It is already a burden with the sales and demand forecast. Since the sales department is either too general with its statements or rarely makes them, the only option, at least for articles with a certain history, is to look at the demand figures from this past. But the only reliable thing about these figures is that the expected figures will not materialize. Someone supposedly clever is said to have once said: “Forecasting replaces chance with error.” People without experience in sales forecasting conclude from this that you shouldn’t go to too much trouble with demand forecasting, because a systematic sales forecast only leads to equally bad results in other ways.
However, the reality is quite different. What we commonly refer to and interpret as sales forecast values is actually only part of the forecast. A forecast always consists of three components: a statement about the assumed basic requirement, a statement about the required security of supply despite the uncertain statement about the basic requirement and, as a result, a required safety stock to guarantee the required security. The statement on basic requirements will be uncertain in the event of fluctuating demand and is of little use as sole forecasting information. What you can usually calculate precisely, however, is the uncertainty. Depending on the degree of uncertainty to be cushioned, this results in a lower or higher level of security. A good forecast therefore actually replaces chance with error. But while the former is random, the latter can be recorded statistically and you can protect yourself against it.
In practice, a required delivery readiness level is therefore specified for each item whose future demand is to be forecast. The basic requirement and the safety stock are calculated using appropriate statistical formulas. Many of us have already gone that far and still can’t get a proper grip on stocks and delivery readiness.
If you know where to look, it’s easy to identify the cause of the divergence between inventory levels and delivery readiness. Virtually all classic statistical formulas for forecasting basic requirements and calculating the necessary safety stocks assume a normal distribution of the frequencies of the various demand quantities. In practice, however, this “normally distributed demand” only exists for a few items. This means that the classic formulas often miss the mark and other, practical approaches are required.
One starting point is to measure the deviation of the forecast values from the actual consumption values, which are known from the past under consideration. The best forecasting method – the safety stock calculation is not even taken into account here – is then selected as the one with the smallest deviation from the actual requirements.
However, the approach of only considering the forecasting procedures and neglecting the safety stock procedures and thus the level of safety margins falls far short of the mark, although it is widely used. This does not significantly improve the quality of the results.
Furthermore, the approach of working with “distribution-free” forecast and safety stock methods, which are not confused by a lack of normal distribution, already applies here.
Only a dynamic simulation of the scheduling situation over a certain period of time in the past, typically 12 months, can really help. Here, the various forecasting and safety stock procedures have to compete against each other on the course of the past of each individual article. The forecast calculation is carried out using the combination of methods that would have performed best for an article in the past.
For such a simulation, it is first necessary to clarify what it means to say that a process combination has performed “best”. Although it is common sense to rely on the forecast deviation again, it is still wrong. Corresponding simulations show that neither the lowest total costs of an item nor the lowest stock levels can be achieved while ensuring the required delivery readiness. It is more expedient to focus on achieving delivery readiness with the lowest possible inventories or the lowest possible total costs.
Finally, it should be mentioned that the simulation approach for testing the suitable combination of procedures must of course also simulate the entire scheduling process. But we cannot and do not want to go that deep at this point.
You already knew that forecasts are difficult, especially when they concern the future. Now you also know that forecasts require a certain amount of effort, especially if they are supposed to predict the future.
