How does the Supply Chain make its forecasts ?
What will the weather be like tomorrow? How many people will go to see the latest James Cameron movie next week? What color will be in fashion next year?
These questions are difficult to answer with precision!
Why? Because the very nature of the future is to be uncertain. As a general rule, we cannot know with certainty what tomorrow will bring.
But this does not prevent us from making assumptions, taking hypotheses, making plans and organizing ourselves accordingly.
In the supply chain, the first objective is to ensure the availability of products, services or resources at the time we need them and in the quantities we need. As some deadlines are unavoidable, some decisions have to be made days, weeks or even months in advance.
This is why one of the core activities of SC is demand forecasting. The challenge is to estimate in advance and as accurately as possible what the future demand will be, and to make the right decisions!
How does it work? How does the Supply Chain make its forecasts?
For decades, they have relied a so-called “deterministic” approach to forecasting. A deterministic forecast is a single number considered as the most likely future. Although this view of the future is only an average assumption, it is this single figure that feeds the whole supply logic: stock sizing, calculation of needs and planning of production, transport, team sizing, etc. Everything is designed to meet this “single” vision of the future.
Sometimes, this forecast is accompanied by an error measure to materialise the uncertainty. But this measure is never used in the calculations.
What happens if the forecast does not come true?
In fact, it is extremely rare for a deterministic forecast to come true perfectly!
There is a 50% chance of being above and 50% of being below. Therefore, according to the importance of the error that we will have committed, we will have to face either a shortage or on the contrary an over-sizing.
Obviously, supply professionals are very aware of this major limitation and they have very quickly developed workarounds that allow them to “limit the damage”.
There are two cases.
When the actual demand is higher than the forecast, there is a shortage. The main tool in this case is the so-called safety stock. In a simplistic way, it is a question of keeping a certain amount of stock “just in case” it is needed.
Sized on the basis of a statistical analysis of demand and past delivery delays, these safety stocks are more or less regularly updated.
When demand is lower than forecasted, appropriate actions are taken a posteriori and whenever possible to dispose of the surplus.
It is interesting to note that, in the end, only the safety stock allows the uncertainty to be partially taken into account through an analysis of the variability of past demand and the reliability of past deliveries.
What are the limits of the deterministic approach?
The Achilles heel of the deterministic approach lies in its vision of a future that is then considered certain.
In fact, in a deterministic approach, whether the future demand is 500+/-10 or 500+/-500 or 500(+200/-100), the supply will remain strictly identical since it only considers the average demand and totally ignores the uncertainty that accompanies it.
The only way to take uncertainty into account is through the safety stock. However, this stock only considers past behavior, generally over a long period of time.
If the uncertainty evolves over time, it will not take it into account.
If by a better control of a process or thanks to more reliable information, a company was able to reduce the uncertainty of its future, it would not change anything to its supply policy.
If on the other hand, due to an X or Y evolution, the uncertainty of the future would increase, there again the supply policy would not change.
Another limitation is that the safety stock only considers the uncertainty of the demand and the respect of the delivery dates. However, many other data are also uncertain. For example, the quantity to be received, the quality of the upcoming delivery and even the current stock levels are more or less incorrect in practice, as evidenced by the inventory discrepancies that occur every year.
Fortunately, there is a new approach to overcome these limitations and allow for much more reliable decision-making.
What is a probabilistic approach?
Probability is the language par excellence for describing uncertain information perfectly.
Far from being limited to the “most probable future”, a probabilistic forecast is a forecast that describes all possible futures, and their chance of occurrence.
For example, instead of announcing that tomorrow’s temperature will be 20°, a probabilistic forecast states that there is a 0% chance that it will be 16° or colder, a 10% chance that it will be 17°, a 14% chance that it will be 18°, a 18% chance that it will be 19%, a 25% chance that it will be 20°, and so on, down to a 0% chance that it will be 25% or more.
Based on this principle, it is possible to perfectly describe the uncertainty associated with any information, such as a delivery date, a quantity to be received, a stock level, the price of certain resources (material, hr, transportation), etc.
Once we have this perfect description of the future and its uncertainty, it is possible to evaluate the different supply scenarios and to select the most suitable one: reducing the risk, guaranteeing a level of service and/or minimizing a cost.
It is understandable that, because it is based on an exhaustive analysis of possible scenarios and allows for all forms of uncertainty to be taken into account, the probabilistic approach allows for perfectly informed decision-making, whereas the deterministic approach only takes uncertainty into account in an imperfect and incomplete manner.
Moving from a deterministic to a probabilistic approach. Which path for which contributions?
A recent study conducted by Vekia (to be published soon) based on the M5 competition dataset shows that, for the same forecast, the switch from deterministic to probabilistic forecasting leads to a 53% reduction in average SC costs.
Moving from deterministic to probabilistic forecasting represents a profound transformation of practices, tools and processes related to forecast generation and consumption. It is now a matter of manipulating probabilities where previously we manipulated a probable future.
The 4th generation of APS, now available in a robust and scalable way, allows to erase its limits by proposing a module that is simple to implement and use and compatible with existing ERP systems.