A sales forecast comes across your desk. Now what? For most companies, demand is only one-half of the story. While one department manager estimates demand, another manager orders product to meet what it is thought the orders could be. Is there a method to bring these two functions together and know what the customer will order in advance?
If the forecast predicts a company will sell 100 units, then--in a perfect world--the company manager will order 100 units, the stock on hand will be zero, and 100 percent of the customer needs will be met. However, the perfect forecast doesn't exist.
Managers know that forecasts are never "perfect." To ensure customers have what they want, when they want it, companies keep additional stock over and above the forecasted demand. For some companies, this may be a set number of days of supply, and for others it may be inventory levels based on some derivation of a reorder point or min-max replenishment. Regardless of the structure, or the justification, these methods have some buffer mechanism or a safety stock component to compensate for what the forecast indicates will happen and what might actually occur.
Most of these methods compensate for variation of historic demand, but they are not directly affected by forecast error or probability. All too often, as forecast error improves, there is no measurable reduction in inventory.
Effective forecasters must develop a method that uses the forecast to truly drive the supply plan. To accomplish this, the following two steps must be performed:
Step 1: Look at the forecasted number and calculate a level of confidence for meeting potential sales using the supply plan.
Step 2: Implement a method to determine the prediction intervals, and put it together with the forecast to factor for potential demand.
It is common to measure the standard deviation of past demand and use it in correlation with a normal distribution service model. For these calculations, measure the sum of variation of the average forecast error. To better represent the typical size of error over the sample, use the root mean squared error (RMSE) for the variation.
Now consider the confidence level--the reliability that the forecast will be within a determined range stated for the next given period. For example, a 95-percent prediction interval implies that 95 percent of the time the expected demand will be inside estimated ranges. Conversely, expected demand may be outside of estimated ranges five percent of the time.
With this prediction interval, determine a corresponding interval factor (z). This multiplier for the variation determines the buffer stock. To calculate the interval factor, use a table based on approximations for common prediction intervals or the Excel function NORMSINV.
Next, using a table of interval factors to calculate and determine, for example, a 95 percent confidence interval, factor for approximately 1.65 times the variation or RMSE. The result is the much coveted safety stock value and when added to the forecast, it provides a 95 percent probability of meeting the customer's needs in the next period.
Rather than just compensating for a standard deviation, this method applies a calculated technique to determine the probability of the estimated demand. Now, when forecast error improves, there is a direct reduction in inventory levels.
After exploring the use of one procurement method, give similar thought to other methods, such as the order-up-to or min-max methods. No matter the method, the key to success is using what is known through the forecasting process to determine safety or buffer stock. Although perfect forecasts may never be attainable, some methods will provide more effective approaches to modeling reorder point that consider forecastability and accuracy.
About the Author: J. Eric Wilson, forecast analyst, Tempur-Pedic, can be reached at (859) 514-4657 or via e-mail at firstname.lastname@example.org.
Timely, incisive articles delivered directly to your inbox.