Root Cause Analysis in Manufacturing Companies – Tracking down errors

by Björn Heinen

Some defects in processes or products can have a strong impact on the profitability of a company. Quality problems can cause high costs in manufacturing companies, for example due to rework, rejects, or a lack of planning and adherence to schedules. It is therefore important to get to the bottom of the cause of the errors to be able to correct them promptly. A tool used for this kind of troubleshooting is root cause analysis. Its result provides actionable insights, and its goal is to prevent or eliminate the triggers of errors or problems instead of merely dealing with the symptoms. To do this, various methods and tools are used to determine the causes of incidents and quality issues.

Root cause analysis is required when problems can no longer be solved with self-service business intelligence. Especially in production environments with very complex process structures, a manual evaluation of the data coming from various sources is often too complex, time-consuming and requires partially automated analytics tools at a minimum. Let’s take a closer look at the methodology and its benefits:

Approach and basic principles

A root cause analysis is guided by concrete principles when analyzing the current state and determining the causes of an existing problem. The aim is to uncover correlations that have a causal relationship (causal discovery). Mathematical and statistical methods are then used to test specific hypotheses about the origin of the error. All invalid hypotheses are then gradually eliminated – for example, if there is a high correlation but no (direct) causality. A common example of causal discovery is shopping carts in supermarkets. For example, if 4,000 of 200,000 consumers have purchased diapers, 5,500 have purchased beer, and 3,500 have purchased both products, there is an 87.5 percent confidence that someone who purchases diapers will also purchase beer. Conversely, there is only a 63.6 percent confidence that someone who buys beer also buys diapers. Accordingly, the data suggests that the factors that lead to buying diapers influence the decision process of buying beer. However, this does not prove whether there is a direct, an indirect or no causal relationship – these are only indicators. All causal relationships within business processes must be systematically investigated and proven with concrete evidence. This analysis is usually done semi-automatically.

After all of the purely correlative relationships have been filtered out, the underyling causal relationships can be analyzed in detail. In the rarest of cases, there is a linear cause-and-effect chain. In practice, it has been shown that several nested cause-effect chains exist. In the above example, it is obvious that no customer arbitrarily thinks that beer is just great with diapers. Instead, it is reasonable to assume that the common causal factor is “young parents having a beer after changing diapers.”

Goals and benefits of Root Cause Analysis

Root Cause Analysis mainly supports the following goals:

  • Problem identification: it can clearly identify where the cause of a failure lies.
  • Development and application of solution strategies: By developing and initiating targeted solution measures, cost-intensive errors can be eliminated, and future problems and disruptions can be prevented.

By applying the solution approaches that have been developed, problems can be eliminated, and new difficulties avoided. Processes in the company run smoothly again. This saves resources and has a positive effect on profitability.

An example of a root cause analysis use case is a small batch manufacturer. This company manufactures engines and their parts. However, a relatively high percentage of them has errors during assembly, and the motors do not run within specifications after assembly. The root cause analysis is able to identify the error and significantly reduce costly rework and troubleshooting.

Closing Thoughts

Fortunately, many companies have now jumped over the hurdle to an open error culture. Root cause analysis forms an efficient tool to identify errors and initiate appropriate countermeasures. To do this, correlations must be extracted from data, which can best be validated using causal models and in collaboration with the business departments. In addition, it is possible to catch a glimpse of the future: The causal model created can then be used directly for predictions.


This article was originally published on the INFORM DataLab Blog.

You may also like

Data Quality Disasters


How Machine Learning and Artificial Intelligence have (not) changed over time


Why Data Science Projects Fail


About the author

  • Björn Heinen

    Björn Heinen has worked at INFORM since 2017 in data science. As Lead Data Scientist, he deals with both internal projects, in which existing INFORM products are extended by Machine Learning functionalities, and external projects, which he manages from development to implementation and integration.

    All posts by this author

    More about the author at:

Our authors

Find all our authors at a glance!

All authors

Back to top