3 key statements for supply chain demand forecasting


Well, let's talk about 3 statements that are obvious and over-simplified as self-evident truths about demand forecasting in supply chains.


Obvious statement #1 "Predictions are always wrong":


The first obvious statement is that forecasts are always incorrect. You hear this often, it is difficult to forecast something that is going to happen in the future. And why is it believed that forecasts are always wrong?   


Well, we can think of demand as essentially a continuous variable, i.e. a variable that can take an infinite number of values between two numbers. If I have some distribution with a probability of 90, then the probability that I pick a number and it has a value of 89.62397 is essentially zero. So, picking an exact point from a continuous distribution is very difficult. It is almost zero. So, the idea of having a correct predicted point all the time is improbable. It almost never happens. So yes, the forecasts are always wrong. 


Moreover, forecasts are highly disaggregated. Generally, in supply chains, it is not only necessary to forecast, for example, the quantity of blue blazers that will be sold in the whole country; you need to know the quantity of blue blazers that will be sold in Bogotá, and not only in Bogotá, but in each specific shop, and on the date or range of days (week) that the purchase will occur. So forecasts are highly disaggregated. Typically, a forecast is a combination of SKU, location and time. And, as it becomes more disaggregated, it becomes increasingly difficult to forecast, which leads to the forecast always being incorrect. 


The other thing to bear in mind is that every estimate has an error band. So, a forecast is just an estimate. And what are you doing with this estimate? You are giving a point with a confidence level or forecast interval. For example, an estimate of 89 plus or minus 2, so our forecast is between 87 and 91.


But what do you do about it? Essentially, the big point is: don't focus on, or obsess about, the forecast point value. Think of it as a guide and keep in mind that each point estimate carries with it an error band or forecast range, which allows you to start planning the safety stock percentages you should use in inventory.  


The main objective of demand forecasting is to reduce the uncertainty caused by demand, supply and processing variability. Forecasting enables strategic, tactical or operational objectives to be met, such as capacity planning, working capital investment strategies, branding plans, budgets, equipment planning, inventory planning, transportation and production planning.


Obvious statement #2 Aggregate forecasts are more accurate than disaggregated forecasts.


Our second obvious statement is that aggregate forecasts are more accurate than disaggregated forecasts. And this makes sense. But when we talk about aggregation, we talk about aggregating the forecast by SKU or looking at the forecast by a product type or by time, so looking at things over a certain period of time, a week, a month, a year, or doing it in a location. So the way I aggregate these factors together can improve my forecast accuracy. To validate this, we use a metric called the coefficient of variation (CV). Which is simply the standard deviation about the mean. This allows us to get an idea of volatility or uncertainty. A higher coefficient of variation indicates higher volatility. And the higher the CV, the more difficult it is to forecast demand and the less accurate the forecast will be.   


To better illustrate this concept let's imagine a café that has 3 sizes of coffee cups: small, medium and large. Of the large you sell 80 per day with a standard deviation of 30. Of the medium you sell 450 per day with a deviation of 210 and of the small you sell 250 with a standard deviation of 110. Now let's imagine that the lid for the three cup sizes is the same. Then you need 780 lids with a standard deviation of 239. This gives a CV of 0.31. This means that the volatility decreased. It is lower for aggregation using a common lid than the coefficient of variability of each size individually. 

Size Daily Sales Standard Deviation Coefficient Variation
Common Lid

Why is that? It is reducing the amount of total variability because we assume that some of the volatility of the large vessels is offset by some of the changes in demand in the small and medium-sized vessels. So these individual peaks and valleys offset each other. This is known as risk pooling. This reduces variability, facilitates forecasting and helps to reduce the safety stock. 


It could also be aggregated by time. What happens with weekly demand? Well, the coefficient of variation goes down. I can also do it on a monthly basis. The important thing is to match the time periods between the different actors in your supply chain. For example, if I can only order caps from the supplier once a month, then why do I want to manage it with a daily demand? Maybe I would agree to manage it on a monthly level. I can also aggregate it by location and it would have the same effect. In summary, when you do risk pooling you reduce the coefficient of variability and when you reduce the coefficient of variability you improve the forecast accuracy, but you should do it by looking at aggregations that really add to your supply chain and not do it simply because you have less variability in your demand.


However, doing all the combinations, granularities and variability calculations can be quite time-consuming and increases the difficulty of making forecasts. It is possible that spreadsheets can become so large that they start to freeze or that it is difficult to keep track of the forecast accuracy of each one. This is why there are tools that allow you to do this type of forecasting in an automated and fast way, which allow you to focus more on strategic analysis and less on data carpentry and formulas.


Obvious statement #3 Forecasts with shorter time horizons are more accurate.

The third obvious statement is that forecasts with shorter time horizons are more accurate. One example is that it is easier to predict tomorrow's temperature than a year from now. And it makes sense, because you have more knowledge about the short term, what is happening right now, than what might be happening a year from now. It also applies to shorter time horizons, say, half an hour is easier than a day ahead. Let's think again about the café, which also sells sandwiches and they need to get the sandwiches ready for the next day. They sell 8 types of sandwiches, so could they calculate the demand for each of them, with the variability involved? What if today they only ordered and cut the estimated amount of ham that is common in all the sandwiches and the next day two hours before lunchtime they start making the sandwich? What would they be doing here? Well, they are postponing. They are postponing the final consumption instead of making a quantity of each of the 8 types of sandwich, hoping to estimate well the demand for each one. Customisation is being done until the end. And this is another example of risk pooling and postponement. The idea is to postpone the final customisation of the item in the supply chain as long as possible to reduce variability.


These three statements simpleYes, they are true, and although often obvious, they should be kept in mind as they lead to actions you need to manage and mitigate in your supply chain. Also, identify how each of these affect forecasting, inventory planning, logistics and transportation and use your expertise and tools to manage them in the best possible way.

[mailerlite_form form_id=6]
Open chat
Chat with Datup
Hello 👋
How can we help you?