Module: Trimming irrelevant history

Executive summary

The forecast algorithms give the best results if the input is as meaningful as possible; in this case, a reasonable sales history is a must.

One of the recommended modifications is to trim such sales history that no longer makes sense: either a trend in sales has changed dramatically or an average monthly sale significantly increased / decreased (i.e., hiked).

The module is also used with success for trimming sales around an introduction of a new product – the product is in fact sold considerably less (before the market accommodates the product), or, on the contrary, it is sold considerably more (especially if the product’s sale started off with a massive campaign).

These fluctuations in sales spoil the forecasts (unnecessarily) and a chance of an error in the forecast increases.

Functional description

Trimming principle

The trimming principle is based on analyzing patterns that are or are not fulfilled.

We have several algorithms that examine sales trends, extremes at the beginning of a data series, and algorithms examining an absolute change in sales after various cleanings.

One by one, all the formulas are applied to the sales data series and the sales history is then trimmed in such a way that the module Sales forecast receives the most relevant and the most recent input.

Using trimming in other modules

This module is but and instrument: it serves as a tool in other modules, as for instance in the module Trimming extremes and in the module Monthly sales forecast.

The trimming might result differently in these two modules, since if some of the sales are trimmed of its extremes, the input to the algorithm for Monthly sales forecast is changed.

In the module Monthly sales forecast, the trimming is used not only to trim sales of a single item, but it is always used to forecast sales for a group of items. Thus, even a shared history of a Top-Down or a Middle-Out group might be trimmed. For a single item, a part of its sale history might be irrelevant, but cumulatively, for the whole group, it is not irrelevant anymore: therefore, a cumulative history is not trimmed at all.