The Pros and Cons of Task Mining

Task mining offers a complementary approach to process mining. It leverages computer vision to analyze user screen activity, providing a more detailed view of individual tasks than process mining, which relies on software event logs. However, task mining has limitations. It struggles to identify complex, end-to-end processes that unfold over long periods or involve multiple employees. Additionally, task mining results can be inconsistent, making it challenging to measure process improvement. While task mining is powerful for analyzing specific tasks, it may not be suitable for comprehensively understanding entire workflows.

Welcome to part 3 of an ongoing series of blog posts demystifying “process intelligence” - the terms, the technologies, and how and why process intelligence can help improve your business. In our previous blog post we discussed process mining, which analyzes software event logs to build a data-driven view of processes, but can miss critical process steps that aren't visible from event logs alone. For our third blog post, we will be looking at task mining: what it is, how it works and compares to other process discovery approaches.

As always, you can learn more about this and related topics by reading the Skan Process Intelligence Playbook here.

What is task mining?

Task mining, compared with process mining, takes a fundamentally different approach towards process discovery. Rather than analyzing back-end software logs, task mining uses computer vision to analyze the ‘front end’ of processes and tasks. Task mining analyzes processes by observing tasks in the same way that users do—by seeing them on their computer screens.

In recent years, task mining has gained popularity in large part due to an impressive increase in both the effectiveness and speed of computer vision algorithms, dubbed “The Industrialization of Computer Vision” by Stanford University’s Artificial Intelligence Index Report 2021. According to the report, the training time for computer vision algorithms decreased by 87% between December 2018 and July 2020. Also, their accuracy has improved, from 85% in 2013 to nearly 99% in 2020.

Task Mining tools apply these computer vision algorithms to screenshots of specific process steps, extracting detailed and specific information as data points that can be interpreted by an algorithm to build a process map. Because task mining can observe every action on the desktop, it can build a much more detailed view of a process compared with process mining, which is limited to observing only those specific actions which can be identified by software with backend logs. Task mining can be a powerful approach when organizations only need to know details of very specific parts of a process (‘tasks’). Additionally, because of task mining’s focus on front-end and UI, it has also held particular attraction for organizations looking to implement RPA-type automation, which is also focused on the front-end/UI.

While task mining overcomes some of the specific challenges present in process mining, it also comes with its own share of challenges. Most notably, task mining has difficulty stitching together all of the specific data points into longer-term processes.

Challenges with task mining

1. Difficulty Identifying Long-Term Processes

   While task mining tools have had success in accurately building a view of processes that take minutes or possibly days, it has had challenges in identifying end-to-end processes over longer periods of time as well as processes that involve multiple employees. In fact, this shorter-term focus has resulted in the name “task mining” itself—referring to the shorter-term, more tactical focus on “tasks,” rather than longer-term “processes.”

2. Limited Context Within a Full Process

   Because of the limited visibility task mining provides within the scope of a larger process, it is often not clear where the tasks being analyzed exist within the larger process (and the potential impact of those tasks on the rest of the process), the frequency of the tasks (particularly across different process variants), and the importance/impact of those tasks on the overall efficiency of the process.

3. Task Mining Is Not Scalable

   Due to the computational intensity of the statistical approaches used in task mining, its scalability is limited in terms of the length and complexity of the tasks being analyzed, as well as the number of employees that are part of the analysis.

4. Inconsistency of Results

   Because of the non-deterministic nature of the statistical approaches used in task mining, it often delivers inconsistent results. This means that the same tasks or process steps can be analyzed twice, but deliver varying results. This makes it particularly challenging to accurately measure the impact of process improvements by comparing pre- and post-implementation metrics using task mining.

5. Privacy and Security Limits Cloud-Based Task Mining Solutions

   Because task mining utilizes computer vision to analyze screenshots, for security and privacy reasons it almost always needs to be deployed in a customer’s own servers. In fact, cloud-based task mining solutions which need to send those screenshots and other potentially sensitive data to external servers are often a show-stopper in large enterprises.

Task mining's use of emerging computer vision technologies allow it to understand process detail not available to process mining tools. However, task mining also comes with its own limitations, notably its struggle with identifying complex, end-to-end processes.

Ready to learn more? Read the Process Intelligence Playbook

Modern process optimization requires a data-driven approach. That means considering all the technologies that can help you do that.

But, where do you start?

Skan’s actionable Process Intelligence Playbook explains approaches to process discovery, how to get started with process intelligence, and illustrates process intelligence in action through real-world applications.

Ready to learn more? Get the  Skan Process Intelligence Playbook  today.