Business Process Improvement (BPI) initiatives are concerned with identifying and implementing new strategies for conducting business operations in a more efficient manner. These initiatives typically employ various process analysis techniques, including evidence-based analysis techniques such as process mining, to identify new ways to streamline current business processes and to increase the efficiency of resources. Current practices in the application of evidence-based analyses of business processes are mostly focused on increasing process efficiency, thus overlooking the need to also optimize the way in which resources are managed. Fortunately, by exploiting detailed logs recorded in today’s information systems (which include records about the activities that were carried out, the employees who carried out the activities, and the time at which the activities were performed), it is now possible to learn the real behaviour of resources in terms of how they carried out business activities.
The goal of this research is to develop an approach to discover the manner in which resources prioritise his/her work using event logs. Such an understanding is important as it allows identification of existing problematic behaviours (if any) and their effects on the overall process performances. Most importantly, insights about resource behaviours will nicely complement existing process improvement strategy as it opens up opportunities to synchronize the way in which processes are designed with resources’ work prioritisation behaviours such that optimal process outcomes can be achieved.
The main question that this research attempts to address is how can we systematically learn resources work prioritisation behaviours by analysing event logs? In particular, we attempt to discover resources work prioritisation behaviours by learning the most suitable patterns of queue discipline employed by resources. A queue discipline refers to the manner in which customers are selected for service when a queue is formed.
As shown in the figure below, the most common queue discipline used in day-to-day life is the First-in-first-out style (FIFO), last-in-first-out (LIFO), priority-based (guided by certain prioritisation rules), and even random (i.e., no specific ordering strategy). The left-hand side figure shows an example of a `Purchase Order’ (PO) process which is made up of five activities: create PO, approve PO, modify PO, confirm PO, and terminate PO. The execution of a process leaves traces that are recorded in event logs. Our approach uses information in event logs to determine the prioritisation order employed by resources in tackling their work items. A work item is represented by the activity name and the case to which the activity belongs (e.g. [c1, aPO] represents an `approve PO’ task that needs to be performed for a case identified as `c1′. Resource B employs a FIFO discipline. Resource C employs a LIFO discipline. Resource F employs a priority FIFO queue because work items with the highest priority are executed first. In this case, activity `terminate PO’ (tPO) has a higher priority than other activities, thus are executed first, while work items that share the same priority (all three work items containing activity `create PO’ – cPO) are executed in FIFO manner.
We proposed an approach that can be used to discover the most likely queueing discipline employed by resources from an event log. The rationale of our approach is by firstly `reverse engineer’ events found in logs to reconstruct the list of tasks for each resource at any given point in time (i.e., rebuilding the worklist of each resource). Then, we observe the order in which work items enter and exit the resources’ work lists in order to arrive at a distance metric. A distance metric captures the deviation of the actual exit order of a work item from the expected exit order of the same work item, given a particular queue discipline. Finally, by plotting the distance metric over a period of time, we are able to determine the dominant queue discipline employed by the resource.
Our approach has been formalised and implemented as a ProM plug-in. Furthermore, our approach has been evaluated using synthetic logs to demonstrate its correctness and applied in a case study with an Australian insurance organisation. Findings from this case study that were deemed interesting to the stakeholders include:
- The revelation that the LIFO work prioritisation behaviour is popular in their organisation – a useful insight as they actually prefer FIFO prioritisation behaviour
- The influence of the way in which tasks are presented to users (on their monitors) on their work prioritisation behaviour
The outcomes of this research result in a research paper that has been submitted to a journal (currently under review):
S.Suriadi, M. T. Wynn, J. Xu, W. M. P. van der Aalst, and A. H. M. ter Hofstede. Discovering Work Prioritisation Patterns from Event Logs. 2016.(under review).