How To Forecast Seasonality

00:08 Introduction
00:25 What do we mean by core patterns when it comes to seasonality?
02:22 What are the difficulties people face when using seasonality?
06:52 How can we know the seasonality of new products?
08:22 If you forecast one month in advance, how can you know if summer is going to be elongated by an extra month?
12:07 How do you know what is seasonal demand and what is just a spike due to a promotion?
16:18 There seems to be many ways to have the data distorted. Is there any way we can have confidence in the results?
19:40 How can companies improve their approach to forecasting and take into account seasonality?


Seasonality is one of the major cyclical patterns that can be used to improve forecasting accuracy. Most supply chain processes tend to be seasonal to some degree, not only because of demand, but also lead times.

Christmas shopping, changeable weather conditions and even the SuperBowl are all examples of how seasonality can have a very real impact on a company’s sales figures. However, despite being one of most frequently used statistical patterns to improve the accuracy of demand forecasts, it is often a field that is widely misunderstood and difficult to get right.

In statistics, the demand of a given product is said to exhibit seasonality when the underlying time-series undergoes a predictable cyclic variation depending on the time within the year. In this episode of LokadTV, we try and understand how seasonality can be forecasted for new products, for which we do not have enough historical data to build more concrete predictions.

We then explore the main difficulties faced by supply chain practitioners when using seasonality: short lived products, the fact that seasons are not set in stone and so seasonality changes from one year to the other, as well as the patterns that may influence seasonality. All of these factors warp a company’s perception of their products' demand.

In addition, we discover the wide number of distortions that can occur to the data and how to take into account exceptional spikes in data. For example, how can we determine what is actually caused by seasonal demand, as opposed to a sudden spike due to a successful promotion?

We discuss how a transition towards a Deep Learning and Machine Learning setup can help companies to improve their approach to forecasting, to be able to better take into account this vital aspect that is seasonality. To wrap things up, we examine in more details what challenges these companies are likely to face when doing so.