You learn from the past, but you live in the present
Capturing and analyzing historical process data is useful to understand what needs to be changed in the way you execute your operations.
Yet in terms of the present – i.e. real time – what we need to address is; (a) knowing if the correction we applied is working as expected, (b) receiving alerts when something is not behaving the way it should, and (c) being able to anticipate and react to undesired situations when they happen.
We have been working on a predictive algorithm of our own to determine if we can provide our clients with accurate and timely information on whether they are going to meet their SLAs or not.
We have trained it with a large volume of data from some of our projects and you can see a summary of the results below. The results are diverse and depend on the structure of the process, so that its application needs to be evaluated on a case by case basis, but in general we can see that the lower the number of variants in the process, the more accurate it is.
Here, the column “% Accuracy” is a measure of the accuracy with which we can predict that an execution will fulfill the SLA. The column “3 main variants %” represents the percentage of all cases in the dataset that follow the 3 most frequent variants, and therefore gives us an idea of how concentrated or disperse the distribution is.
There is a lesson we can extract from this, which although intuitive, is no less relevant: reducing or avoiding complexity (and errors are part of this) makes your process more manageable.
This brings us back to one of the fundamental principles of the classic quality control approach: less variation equals more predictability, which suggests we should focus our efforts in control.
We can also look at this from the perspective of the process variant: Is the SLA met executing the wrong variant? Is the SLA not met and we have not followed the correct route? Or are we following the right path but are not getting the expected result (SLA)?
There a are many options to turn your data into actionable results
You can test this out with your own data in your day to day…
Want to try? Book a meeting
Check some of our videos:
- New ➽ ICU discharges: where time slips away
- Fast Track vs Standard Path: Which is Better for Diagnosing Colorectal Cancer
- What causes discards in parenteral nutrition preparation? Key factors revealed!
- How Process Mining Uncovers Medication Administration Issues
The Prompter.io is our open project to share our experience—and that of others—in integrating language models and data-to-text techniques into process intelligence. Don’t miss it!
