Forecasted Solutions

BLOG

Forecast Competition

Statistical modeling has been around forever, like most in the predictive forecasting world our forecast models compete for the best possible forecast, we use a variety of error methods (MAPE, MAD, MSE, BIAS) to determine the lowest error.

However, there are a few things that set us apart from the competition.

  • We do not limit you to one computer generated forecast, actually we recommend several computer-generated forecasts, more on that in a bit. Our base “computer” forecast runs at the lowest level, Customer/SKU. If you use base levels like Warehouse, plant etc that is the lowest level (Customer/SKU/Warehouse etc).
  • Each Forecast you can set the level it is forecasted at (With the exception of our base “Computer forecast” it is at the lowest level). You can add a forecast that is ran at the SKU level, set one up to run at one to many attribute levels. There is no limit to how you can generate new forecasts. Each higher-level forecast is pushed down to the lowest level based on our “Computer” forecast for distribution.
  • Each Forecast you can set the error measure to use in the competition, (MAPE, MAD, MSE, BIAS) as well as the time frame for the error checking (Full History, Last Full Year, Last Half Year).
  • So not only do our forecast compete for the best forecast, if you have more than just the base “Computer” Forecast then a new forecast is automatically added called “Optimized” Forecast. After EVERY forecast run, we look at the last 6 months absolute error for every combination of low-level forecast. The forecast that wins at the low level will be set as the Optimized forecast going forward for that low-level forecast. One SKU/Customer/Warehouse forecast may have the lowest error generated at the Customer level while another SKU/Customer/Warehouse has a lower error generated at the SKU/Warehouse level. In this way not only do the different statistical methods (Including our own proprietary ones) compete for the lowest error but our different forecast at different levels then compete for the lowest error.
  • Every forecast management page has an “Error” tab where you can go look at all the forecasts at the aggregation level you are viewing and see the Summed Absolute error at that level for each forecast. You can even drill into each period and see all the low-level items with their individual errors, and from there navigate to a new page showing the details. This allows you to quickly find and identify issues and quickly resolve them!