Process Mining Takes Some Beating. Is it Good After All?

I got this data set from a client’s IBM BPM platform. Great, just what you need to put process mining to a good test.

The file has records. Loading file was a no-brainer, I use Timeline PI, it needs , , and to connect events into stories (aka “timelines”), I added user name, category, type and other fields as attributes. They may help me understand why a process stops or slows down.

Now mapping looks good and Timeline PI is ready to dig in! It connects events into stories. For the file of this size it took about 20–30 seconds to load.

Timeline is an instance of a process which runs from start to end. When same process is executed multiple times each time it produces a unique story.

1. Configuration

Home page you see when data is loaded looks pretty much like a standard dashboard, but it has some unique quirks and features. For each of stories Timeline PI identified first and last events. Top section has summaries of process duration, event counts, time gaps and cost. You can set a tag or logo for each process type, that helps making timelines more visual.

2. Discovery

Now let the fun begin. Here is some highlights Timeline PI is ready to share straight away.

Note how Timeline PI shows the stories, it lines up activities along time, dots represent duration.

As I mentioned in my article here, process discovery is not set in stone but it follows a certain protocol:

Check if all steps are followed in prescribed order

See if there are no delays when steps are executed in order

Check if any activities fall out of standard execution plan

Identify patterns which can be automated

3. Quality

Definition of quality rests on rules and standards. Since there is no definition, I assume that process which follows same order a number of times is the standard process.

Look at these stories. I have fifteen instances, they belong to the same process category and have six steps. I can click on a timeline to check if steps are the same. What you see below is detailed description of each story with indication of date and time of activity, user name, type of activity, application name and BPM process snapshot id.

I can find exceptions by making these steps into a protocol.

Once protocol is defined, I can see what breaks the rule. Timeline PI offers choices based on patterns it could see over the data set.

Let’s check for missing step. If chosen, it zooms in to all stories where Request Processing is missing, now you can see which users were involved.

I can also create a dynamic flow to check how stories developed over time.

See how 5 processes fail to follow the standard protocol and go straight from Intake to Unexpected Outcome, I can zoom in to see the details.

From here I cannot tell why Sarah terminated the process, but description reveals enough information to bring it up to Sarah and Gandalf.

4. Performance

15 timelines defined as standard process look good in terms of order of steps, but what about time? Let’s check if we can add duration.

You cannot tell from example on the left if steps we processed on time, but example on the right adds process duration. You can see six steps where second step was executed much later than the first one. Now I will put standard and delayed processes side-by-side and see how they differ. To do this I need to build a query.

Based on most processes in this category I assume its ok to complete all steps within 24 hours. So I created a query to check for processes which take longer than a day to move to the second step.

The query brings back the subset of 5 processes, which I will use to compare to the standard flow.

I can do a few things from here, for instance, I can check how delayed processes compare across users or snapshot versions.

Below is how count of delayed (orange) timelines changes compared to standard (gray) by user name

or by snapshot version.

You can drill down to specific stories to see other attributes and understand why process stopped, delayed or who was involved.

5. Anomalies

Definition of anomalies comes hand-in-hand with definition of normal. It’s very similar to protocol violations, but I see anomaly as a broader term that defines whatever falls out of common behavior pattern.

If you remember, duration of all timelines in the set goes from 0 to 26 days 20 hours. However, if we check for 90% max duration drops to 19 days. We can treat remaining 10 percent as anomaly and investigate.

If I set selection to late-late running 10%, I can zoom in to 11 timelines below. All belong to

I can drill down to each step from here.

6. Fit-for-Automation

Besides Process Mining I do robotic process automation (RPA). Most common questions I hear from clients is “where do we go from here”? Many customers struggle with making a proper process definition, but process mining can help understand which steps consistently follow same pattern, therefore could be automated.

It takes a few clicks to see most suitable candidates. The flow on the left seems to be ready for automation, it has 6 steps and it covers 14% of all processing volume, which means automation can yield substantial benefits.

7. Conclusion

You should be able to tell by now if process mining exercise was a success or not. In my eyes, it’s a quite mature technology which sets foundation for long term improvements. You can use it to explore all sides of business, I will definitely continue my research, stay tuned for more stories, next time I will show how Timeline PI creates machine learning models and forecasts.

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store