On the difficulties of deeper process mining

Process mining techniques tends to concentrate on the events and events chain, however all manual events have duration. Even automated process steps have one, but in most use cases you can use zero duration. Why should we measure event duration?

1. To access and optimise the resource usage or quantify the cost of process loops or less effective process variants. This is one of the top "quick wins” for process mining projects
2. Decrease the process step/process fragment execution time without process change. Process model created by process mining tools will let you go further compared to isolated steps analysing step
3. Access the effect of GUI updates/software changes and improve employee productivity. For SAP users, this will include migration to Fiori, Screen Personas or creation of new custom Z transaction. 

 In my practice, I’ve encountered three approaches to process step duration measurement:

1. Continuous monitoring of each process instance progress based on traces in IT systems
2. Manual process steps duration measurements done at regular (or irregular) intervals on the sample of employees/roles
3. Generic automated/manual steps separation with some high-level estimate on manual step duration

First approach allows for employee performance optimisation. You still can use data from sampled manual measurements to look for improvement options, but you should allow quite a big margin for the lack of accuracy and quite likely such approach is already implemented in your company.

Measuring steps in each process instance enables for effective research tools such as step duration distribution analysis to find persons/business units to concentrate the optimisation efforts on. Monitoring the steps duration dynamics allows to check if software changes implemented hit their target. With process mining tool, you can relate the step/process fragment changes duration to processing result or amount of resources spent on later stages of process.

New opportunities come at cost. Most information systems store data on events and event times, not event duration. You must deduce the duration using existing data via special algorithms such as Complex Event Processing. In some cases, the cost of development can be higher than cost of frequent manual sampled measurements and would not give you extra benefits.

Before diving into process step duration calculation, I try to access the situation checking for:

1. Show stopper: Does IT system(s) store timestamp or only dates (without time) of events?
2. Show stopper: Am I allowed to access personal performance by legislation of country? In Russia the answer is yes, however I am not sure about other countries.
3. Success inducer: What is the FTE cost and number of employees involved in process? 15000 “cheap” sales reps in bank branches can work as good target for optimisation based on step duration analysis, but 40 “costly” accountants in back-office - not
4. Success inducer: How accurate were recent manual process step duration measurements? Did step duration change a lot during the time, at specifics season, does business recognise the presence of high load hours? High variation of step duration makes limited duration sampling less reliable and leads to increase of “safety margin” in resulting KPI thresholds.
5. Success inducer: Is there the business case linked to the resource time consumption you cannot measure easily? Like time spent by top managers or time spent be employees in remote destinations?
6. Success inducer: Does business express concerns about the presence of key employees and/or key role work overload? It is important to ensure that perceived understanding of the situation matches the real life.
7. Success inducer: Are there issues with ergonomic of software in use? Most of the specialised user experience test tools works just fine with single screen analysis, however meaningful business steps usually involve work on multiple screens and come cycling between them? Rendering those tools less effective.
8. Success inducer: Does number of user errors in process tend to fluctuate and business agree that speeding up one step could lead to slowing of another one due to need to fix those errors?

If you encountered no show stoppers and enough of success inducers it is time to do analysis of source data and estimate process step duration calculation costs. Implementation of the step duration calculation an easy process and worth separate article, here I want to mention only fact: good old SQL/ETL platforms functionality will not be enough in most cases. If you can justify development costs by expected process/personnel efficiency improvements - proceed. 

I believe that process mining projects which include process step duration analysis are challenging ones, but can help business a lot. What is your experience?