Process Mining in 2021 and Beyond

In the past decade, process mining has become a mainstream approach to analyze and improve business processes. With hundreds of documented case studies and probably thousands of other success stories, process mining is now an integral part of the Business Process Management (BPM) discipline.

At the same time, process mining is a dynamic and rapidly evolving field, with the best yet to come. While the past decade has seen an emphasis on visualization and dashboarding (automated process discovery, performance dashboards, animation), the next few years will see process mining evolve into the realm of AI-based process optimization. I specifically foresee five trends coming of age in 2021.

Trend 1. Robotic Process Mining

Robotic Process Mining is an emerging sub-field of process mining focused on discovering repetitive work routines that digital workers perform in their daily work. Examples of such routines include filling in an online form with data coming from one or more documents or copying data from an email attachment into an internal information system.

The starting point of robotic process mining is a UI log: A recording of user interactions between process workers and various productivity applications and information systems. A robotic process mining tool analyzes UI logs produced by several workers over an extended period of time, in order to discover sequences of steps that are frequently repeated. These are called digital work routines. Each routine is analyzed to determine if it can be automated, for example via a Robotic Process Automation (RPA) bot or via an application orchestration script. The ultimate aim of robotic process mining is to relieve workers from non-value-adding routines, and enable them to focus on what matters for the customers. This short animation summarizes the goal of robotic process mining.

Trend 2. Causal Process Mining

Causal Process Mining is an emerging sub-field of process mining that seeks to discover and quantify cause-effect relations from business process execution logs. Such cause-effect relations may help process managers to identify business process improvement opportunities. For example, given an event log of an order-to-cash process, a typical causal process mining technique allows us to discover that when a customer is from Southeast-Asia, then assigning activity A to worker X or executing activity A before activity B (rather than the other way around) increases the probability that this customer will be satisfied by 10%. The goal of causal process mining is to identify interventions that make a difference in terms of specific performance metrics. Read here to know more about causal process mining in this short article.

Trend 3. What-if Process Mining

What-if Process Mining is concerned with using event logs to understand the effect of one or more changes on a business process. Traditional process mining techniques are focused on "as is" analysis questions such as "what are the current bottlenecks in the process", "where are the current sources of waste", where are the rework loops? In contrast, What-if process mining addresses questions such as: "What would happen if the number of customer orders doubles next month?", "What would happen if 10% of our workforce is slowed down by Covid disruptions?", "By how much would the waiting time or the process costs be reduced if we automate a given task in 90% of the cases?". The keystone of What-if process mining is data-driven process simulation: The ability to automatically discover high-fidelity process simulation models based on execution data. Traditionally, process simulation has been a highly time-consuming task, which requires highly specialized knowledge about probability distributions and statistical analysis. Data-driven simulation automates this task by automatically discovering, tuning, and validating simulation models, ensuring that these models faithfully capture the observed process. In the past couple of years, we saw the first data-driven simulation tools appearing, including the Simod open-source toolset, which discovers BPMN simulation models from event logs. As we move forward into 2021, we'll see surely many more developments in this field.

Trend 4. Prescriptive Process Monitoring

The same event logs that we use for process mining can also be used to train machine learning models to predict, at runtime, negative outcomes in a process, such as: Will a running process instance complete on time or late?, Will the customer be satisfied upon delivery of their order?, Will they return the products?, Will a supplier pay their invoice on time or with some delay? There are several tools to generate predictive process dashboards, including open-source tools such as Nirdizati. No doubt, there is a lot of potential business value in such predictions. They make it possible for operational managers and process workers to see the issues coming and to take preventive actions. In practice, however, a predictive monitoring dashboard is of little use on its own. Yes, it tells us that 10% of the open purchase orders will lead to a customer complaint of some form, for example due to delays or due to defective or incorrect products. But it does not tell us what should we do about it and when? How should we allocate resources to prevent such issues in a way that maximizes business value? When should we act upon a prediction? Prescriptive process monitoring is an emerging technology that exploits predictive process models in order to recommend actions (call them "interventions") that optimize a given process performance indicator. Prescriptive process monitoring is cost-aware: It optimizes the trade-off between the cost of executive preventive actions and their overall benefit, for example the cost of expediting or prioritizing a customer delivery, versus the benefit of delivering earlier for that particular customer (possibly at the expense of other deliveries). Prescriptive analytics technology has reached a high level of maturity in the field of e-commerce — think about the recommender systems used by major media sites such as Youtube. I foresee some of this technology making its way into prescriptive process monitoring engines in the coming years.

Trend 5. Automated Process Improvement (AutoPI)

Current methods for process mining are largely based on visualization and dashboarding. They require an analyst or subject-matter expert to navigate through a series of visualizations in order to identify issues in the process and possible ways of addressing these issues. A business process visualization may indicate to us that there is an issue (for example a bottleneck or a rework loop) but it does not tell us how to address it. Should we re-allocate or re-train resources?, Should be automate some tasks or some handoffs, and which ones?, Should we change the way a task is performed?, Should we do a task earlier or rather postpone it to later in the process?, Should we send a reminder to a customer earlier in the process?

Automated process improvement is a gestating technology to automatically explore a very large space of potential business process changes, in order to discover combinations of changes to optimize one or more process performance indicators, like defect rate, cost, manual effort and/or throughput. AutoPI is still in its infancy, but I would expect some prototypes and pilot cases to emerge in the coming year or two, with mature tools emerging in the mid-2020s.

Conclusion

The above are just some of the exciting developments we should expect to see in the field of process mining in the coming year(s). There are, of course, other challenges that will be tackled in the coming years, as discussed in the Industry Panel of the ICPM'2020 conference. We'll certainly see a few advancements to tackle technical challenges such as privacy-preserving process mining, cross-process analytics, and automated validation and enhancement of event log quality. We'll also see many developments ahead in the area of strategic positioning and governance of process mining, such as process mining maturity models, or the emergence of management concepts such as the "Enterprise-Wide Process Control Room", where processes are managed holistically and proactively, as opposed to launching process mining investigations reactively as issues arise.

So if you are into process mining, there's a lot more to come, and for BPM practitioners still hesitating to integrate process mining into their practice, it's an excellent time to do so, as you can take advantage both of the existing maturity in the field and of the many new developments to come.

Acknowledgments and License

This piece is written in my capacity as professor of University of Tartu. My research is funded by the European Research Council (PIX project) and the Estonian Research Council. I am also co-founder of Apromore - an open-source process mining solutions provider. I have tried not to be biased by this latter affiliation.

This article is licensed under a Creative Commons Attribution Generic License CC-BY 2.0.