Bottleneck and Resource Analysis on IT Help Desk with Process Mining

. This study utilizes data from the help desk log of an Italian company to create a model of the company’s business processes. The primary objectives are to show the benefit of the implementation of process mining to identify bottleneck activites within the process and analyze the workload distribution among resources. The research reveals that the most common bottlenecks occur during the transition from ‘resolve tickets’ to ‘closed’ accounting for 99% of cases, and another activity from ‘assign seriousness’ towards ‘take in charge ticket’ experiences bottlenecks in 91% of cases. Furthermore, a decrease in the number of cases was discovered after October 2012. Prior to this period, the average number of cases per resource was high, leading to a high average number of active resources per day and average number of events per day. However, after October 2012, the average number of cases per resource decreased by approximately 74.6% to 47 cases per resource. The average number of active resources also decreased by 25% to 3 active resources per day. Additionally, the average number of events per resource decreased by 40% to 3 events per resource per day. Regarding the resource workload, the analysis reveals that ‘value 2’ has the highest workload, having worked on 4,235 events. This is followed by ‘value 5’ with 3,748 events, ‘value 1’ with 3,028 events, ‘value 9’ with 2,073 events, and ‘value 13’ with 1,420 events.


Introduction
With the advancement of the digital era, numerous applications have emerged to assist the daily life, both on a personal and organizational level.However, it cannot be denied that no matter how sophisticated an application may be, it sometimes encounters obstacles, either due to human error or issues within the system itself.Hence, service providers offer help desk services.This service serves as a platform for users to lodge complaints or report difficulties encountered while using the applications they utilize [1]- [3].
The help desk will create a ticket for each user complaint.Based on these complaints, an assessment will be made regarding the seriousness of the issue, and the complaint will be escalated to the relevant department [4].It is possible that certain problems can be resolved directly at the help desk if their seriousness level is low.However, if the seriousness level is high, it may take some time to address the issue.This is common considering that help desks often receive numerous reports on a daily basis, which can lead to bottlenecks [5].
The case study in this research focuses on the ticket management of a software company's help desk in Italy.The event log, in CSV format, contains 21,348 rows, recorded from 2010 to 2013.It includes 2,1348 events, 4,580 cases, and 14 activities.Some data has been anonymized to protect sensitive information, such as the names of employees in the 'Resource' column, which have been modified and do not reflect their actual values.
The objective of this case study is to perform bottleneck analysis and resource analysis using process mining techniques.In the bottleneck analysis, we aim to determine if there is a backlog of ticketing in the help desk service and at which stage it occurs.Meanwhile, in the resource analysis, we seek to understand how tasks are distributed among the available resources, whether each person has an equal workload or not.According to Brocke [6], this research can be categorized at the individual level.

Ease of Use
There have been numerous studies on process mining.Process mining is a technology that helps companies understand their business processes.Intensive research has been done to discover various process mining techniques that can help in the automatic discovery of process models from event log data.There are three types of process mining: Discovery, which refers to the application of process mining techniques to create process models from specific event logs; Conformance checking, which compares an existing process model to the event logs used to create the model and calculate metrics to determine the quality of the process model; and Enhancement which improves process models by incorporating case information.[7,8].
Process mining is a discipline that aims to discover, check conformity, and improve processes by leveraging knowledge extracted from event logs.Event logs are utilized to discover process models, which can then be used for conducting congestion or bottleneck analysis.During the analysis of business processes, bottlenecks can be identified [9].
Bottlenecks generally occur when there are components within a business process that slow down or hinder analysis, thereby impeding the efficient identification and the understanding of the actual flow of the process [7].Bottlenecks can be considered as subprocesses within a system that slow down or halt the overall process.If bottlenecks can be addressed or improved, the overall process performance can be enhanced, leading to improved efficiency or cost reduction.[9].
Process mining can have different perspectives, depending on the subject of analysis, and one example is the organizational perspective.The organizational perspective focuses on information about resources in the log, such as the actors or individuals involved and their relationships.This is useful for determining models of role interactions, organizational hierarchy, social networks, and analyzing behavior patterns [8].In addition to identifying bottlenecks, resource analysis is another task that can be performed during business process analysis.This includes identifying, modeling, and analyzing the utilization of resources such as people, machines, or material at each process step.Through this analysis, organizations can identify potential barriers, allocate resources more efficiently, improve productivity, and optimize resource allocation to achieve better business goals.

Method
First, we identified the workflow of the ticketing system in the help desk service.Then, we modeled the event log using process mining techniques with Celonis Proactive Insight Engine [10], resulting in a process model.
Next, on the 'Process Overview' sheet, there is a 'Throughput times' tab that provides information about throughput time based on cases, as well as information about activities bottleneck with their respective durations from longest to shortest.Additionally, it indicates how many cases are affected by each activities bottleneck.
Lastly, we will conduct resource analysis.This analysis can be done by adding a new sheet and selecting the PI Social feature.The PI Social sheet will display activities, performance, and characteristics of the resources (referred to as users in Celonis).

Discovery of Process Model
The process modeling was conducted using Celonis.As shown in Figure 2, the most common path in the process consists of the following activities: process start, Assign seriousness, Take in charge ticket, Resolve Ticket, Closed, Process End.These activities are frequently encountered in each process.100% should be sufficient because bottlenecks usually occur in activities that are frequently bypassed.Meanwhile, the filtering connections can be kept at the default value to avoid a spaghetti process model (Figure 3).

Bottleneck Analysis
To conduct bottleneck analysis, we first identified the completion time for the entire process.As shown in Figure 4 below, the average time required to complete the entire process is 41 days.This indicates that bottlenecks are likely to occur in cases that have a completion time exceeding 41 days.From the filtered cases mentioned earlier, we conducted a process simulation in Figure 6.There are bottlenecks in the 'Assign Seriousness' and 'Resolve Tickets' activities.This can occur due to a high volume of incoming tickets, overwhelming the available resources.In such cases, it is possible to consider adding more resources, but this decision ultimately rests with the company as it would result in increased costs.Additionally, this bottleneck could arise because these activities are considered low priority by the management, leading to their evaluation only once a month or once a year.We can see the list of activities that were falling on bottlenecks on the Process Overview's sheet at the Troughput times's tab, as shown in Figure 7, where a list of bottleneck activities is displayed.As indicated in the previous process simulation, the Assign Seriousness and Resolve Tickets activities were falling on bottlenecks with a throughput time of 4 workdays and 29 workdays, respectively.Interestingly, these two activities impact 91% and 99% of the total cases in the IT Helpdesk process.

Resource Analysis
Before diving into the detailed part, we first look at the general overview of resource interactions.In this overview section, a dashboard is presented to provide a general picture of resource interaction information.In Figure 8, it provides information that on average, there are four active resources in a day.Furthermore, on average, each resource handles four events day.Each resource is involved in an average of 208 cases, and on average, each case involves three resources.There is a notable fact that at the end of 2012 and 2013, the number of user complaint reports were lower compared to previous years.This is evident from the lower average number of cases per resource during the months of October to December (Figure 9).However, this trend was contrary to the end of 2010 and 2011, when there was an increase in the number of user complaints from October to December.Based on these facts, we assume that the company may have performed updates or improvements to its software product around October 2012.
Looking at the pattern of cases per resource, it would also impact the average events per resource.At the end of 2012 and 2013, the average events per resource were lower compared to the end of 2010 and 2011 (Figure 10).When comparing the average number of cases per resource before and after August 2012, it is found that indeed the average number of cases per resource decreased after the software "improvement," resulting in fewer user complaints.This is indicated by a very drastic reduction in cases per users.(Figure 11 and Figure 12).Next, let's analyze the workload.In Figure 13 there are 22 resources recorded to have worked from January 13, 2010, to November 28, 2013 log).Looking at the workload distribution, Celonis calculates it based on the number of events.The top five resources with the highest number of events are: Value 2 with 4,235 events, Value 5 with 3,748 events, Value 1 with 3,028 events, Value 9 with 2,073 events, and Value 13 with 1,420 events.When looking at the details of the work done by Value 2, it includes the average number of events per day, the average number of activities per day, the average throughput time, last active, and who they received cases from and passed cases to.Additionally, from this resource analysis sheet, we can also determine the working hours of the resource and their workload during specific time ranges.

Conclusion
Process mining is a valuable tool for uncovering, comprehending, and analyzing business processes.In this particular study, an Italian IT Help desk company serves as case study to conduct a thorough examination of bottlenecks and resource utilization.The research findings highlight two primary areas where bottlenecks are most common: the transition from the 'resolve tickets' stage to the 'closed' stage, which accounts for 99% of cases, and the progression from 'assign seriousness' to 'take in charge ticket,' which experiences bottlenecks in 91% of cases.If the company has a larger budget, this might be overcome by increasing the number of employees, to be positioned in units that do the jobs that experience the bottleneck.
Furthermore, a notable decrease in the number of cases occurred after October 2012.Prior to this period, there was a high average number of cases assigned to each resource, leading to a substantial workload with a large number of active resources and events per day.However, starting from October 2012, the average number of cases per resource decreased by approximately 74.6% to 47 cases per resource.Similarly, the average number of active resources per day decreased by 25% to 3 active resources, and the average number of events per resource per day decreased by 40% to 3 events.In resource workload analysis, the study reveals that 'value 2' bears the highest workload, having managed 4,235 events.Following that, 'value 5' dealt with 3,748 events, 'value 1' with 3,028 events, 'value 9' with 2,073 events, and 'value 13' with 1,420 events.

Figure 1 .
Figure 1.Workflow of research

Figure 2 .
Figure 2. The Common process Figure 3. Process model with 100% activities and 91.7% connections Since we intend to analyze bottlenecks and resources, filtering the activity to

Figure 4 .
Figure 4.The average time required to complete the entire process Next, in Figure 5, we applied filtering in the Celonis tools to select cases with completion times exceeding 41 days.It was found that 48% of cases (2,203 cases) had a duration of completion exceeding 41 days.

Figure 5 .
Figure 5.The process that has average throughput time bigger than 41 days

Figure 6 .
Figure 6.Simulation of the process to see model to see the bottleneck's position

Figure 7 .
Figure 7.The list of bottleneck activities

Figure 9 .Figure 10 .
Figure 9.The trend of case per users

Figure 11 .
Figure 11.Case per resource before October 2012

Figure 13 .
Figure 13.Resource's workload by event count

Figure 14 .
Figure 14.The detail of resource's performance