How “Outlier” can solve demand planners problems in IBP : Part-2
SHARE facebook



By: Krypt Admin - December 9th, 2020

Authored by: Abdul Sattar

In Part-1 of this blog we got to know what Outlier is all about in a layman term and demand planner’s problems. In this Part-2 of the blog, we will deep dive into usage of algorithm, how Outliers can be used and how to correct them.

When to use which algorithm?

Let Mathematics answer that Question –

Since IQR test is based on IQR value, which is the distance between Quartile3 and Quartile1, hence –

  • All the data points are not considered for defining the boundary
  • Most of the outliers are either near lower end of Q1 or closer to Q4 hence outlier has less Influence on it. – > IQR is robust to outliers.

Whereas on the other hand Variance is Non-Robust i.e., gets more influenced by the outliers and for defining boundary all data points are considered as it’s based on Variance.

Variance can be good method when data points are normally distributed but when data points are shifted towards one end it is advised to use IQR method.

Demand Planner’s Perspective

Demand Planner ran pre-processing algorithm for a product family having different trend and came up with following analysis:

  • Both Tests hold good when a product is in maturity phase as shown below

As seen here, both IQR and variance test turned out to be reliable case for detection of outlier.

  • When a product is seasonal, Variance test would be a better choice as IQR might see season as Outlier.

In case of increasing seasonal, Variance test turns about to be reliable as it detected all the outliers, whereas IQR missed a few.

  • When a product is having sudden change or has moved between phases in a Product Life Cycle – it falls under the category of paranormal distributed data (Opposite of normal distributed data), and it should be handled using IQR method.

Black Bars highlight the data points treated as outliers by Variance as it is not normally distributed and hence IQR is better method in comparison.
*Note: Both the Pre-processing algorithm doesn’t hold good for intermitted demand or highly volatile demand.
*Note: Pre-requisites to Outlier detection is substitution of missing values, if it is not run then system will treat them as outlier.

OUTLIER Correction

When creating a pre-processing forecast model for Outlier detection SAP IBP provided the following ways to correct them


Outliers are the anomalies in historical demand patterns, which can reflect different meaning from the entire data set. And hence, it is crucial for organizations to not just being able to detect the outliers automatically but correct them as well without human intervention.
Once the data is without outliers, they can be fed as input for statistical forecasting.

How can Krypt help?

Krypt is an SAP preferred partner and has assisted in the success of global businesses across industries & geographies. We have helped businesses through successful integration & implementation of SAP IBP, SAP GTS, SAP TM & SAP EWM.

If you are a global consumer products company looking for successful IBP deployment, and wish to explore more about our products/services, please request a DEMO or do contact us.

Follow us on:



Subscribe to Our Blog