Inventory Management: How to Improve Demand Forecasting

Want More?

Download the PDF

Want to stay up to date on the trends and issues impacting your supply chain?

I understand that Tompkins will only use this information to contact me about business opportunities. By completing this form I am confirming that I have read and accept the Privacy Policy.

This field is for validation purposes and should be left unchanged.

Contact Us

We would love to hear from you.
Call Us: 561-994-0012

Want to stay up to date on the trends and issues impacting your supply chain?

I understand that Tompkins will only use this information to contact me about business opportunities. By completing this form I am confirming that I have read and accept the Privacy Policy.

Published January 25, 2022

Share on:

While information sharing can help predict upcoming demand for some inventories, the quantity and/or quality of the information may be limited, creating the need for forecasting. For those inventories that directly serve end customers, information sharing is not an option and demand forecasting is a necessity.

Routine demand

Routine demand can be characterized as having demand patterns that are uniform, seasonal, constantly increasing or decreasing, or a combination of any two of those patterns. Routine demand can be forecast relatively well for existing products that have a 12-month history and all of the methodologies are appropriate in certain applications.

Causal forecasting tools attempt to develop statistically significant correlations between past demand and its underlying driving force(s) so that future demand can be forecast based on the actions of the driving forces. Causal forecasting tools are complex and include least squares regression models and econometric models of all types. They are almost exclusively useful for inventory in the aggregate, as opposed to individual SKUs, and are beyond the needs of most users.

Nave forecasting tools extrapolate historical demand data using simplistic (compared to causal) mathematical extrapolation for individual SKUs. Nave models are based on the use of moving averages, which requires extensive data, or exponential smoothing, which requires significantly less data.

While experienced personnel can apply judgment to routine demand SKUs, the results are not exceptionally good, even when 12, 24 or 36 months of sales history is available to them. Compared with nave forecasting tools, differences of 30-40% will be evident in individual time periods and reality may be even more variable.

In routine demand forecasting, it is particularly important to address significant increases or decreases in demand that exceed preset limits. If these are not addressed, particularly before the next purchase or manufacturing work order is released, the risks of stockouts are significantly increased.

Beyond the type of naive or causal model used, the planner has significant control over its application, which include:

  • The length and point in time of the basis period used
  • The hierarchal level at which the forecasts are to be made (supplier, product category, SKU)
  • The extent to which the planner chooses to personally review seemingly extraordinary events

Very few products have demands stable enough to estimate future demand based on judgment. Discrete events, in particular, are difficult to model and negatively impact the results. Further, even with good forecasting software, the planner is routinely faced with situations that may represent the beginning of a trend that is not yet fully developed or a one-time, fluke event that will never be repeated. If the planner is incorrect in their interpretation, the result will be undesirable-a product shortage or excess stock.

Event-based demand

Forecasting demand associated with individual events is much more difficult than forecasting routine demand. Sales, advertising promotions, holidays and similar one-time events-even if repeated quarterly, semi-annually or annually-should be forecast separately from the routine demand if the short-term peak in demand exceeds the planned safety stock.

If the planned safety stock will likely be exceeded, there are several reasons for creating a separate event-based forecast for the SKU. First, forecasting the event separately will ensure that it gets addressed directly. Second, separating the event history from routine history will ensure that the routine forecasting does not see the event demand as the beginning of a trend or an event that may reoccur. Third, if there is a stockout, the routine history will not have to be edited to capture the lost demand.

Inventory reduction through improved forecast accuracy

Accurate forecasting is key to both sides of inventory management-minimizing working capital while simultaneously providing the desired level of customer service. Free, trusting and open communications with customers is the best way to ensure accurate information.

Depending on the particular software application, there may be one or more ways to forecast or reforecast upcoming demand, including the following:

  • Multiplying the calculated average demand by a factor to arbitrarily increase or decrease the average
  • Multiplying all of the history referenced by a factor to arbitrarily increase or decrease the historical demand for each period
  • Overriding software-identified trends
  • Limiting the history referenced to the most recent periods or the upcoming periods in the previous year
  • Modifying (correcting) history to remove one-time demand
  • Entering an arbitrary demand for each upcoming period

To develop accurate forecasts, all forecasting personnel must:

  • Be well versed in all of the possible techniques available in the selected software
  • Understand the situations in which each of those techniques are most appropriate
  • Have the information available to remove extraordinary, one-time events from past demand and incorporate best estimates of anticipated demand for future one-time events
  • Be free to use their best judgment, consistent with their past results

Without historical demand data from the supplier, new products are notoriously difficult to forecast because of the absence of any sense of what to expect beyond generalities. Many new products are forecast based on the history of similar products. This may be a good approach, provided that the base product is carefully selected.

The important concept to recognize is that manufacturer, product family, physical characteristics, application, use or similar points in common may have little to do with the profile or absolute values of the upcoming demand. There are more complex points that may be much more useful in selecting a product to mirror, such as the depth of the existing product line in terms of manufacturers, models and price points; the relative availability of the product in the market; the price; and the advertising to accompany the product release.

One of the most fundamental ways to improve the demand forecasting of new products is to measure and report forecast accuracy. Forecast variance may be reported in the aggregate or-depending on the software application-it may be drilled down to separate the variance for items such as regular products, very new products, very old products and seasonal products, if more definitive data is needed for meaningful feedback.

Demand forecasting can be extremely challenging, especially during times of uncertainty. Contact us today to learn how to improve your demand forecasting in today’s unpredictable environment

Newsletter Signup


Sign up for our latest Insights and News.
Join over 50,000 others, it’s completely free!

I understand that Tompkins will only use this information to contact me about business opportunities. By completing this form I am confirming that I have read and accept the Privacy Policy.