In this blog post I am not going to survey all the methods used for forecasting: it would be too technical, too long and may be not so relevant to the reader. I will just briefly outline some of the basic ideas behind most forecasting methods and their relevance to business.
First of all (I will never stop recalling this…): forecasting methods are aimed towards the best possible prediction given the available data. It is no magic, it is science. It cannot deliver any result if there is no causal relationship between your future, unknown, data and some other, less unknown, data.
Let me give you an trivial example: you cannot use forecasting methods to win at the lottery. The result of the lottery, as well as any heads/tail-like game, is purely random. The outcome is not correlated with any other phenomenon and a lot of people wasted a lot of money assuming some predictability. No forecasting method will help you in this field.
However, in business…
But if, as I guess, you are in business, things might be really different. If you are selling, e.g., swimming suites or ice creams, or fresh beverages, the fact that this 2017 summer is so hot is relevant and the fact that in the next few weeks it is likely that the temperatures will rise again (oh, also this is a forecast!) lets you guess that your sales will continue, or even they will grow. This is commonsense.
What forecasting methods do is to go beyond common sense and implement the most advanced analytical techniques to get the most out of your data. By this I mean, in practice, finding and exploiting correlations: some are quite evident from the context of your business. Some others are far less intuitive and very difficult to find without a powerful data science engine, capable of looking inside massive datasets (financial, economical, demographic, weather related, …). Other correlations can also be found within your own data: there might be different products with similar sales patterns or, on the contrary, cannibalization effects by which selling one of your product will decrease sales of another one; or there might exist correlation in distinct geographical areas or in different sales channels, like e-commerce and retail.
Forecasting methods aim to exploit these correlations in order to predict what is predictable, in the best possible way, with all available data. Many classical and elementary forecasting methods, like those easily found on spreadsheets, just try to correlate future behavior with past history. This is already a very important step and by far not trivial: look inside the Intuendi demand forecasting solution and compare it with standard software: you will soon notice that there might be a big difference in the quality of forecasting even when it is based only on past historical data of a single product.
But of corse things become more and more relevant when multiple correlations come into play: correlations within a family of product, within a region, within different channels; correlations with external data; correlations with predictable events (holidays, planned discounts, promotions).
We at Intuendi are committed to deliver to you the most accurate forecast with all of the available analytical tools in order to deliver value to your business. We do not promise magic. Just accurate predictions.