So, you want to forecast demand.
Imagine you’re opening your own business, a bakery and you’re making the best croissants in town.
It’s the first day and you’re just getting started. How many croissants do you need to make? It’s a hard question, because you don’t have any history, you don’t know how many customers are going to show up.
Nevertheless, a system of forecasting is necessary for any business. You need to figure out what customers are going to buy and forecasting is the way to get there. Imagine that it’s a year later and you are still in business making croissants, does it get any easier? You have some history to go on, but you still don’t know what customers are going to do.
Forecasting is an essential part of every business. Let’s define forecasting. It’s the prediction, we’re going to guess what demand will do. Estimation, we’re going to guess what level it will be at. And projection, carrying it into the future of demand. There are several types of forecasting approaches, what we’re going to use here is time series and time series relies on historical data. We have demand recorded over time from past until the future
and we’re sitting here at point t, and this demand data is going to be recorded in equal size intervals of time. So that maybe a day, it maybe week, it maybe a month or even a year.
So, we have this equal size periods of time and we are going to forecast in those same intervals into the future.
Our goal is to build forecasting model based on past demand, which we denote as d typically and we are going to forecast into the future and we denote that as F.
Once future data comes in, we’re going to continuously refine our forecasting approach and hopefully have a forecasting method that minimizes the error and gets us the best prediction possible.
Time series are mad up of several different patterns and noise, which I’ll discuss in a second. The different types of patterns that we typically see in time series are a base level of demand gives us a starting value, any trend that might persist. So, either demand goes up over time or it goes down over time.
Seasonality. Seasonality is a pattern that repeats every month, every quarter or every year.
Longer range cycles, everything that’s longer than a year, we call a cycle. It works like seasonality.
And finally, we have noise. Noise is random. It follows no rhyme or reason, but it’s always there.
A good forecast will try to resolve all of the pattern without trying to forecast noise. We cannot forecast randomness, so the best forecast does not even try. Before we get started with our different forecasting methods, a few words of caution.
The best forecast is not always the most complicated one. In fact, if you can have an equally good forecast that is more simple, it’s a better one to use.
Now when it comes to forecasting, it’s equally art and science.
We have scientific methods available, but using it right and picking the right one is equally important and it takes a lot of work to continuously monitor how different methods are doing and picking the right one at the right time.
A lot of people will tell you maybe that forecasting is always wrong, because you don’t always hit the number exactly that you predict.
But remember this, if you just get a little bit closer than before, you have an advantage. And if the forecast is right or better, then everything else in the whole supply chain will work so much better.
Therefore, forecasting is one of the critical pieces of supply chain management.
So, you want to forecast demand.