AI-Powered Root-Cause Analysis

Challenge

When you bump into an anomaly in your dataset during a process analysis -- for example, the Order-to-Cash (OTC) process is taking too long -- you know it’s just the beginning of your work. You need to dig deeper and learn what is behind, why it happens, which specific attributes (suppliers, plants, materials, etc.) are causing it.

In other words, your goal is to identify the combination of attributes having the most negative impact on the problem.

Example

Let’s say you would like to get to the root cause of the lengthy OTC process where the average case duration is 46 hours. The dataset also provides information about case level attributes such as supplier country, supplier city, material, total amount, and cost center.

How can you find out which of these attributes, or a combination of them, causes the process to take so long?

Solution

To get the answers, run the AI-Powered Root-Cause Analysis, which is based on intelligently selected Machine Learning algorithms.

With the Root Cause Analysis, you will be able to compute the combination of the influencing attributes and the root causes of faults and problems your process experiences. It helps you uncover hidden relations between attributes.

Example

By looking at each value of each OTC process attribute separately, you'll see that the highest influencer of the OTC process duration is when the supplier city is Graz. On average, it increases the duration of the case by an additional 15 hours. This initial analysis also tells you that the other values of attributes influence the target metric far less.

However, when you start splitting the Root-Cause Analysis tree, you’ll notice that the Graz plant is performing poorly only when the material attribute is aluminum, while performing above average with every other material.

Application

The steps to run a Root-Cause Analysis are simple:

• Select any metric that you want to explore (even custom-created, boolean or string metrics)
• Select some case level attributes or custom metrics which you think can influence the metric
• Start analyzing and continue splitting the dataset to get to the root cause
• If you feel the default split is not right, you can adjust it as you see fit

The AI algorithm computes a tree-like structure where each node makes the best statistical split of the dataset into two smaller parts, based on correlations.

For better visualization, horizontal and vertical layouts of the tree are supported. As the root-cause tree can get quite large, a diagram preview will allow you to zoom in and out, so you don't get lost.

Control

Root-Cause Analysis, following a process discovery phase, helps you investigate operational issues on many levels objectively, based on hard data. It also allows you to convert all the findings into filters for further analysis of particular cases.

Now you’ll know which plants are triggering the longer cases duration, which suppliers are causing delays, why some decisions are stuck in rework. Such an analysis will later create the basis for your process improvement initiatives.