One big problem with supply chain forecasting is that companies don't draw on enough resources when creating their forecasts. "Done well, forecasting contributes to supply chain performance, customer satisfaction and financial success," observes Terry Harris, managing partner with Chicago Consulting. "Done poorly, little else matters."
Based on 20 years of experience, Harris offers the following "forecasting fallacies" that illustrate just how hard it is to predict the future with any kind of consistency, and how it gets even harder if you don't have a clue to start with.
1. "Our problem is crummy forecasts." In many companies, the forecast is blamed for far too many ills. The financial organization blames forecasts for high inventory cost and dead stock. The marketing and sales organizations blame forecasts for poor customer service and lost sales. Manufacturing blames forecasting for changing too often. Rarely is this blanket blame justified, and seldom is it helpful. By its very nature a forecast is an acknowledgment that the future is not perfectly known -- there's an element of uncertainty, of randomness. It's rare that a business operates in an environment stable enough to not need estimates.
2. Forecasting is not stocking. Forecasting and stocking are two very different activities.
Too many equate forecasting with stocking or deciding how much to have on-hand. Once a forecast is made the stocking decision, in most organizations, is automatic and invisible. The stocking decision is typically buried in the bowels of a computer system. It's out of sight so it's out of mind. We don't think about the stocking decision whereas we think a lot about the forecast.
3. Having the right stocking method is more important than having the right forecasting method. Think about the variety of stocking approaches, such as:
- by frequency (two weeks' worth, for example)
- the forecast error approach
- the lead-time error approach
- focusing on the fast movers
- having an item's cost, margin or "criticality" influence the decision
- classifying items into As, Bs and Cs and applying one of these methods to each class.
Actually the stocking decision is the only real lever manufacturers have to influence inventory performance. Many think forecasting is a decision process, but it's better characterized as a "best efforts" process -- we try hard to forecast well, but it isn't a decision. Stocking, on the other hand -- determining how much to carry -- is a real decision. You can make that decision independent of the forecast and in any way you wish ... as long as you're willing to accept the consequences.
4. There are "perfect" forecasts. Think about flipping a coin 10 times and predicting the number of heads. The forecast of five is a perfect forecast. Any other number would be worse -- it would be wrong more often, and its error rate would be higher. Moreover, if you knew that the number of heads came from the flip of a fair coin (as opposed to having to observe the historic pattern of actual heads and tails), you would forecast five no matter what that observed pattern was. Of course the number of heads from coin tosses will converge to 50%; but at first, observing the process, there would be fluctuations over and under 50%, leading to forecasts different than five heads in 10 flips.
5. Even "perfect" forecasts are often "wrong." Normally the number of heads in 10 flips of a coin is not five. Actually five comes up only 25% of the time, so 75% of the time the forecast of five is wrong. No matter, five occurs more often than any other number. It's still the best forecast -- a "perfect" forecast. Perhaps a more important point is that as long as the forecasting format requires that we provide one number as the forecast (virtually all forecasting systems work this way), then five is the only logical number.
6. Only minor improvements can be made by using the "optimal" forecasting technique. The "focus forecasting" approach has frequently been touted as nirvana. In this approach items are forecast using several methods every period -- every month, for example. Then the method with the lowest error is selected as the method to use. But the difference in the best method and the second best are trivial, with typical error rates within two or three percentage points of each other. Very little will be gained, if anything, by picking the "best" method. There's nothing permanent or intrinsically best about any method. "Best" is a fleeting concept. The real issue, and the one that business must deal with, is that you cannot do better than about 37% accuracy.
7. While you can forecast future demand, you cannot "know" it. Once you "know" the future, it's an order (or not). If you really know what will happen, if you really know that a customer will order something, you might as well call it an order and deal with it as such. It's great when this happens, but most businesses have an element of randomness to their order patterns. Randomness means you can't know what will happen.
8. The most important aspects of any inventory are the service it provides (fill rate) and the capital it requires. Nothing else is more important, not even forecast accuracy. The purpose of finished goods inventory is to be available for "random" demand. (If it's there for known, or existing, orders it's not really inventory -- it's committed, closer to being "in-transit" or even receivables.) You can achieve any performance (fill rate or availability) you want as long as you're willing to accept the consequences, i.e., having capital now and in the future tied up in dead stock due to previously overstated forecasts. So the benefit of inventory is fill rate and the cost of it is capital. Given enough capital you can achieve any fill rate you wish -- just stock enough. As explained in the following point, however, there are lots of choices.
9. There is an optimal performance -- one with the best service (fill rate) and capital. The dot in the chart below represents a specific capital and fill rate -- one that is being achieved. However, consider improving this performance by moving toward the lower right. It's obvious that you cannot be at 100% fill with zero capital -- it takes some inventory to provide a specific fill rate. As a matter of fact, you cannot even have very low capital and very high fill rate. So, as you strive to move toward the lower right, you get to a boundary, a barrier unable to be crossed. There's an "unattainable region." This boundary is the "optimal" performance level. It's the best fill rate for a given capital and the best capital for a given fill rate.
10. Most inventories require too much capital and provide too little service at the same time. Unless an inventory is performing at the peak, at some point on the curve illustrated in the chart, it by definition has too much capital and too little fill rate. From any point off the curve, you can move down and to the right, reaching a point on the curve and improve both fill rate and capital.
See Also