Academic journal article The Journal of Business Forecasting

Forecasting Inventory: Using Demand to Determine Supply

Academic journal article The Journal of Business Forecasting

Forecasting Inventory: Using Demand to Determine Supply

Article excerpt

Very often businesses do not fully take forecast accuracy into account when determining inventory levels ... describes the best way to calculate inventory that accounts not only for the forecast accuracy but also for variations in demand and degree of certainty needed ... very often when forecast error improves, no measurable reduction in inventory is made.

Do you spend hours mulling over data and/or software determining tiie clearest picture of what sales are going to be? For most companies, demand is only half of the story. While one department estimates demand, another plans supply to meet what it believes demand would be. You can come up with the best forecast that money can buy, unless that forecast truly drives your supply chain it would be like being all dressed up without a date to the prom

Three questions are often overlooked:

1. Based on a forecast, does a company really know how much to produce?

2. By how much will your company's inventory decrease with a decrease in forecast error?

3. How certain are you that your inventory will be enough to meet true customer demand?


In a perfect world, if a forecaster predicts that we will sell 100 units, then the company would produce 100 units. In that scenario, our stock on hand would be zero and customer sendee will be 100%. The first problem is that everyone knows that forecasts are always wrong. To ensure that customers have what they want when they want, companies keep additional stock over and above the forecasted demand. For some companies, this may be a set number of best-guess days of supply; for others, it may be inventory levels based on some valuations in orders or min-max replenishment plan. Regardless of the method used, they all have some mechanism for having a buffer or safety stock to compensate for the error resulting from what the forecast indicates will happen and what actually happens.

One issue I have observed with most of these methods is that they may compensate for variations in the historic demand, but they are not directly impacted by forecast error or probability. Many times when forecast error improves, no measurable reduction in inventory takes place.

Let us assume the current forecast model predicts sales of 100 units over the next month. The standard deviation for this item is 20 units. Based on the historic demand the company will produce 140 units [(20 ? 2) +100] after compensating for standard deviation. It is assumed here:

Supply (Number of units produced) = Safety Stock + Forecasted Demand


Safety Stock = 2 Standard Deviations for the 95% Confidence Level.

Let us say that we end up with actual sales of 1 16 units. The customer is satisfied because we have one 100% customer service, but we are left with 24 units of excess inventory. Let's further assume that forecasting has improved drastically. The forecast analyst has compensated for seasonality and produced forecasts with a new model that has a RMSE (Root Mean Squared Error) of zero. The historic standard deviation has not changed much for this item; it may still be 20 units. While the forecast analyst says now that we will sell 116 units, the company still produces 156 units [(20 ? 2) + (116)]. Forecasting has achieved 100% accuracy. The customer is still satisfied because they received what they needed. However, management is wondering why we now have 40 units of excess inventory in stock if the forecast is so much better.

We know that the purpose of any buffer stock is to compensate for a forecast, yet the above situation occurs quite frequently. Over and over, people, systems, and processes fail to make the direct connection of safety stock to the forecast. It is clear that companies need a method to balance customer service with the cost of carrying inventory. It is also clear that safety stock should be based on forecasts as well as on variations in the demand data. …

Author Advanced search


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.