, automatic outlier cleaning is performed before the statistical forecast is created. In the report out of the Forecast Assessment and the Forecast Fast Scan, a detailed summary of outlier correction will be provided, with full details of meticulous cleaning. This forms the basis of the outlier discussion between the customer and the EyeOn Consultant: Is it an outlier, or is it a regular pattern for this DFU? Insights like these will help us further to tune the inputs of statistical forecasting to your business to align the statistical forecast with your supply constraints and growth ambition.
Seasonality Detection
The following input determines whether seasonal patterns are detected on the outlier cleansed actuals. This will generally be done on the DFU level and a higher aggregation level, often product group (versus product) or region (versus country). Usually, the customer knows when sales are higher or lower due to seasonality. However, we’d like to verify that statistically. Examples of root causes of seasonal patterns are summer holidays or Christmas.
Seasonality is often detected monthly, but it could also be relevant weekly. We have seen an example for a customer where the week number of the year showed a highly seasonal pattern. The insights from seasonality detection are discussed with the customer and will be further customized to their use case if needed.
Trend Detection
For each DFU, trend detection shows whether it is statistically significantly trended. If it is, it will provide details such as increasing/decreasing, slope, and slope over mean. In a recent Forecast Fast Scan, the customer had an ambitious growth target in terms of volume in the next few years. We performed a thorough trend analysis to determine which portion of the portfolio already showed an increasing trend to validate the customer’s gut feel and identify future growth champions.
Based on this thorough pre-analysis, a coherent, high-quality statistical forecast is generated. In both the Forecast Fast Scan and the Assessment, the possible added value of introducing statistical forecasting into your S&OP process is quantified, and it will give much more insight into your portfolio to help you steer only where needed.