Causal Process AI – the new way of process improvement

IBM Research, a partner in the AUTO-TWIN EU project, has released the first open-source library for Causal discovery in business processes, namely Causal Process AI.

The SAX4BPM library enables the discovery of causal process graphs from event logs. Causal process graphs are an important addition to process models because they account for the causal dependencies among activity executions as reflected in the event log. Understanding the cause-and-effect relationships between activity executions in a business process is crucial for process management and particularly for implementing effective process interventions.

What are process interventions in the context of causal process inference?

In causal process inference, an intervention refers to a deliberate action or manipulation of a variable, such as activity duration, allocated resources, or order reshuffling, to understand its causal impact on other process variables. Interventions are essential for estimating causal effects, as they help to distinguish between arbitrary sequencing, manifested by plain timing correlation among activities, and causal dependencies between activity executions. This distinction enables more accurate inferences about cause-and-effect relationships among the various activities in a process.

An intervention in a process involves actively changing a process variable, rather than just observing it. This is in contrast to passive process observation, where you only analyseactivity sequencing between process executions without controlling them.

Why is this important?

In causal inference, interventions are central to the counterfactual approach, which seeks to answer the question: What would happen to the outcome if the value of the variable were changed? The goal is to compare what actually happened (the factual) with what would have happened under the intervention (the counterfactual), thereby identifying the causal effect of the intervention. Applying the counterfactual approach to business processes can help answering questions such as: What would happen to the eventual outcome or execution time in a given process if a certain activity is expedited or postponed? This type of inference cannot be achieved solely with current process mining technologies, which only capture time precedence (correlations) in the data.

Let’s look at the flight departure process.

We know that although check-in always occurs before flight departure, a delay in the check-in is not likely to cause a delay in the flight take-off, unless you are the pilot. In other words, these two activities don’t have a causal relationship between them and shortening the check-in process typically does not affect the time departure.

By integrating Causal Process AI into business processes, enterprises can make smarter decisions and improve overall performance.

Ready to explore? Check out our SAX4BPM library here.