How Well Can You Predict the Future?
By Jon and Matt Schreibfeder
The two basic questions faced by every purchasing agent and buyer are:
1)When do you replenish your inventory of a stocked item?
2)When you issue a replenishment order how much do you buy?
Both of these questions rely on having a good idea of the quantity of each product that will be sold or used in the future. This prediction is commonly referred to as a demand forecast. Last month, we discussed stocked items that cannot be accurately forecast. That is, products that experience sporadic usage. These items do not have a recurring, predictable pattern of sales or usage. They need to be maintained in stock based on a multiple of the normal or typical quantity sold in one transaction.
But products that are sold or used on a recurring basis can be forecast; and you need to calculate the most accurate possible prediction or future demand for each one. If you do not have accurate demand forecasts, you will probably either be over stocked (because you ordered material too early or you ordered too much) or you will experience stock outs (because you placed a replenishment order too late or your ordered too little).
The accuracy of your demand forecasts is critical to the success of your business. Unfortunately, most manufacturers, distributors and retailers do not do a good job of forecasting future demand of products. EIM conducted a study several years ago for the National Association of Wholesale Distributors (NAW). In this study we analyzed the demand forecasts and subsequent actual usage for every item in the inventory of 24 representative companies over a three-month period. The companies we analyzed came from a wide variety of industries, ranging in size from $2 million to $20 billion in sales. They used a number of different software packages. To calculate the forecast error, we used the equation:
Absolute Value of (Demand Forecast – Actual Usage)/Smaller of Demand Forecast or Actual Usage
Actual usage represents all outgoing shipments including sales, transfers and work orders. Here are several examples of how we calculated the forecast error:
Item Demand Actual Equation Forecast
Forecast Usage Error % A100 50 100 Absolute Value of (50 – 100) ÷ 50 100.0%
B200 100 50 Absolute Value of (100 – 50) ÷ 50 100.0%
C300 95 100 Absolute Value of (95 – 100) ÷ 95 9.5%
Look at the first two items in the above table. Note that we receive the same forecast error with a demand forecast of 50 and actual usage of 100 as a demand forecast of 100 and actual usage of 50. We believe it is as bad to be overstocked as under stocked.
It was surprising that the mean average forecast error for all the distributors who participated in the study was 681%. We calculated this figure by totaling all of the forecast errors for each distributor and dividing the total by the number of items forecast by that distributor. The median forecast error for all of the distributors was 382%. The median was determined by sorting the forecast errors for stocked items in ascending order and taking the middle value. That is, for the average distributor in our study half the forecast errors were less than 382% and half exceeded this value.
You might wonder how is it possible to have a forecast error that exceeds 100%? It is easy. What if you were to forecast demand of one piece of a product this month and have actual usage of 7 pieces. We would calculate that to be a forecast error of 600% [(Absolute Value of 7 – 1) ÷ 1].
When we talk with many companies, we hear claims that their sales continually come within 5% – 10% of forecast demand. A common reason is that management at these firms tend to look at total aggregate sales of the company against the total demand forecast for all items. In our study (as well as experience with other customers) it is not uncommon to find a similar number of products whose demand is overestimated as underestimated. Unfortunately, your customers order specific products, and the most accurate measurement of demand forecast accuracy is to examine the results on a “skul” (i.e., stockkeeping unit within a specific location or warehouse) level.
We found that improving the accuracy of demand forecasts is one of the most effective ways of improving corporate profitability. An informal survey of our customer base found that those companies in the upper quartile of profitability in their industries tend to have a median average demand forecast error (as calculated above) of less than one tenth the forecast error of the companies in their industry.
In the next few newsletters, we will explore ways you can improve the accuracy of your demand forecasts. But as an initial step in improving the productivity and profitability of your investment in stock inventory, why don’t you calculate the forecast error of every stocked inventory item that experiences recurring usage for each of the past three months. If an item has a high median forecast error (i.e., above 75%) examine the product and see if you determine the reason behind the forecast inaccuracy.