All of us involved in the problem of demand forecasting must sooner or later encounter the phenomenon customarily referred to as the Calendar Day Effect. Its definition seems simple and usually means a periodic increase in sales occurring before and/or during a certain date in the year and usually a decrease in sales after that date. Examples of calendar days include. Public holidays such as Christmas, Easter, customary “holidays” such as. Valentine’s Day or Women’s Day as well as annual shopping events such as. Black Friday. Depending on the industry in which we do business, we may be susceptible to these or other calendar days. Since the aforementioned calendar days occur every year, in principle, we should not be concerned about periodic increases and decreases in sales, since they are highlighted in the sales history, based on which we forecast. Well, the above statement turns out to be true only for those events that actually have a permanent place in the calendar. However, there are quite a few holidays and annual trade events whose occurrence changes from year to year, for example. Easter or Corpus Christi. Forecasting products susceptible to the effect of the aforementioned calendar days is already quite a challenge, especially if we have thousands of products in our portfolio spread across multiple locations, and each product group behaves differently. This is because it is difficult to imagine looking at the sales history of each product and manually applying the manually calculated adjustments to the current forecast to account for the effect of calendar days.
Certainly, a more attractive approach for both time (forecast preparation time) and quality (forecast accuracy) reasons is to use regression models to generate a statistical forecast. These models are very sensitive to the data they must be fed. Therefore, in the case of modeling the effect of calendar days, it is necessary to prepare not only data describing the calendar due to the dates of occurrence of the said holidays or events, but also information, for example, on the timing of periodic increases and decreases in sales relative to these holidays or events. While the proper description of the data is a relatively time-consuming activity, fortunately it can be done basically once a year. Done once and done well, it will allow us to avoid the risk of falling into out-of-stock situations during periods of increased demand and overstock during periods of decreased demand.
If you would like to learn more about how the use of machine learning and artificial intelligence can help you improve the accuracy of your forecasts, contact us. We will be happy to help your company.
The main feature of the model based on the so-called. “ordering point,” also known as an information-level ordering system or continuous review , is a condition […]
According to Gartner, the level of automation in supply chain processes will double in the next five years. At the same time, global spending on IIoT […]