Forecast via Clustering

Forecasting in SAP is not only devoted to Advanced Planning Solutions that wear an unaffordable price versus small and medium companies.

XSBS can also help you so doing, bringing experimented concepts of forecasting, straight in Excel, connected to your SAP demand management (Transactions MD62).

Basically when you do forecast you first have to consider historical data to perform a cleansing of outliers and missing values. To do so you may call for manual maintenance of data with graphical help. That is cumbersome time consuming and fragile, although this method is frequently adopted. Better if the solution can do this automatically, indeed.

Secondly you must select and adapt your forecast algorithm with parameters like alpha, beta, gamma for exponential smoothing etc. That means you know which algorithm best fit your case. This method is often adopted, but requires high skills in statistics.

Thirdly you have to designate what algorithm is to be paired with which object, say product here below, in something like a table so that you can launch calculation for a group of product together, not only one by one interactively.

The fourth way, available in XSBS Forecasting application, consists of a Clustered Forecasting technic. The base idea is to create a dynamic classification of your object to forecast, say product in this example, along given indicators.

E.g.

Sporadicity of sales like Sporadic, Intermittent, continuous demand

Lifecycle of the product, like launched, mature, EndOfLife

Volatility of the demand like high medium and low.

A cluster is a combination values above like : Sporadic and Mature and highly volatile. Therefore at any point in time each and every product is in a cluster based on its own activity

Imagine now you assign a statistical model against each cluster, it means your forecast calculation will automatically select the proper algorithm for your product depending of its activity. Imagine a product was in launch phase being quite variable, then has grown up to maturity with a fairly stable demand. With clustered forecast, it will automatically switch to more appropriate models at the same time changing of cluster.

In this concept you only need to initialize once clusters with appropriate forecast models and then use! Forget considering statistic, unless you denote your forecast are not good enough.

That is clustered forecast! That is available in XSBS Forecasting application! Check here above the ribbon options.