Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 DOI: 10.1515/orga-2015-0013 How Close to Reality is the „as-is" Business Process Simulation Model? Bartlomiej Gawin, Bartosz Marcinkowski University of Gdansk, Department of Business Informatics, Piaskowa 9, 81-864 Sopot, Poland, bartlomiej.gawin@ug.edu.pl, bartosz.marcinkowski@ug.edu.pl Background and Purpose: Business process simulation (BPS) model is based on real-life data form sources like databases, observations and interviews. It acts as "as-is" business scenario can used for reengineering. The main challenge is to gather relevant data and to develop simulation model. Research aims to elaborate BPS model and to systematically assess how close to reality it is. Design/Methodology/Approach: The research has been performed in Polish telecommunications company. Authors investigate technical process of expanding cellular network. After elaborating "as-is" model, authors use ADONIS simulation tool to run a series simulations and confront simulation results with actual historical events. After this, assessment whether computer simulation model can precisely map real-life business process - and consequently act as credible basis for process improvement - is made. Results: The simulation model has been constructed with data from the WfMS database, observations, staff knowledge and their experience. Fully equipped simulation model is found to allow reconstructing the historical execution of business activity with low margin for error. Some limitation were identified and discussed. Conclusion: BPS is not a popular approach for process reengineering and improvement yet. Data collection issues for BPS that require adopting process mining techniques and additional information sources are among the reasons for that. In our study, computer simulation outputs are compatible with historical events. Hence the model reflects the business reality and can be taken as a reference model while redesigning the process. Keywords: business process, business process simulations, data acquisition, workflow systems 1 Introduction Business Process Management (BPM) is an attention-attracting topic for management staff, enterprise modeling communities and scientists. BPM framework supports the design, enactment, control, analysis and improvement of business processes. The procedure has been designated by E.C. Deming and W. Shewhart as the PDCA method (Moen and Norman, 2009). The PDCA method can be used as a basis for practical solutions that are developed by business software vendors for supporting complex business processes management. Finally, the business process management cycle has been enriched by business process simulation (BPS) phase. This stage is usually supported by an IT simulation tools and is an important part of the evaluation of (re)designed processes (Tarumi, Matsuyama and Kambayashi, 1999; Suzuki et al., 2013). Computer simulations of business processes apply to both newly created processes and processes that are already in operation in commercial environments (Workflow Management Coalition, 2013). In the former case - design time analysis (Van der Aalst, Weijters and Maruster, 2004), simulation is mostly focused on examining abstract steady state situations called "to-be" scenarios, which is helpful for initial design for business processes but is still less suitable for operational business process execution (Rozinat et al., 2008). So, the a-priori simulation model consists of theoretical inputs, such as the shape of business process model, organizational structure and some parameters including activity costs and duration times as well as decision point probabilities, resources availability, etc. Because analysts have direct access to all mentioned theoretical inputs, they can explore different contrived scenarios with respect to the theoretical effect. In the latter simulation case - runtime analysis (Van der Aalst, Weijters and Maruster, 2004) Received: May 25, 2015; revised: June 27, 2015; accepted; July 10, 2015 155 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 business process has been commercially executed for a long time and this enactment is supported by information systems - Workflow Management Systems (WfMS). Usually, WfMS engine performs the workflow logic based on an implemented process model. Every commercial execution of the process model is called business process instance. WfMS performs process instances and, simultaneously, archives sets of information in a database (Gawin, 2009). Thus, produced event logs usually contain data about cases (process instances) that have actually been executed in an organization. Event logs record the times at which tasks were executed, the persons or systems that performed the task and other kinds of data. Such logs are the starting point for process mining (Van der Aalst, Weijters and Maruster, 2004) which means discovering knowledge about processes and discovering data that can power simulation models. While powering data comes from the workflow management system database, the simulation model reflects the "as-is" situation (Rozinat et al., 2008). One of the main challenges is to create simulation models that accurately reflect the real-world business process executions. For the "as-is" situation, both the simulated and real-world process should overlap as much as possible. The real business continuously needs process improvements to achieve better performance (e.g., better response times, less costs, higher service levels) (Van der Aalst, 2010). By modifying real parameters and performing various scenarios of simulations, analysts can estimate business results of the process time requirements and costing, staffing needs to be established, the identification of bottlenecks as well as calculation of resource loads, with which the company intends to carry out the process. With the use of simulation, the (re)designed processes can be evaluated and compared. Simulation provides quantitative estimates of the impact that a process design is likely to have on process performance, and so a quantitatively supported choice for the best design can be made (Jansen-Vullers and Netjes, 2006). The most popular information tools for business process modeling and simulation include ARIS, ADONIS and iGrafx, but the simulation experience is still limited. Some organizations - e.g. Wipro (Srivastava, 2010), Qwest (Teubner, McNabb and Levitt, 2008), Slovenian Ministries (Jaklic and Stemberger, 2005) and Motor Company (Hauser, 2007) provide case studies of projects involving simulation tools and a number of these projects have proved successful. This paper seeks to identify, systemize and elaborate implementable techniques of preparing a business process model based on the historical data, which truly reflects the real process behavior. The remaining part of this paper is organized as follows. Section 2 overviews related work on business process simulation types, tools and process mining techniques. Business case study along with research process is presented in section 3. Section 4 discusses data inputs for business process assessment and powering business process simulation model. In section 5 authors execute multi-instance business process simulation to compare simulation results with historical process outputs. Research is discussed in section 6, followed by conclusions. 2 Related Work 2.1 Business Process Simulation on PDCA Cycle Business process simulation integrates seamlessly with the PDCA cycle and can be performed in two stages: between process modeling (Plan) and execution (Do), as well as between process revision (Check) and practical improvement (Act). Figure 1 illustrates the proposal of PDCA cycle extension with BPS (Business Process Simulation) actions. Output of the PDCA Plan stage allows abstracting real business needs and representing them in graphical or/and Figure 1. Extended PDCA cycle 156 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 descriptive form (Becker, Kugeler and Rosemann, 2003). The most popular visual notations include BPMN (Object Management Group, 2013), eEPC (Scheer, 1992), IDEF3 (Mayer et al., 1995) and UML Profile for Business Modeling (Johnston, 2004). To perform the first BPS action (described as BPS_1), the graphic model must be enhanced with quantifiable data. At this stage, BPS_1 aims to achieve the vision and estimated values (Workflow Management Coalition, 2013) of the business process, which means theoretical (pre-execution) process costs, execution times, resource utilization, etc. In the process implementation/execution stage (Do), the business process is enacted on WfMS and outcomes are recorded in a database to be analyzed in the succeeding phase (Check). In the latter, experts, analysts and managers review executed processes and evaluate them with regard to strategy. They operate on real data - historical execution values (Workflow Management Coalition, 2013) from WfMS database and consider also comments from process participants. Collected data and information power the "as-is" simulation model (stage BPS_2) to "catch the reality" in digital form. By defining KPIs and creating business cases, analysts and managers identify the requirements to create and implement enhanced processes (post-execution optimization). But, before commercial launch of corrected workflows, the BPS_2 stage allows many "what-if" simulation scenarios which reflect planning business reality based on real historical data. BPS can be shown as a simple function: y=f(x), where function f presents a model that transforms inputs x to the result (output of that model) y (Bosilj-Vuksic, Ceric and Hlupic, 2007). By entering different x values, function f generates different y results and can be considered as "what-if" simulations. Another way is to conduct "what-if' optimizations by setting a target value for y, then searching for the values of x that result in the target value for y. Simulation allows to capture process dynamics and helps to investigate random variable influence on process development. Process simulation could play an important role in supporting business process change management approaches such as TQM, Just-in-Time, Business Process Re-engineering, Process Innovation and Knowledge Management (see Hlupic and Vreede, 2005). In this context, simulation can be used to investigate knowledge management processes, to simulate missing data needed for knowledge management, or to evaluate alternative models of knowledge management strategies. Simulation is used to describe a broad range of capabilities. These involve reproducing or projecting the behavior of a modeled system (Barnett, 2003). Models for simulation can be classified based on system of interest: a physical system (e.g. supply chain or production line), a management system (e.g. CRM or workflow process) or a meta-model (e.g. rules that establish whether a model is formulated properly). In 2013, Workflow Management Coalition (WfMC) has published Business Process Simulation Specification (BPSim) (Workflow Management Coalition, 2013). Authors stress how important it is to simulate and analyze business processes in a safe isolated environment before they are deployed and identify reasons for simulation and analysis still not being systematically used in most process improvement projects. Lack of mature standards for BPS in contrast to standards for simple process modeling is pointed out as the main reason. Framework BPSim introduces a standardized specification that allows process models to be augmented with data in support of rigorous methods of simulations and analysis. Provided meta-mod-el is captured using UML and supports both pre-execution and post execution simulations. 2.2 Process Simulation Types and Tool Support The main idea of business process simulation is to execute the model repeatedly to reflect the real business behavior. Contemporary IT literature distinguishes two types of BPS (Russell et al., 2005; Van der Aalst, 2010): Transient analysis and Steady-state analysis. In the former case, answers for operational questions (i.e. times, costs and probabilities predictions) in the near future are provided. When the transient analysis starts with initiated (and still not completed) process instances, then the model takes into account queues of work and temporary resources unavailability. For steady state, the initial state is irrelevant - the simulation model resets any cases in progress and it takes time to fill the system with tasks. Steady state is relevant for answering strategic and tactical questions rather than predicting the near future. Simulation types have become a practical rather than theoretical domain. Software vendors provide copyrighted simulation algorithms that differ regarding parameterization. Generally, process-oriented software falls into three types of tools (Jansen-Vullers and Netjes, 2006): business process modeling tools, business process management tools (BPM) and BPS tools. The former enable creating multidimensional process models in one or more available notations. As a result, static reports can be generated for process documentation, manuals, instructions, functional specifications. BPM systems can be perceived as tools that support managing business processes across the whole PDCA life-cycle (i.e. FLOWer). The core part of BPM system is workflow engine. BPM is defined as supporting business processes using methods, techniques and software to design, enact, control, and analyze operational processes involving humans, organizations, applications, documents and other sources of information (Van der Aalst, Ter Hofstede and Weske, 2003). 157 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 BPS tools support for process measurement and simulation, based on diverse techniques and providing broad range of outputs. BPS practitioners appreciate modeling flexibility, animations and simulations effects, statistical capabilities, variety of output reports and how user-friendly a simulation tool is. (Bradley et al., 1995) propose seven different categories to evaluate BPS tools, i.e. general capabilities, hardware and software features, documentation, user-friendliness, modeling-oriented potential, simulation capabilities as well as output analysis capabilities. 2.3 Process Mining and "as-is" Simulation Model With reference to Figure 1, it can be said that in case of BPS1 analysts need fictive inputs, mostly needed by business simulation model to design and predict future behavior. While performing BPS2, they need a simulation model which captures historical process shape and parameters, because "as-is" simulation models should reflect the reality as strictly as possible. In order to obtain useful simulation model to perform different process scenarios, process mining techniques extracting relevant data regarding processes from WfMS database are often used. WfMS operates with Business Process Participants in a client-server architecture, where the client side is usually a web browser (rarely a separate application) that can be operated from the employee's device. Based on the process definitions, workflow engine executes process models as flows of forms and documents that contain information regarding the tasks for employees. The aforementioned forms - Transaction Sets - are managed in accordance with defined routes. In commercial use, dedicated business processes are initiated on the engine several times, but based upon a single process definition. Running the same process numerous times means that workflow systems sequentially record process instances in the database. Process instances, recorded over the years in the event log, contain a huge amount of workflow data. Exploration of the data, also known as Knowledge Discovery in Databases, is a multi-step process that involves raw data transformation from the event log into actionable knowledge about the organization. In addition to the actual projection of the company's operations, the data may also be used to feed the simulation-oriented computer process models. The process mining method involves discovering knowledge about business processes, which are, by nature, dynamic phenomena. Business process analyses are reasonable provided that observations refer to attribute values not at specific moments, but over a long term. The discovery process involves the transformation of raw data into useful information which is used at a later stage to improve the systems and processes. The most valuable knowledge on organizations involves the hidden patterns, rules, trends and correlations in data structures, which are formed auto- matically over a long time period during data archiving in database systems. Discovery of these non-trivial relationships between attributes provides unique insight into the operation of the company, inaccessible using less refined methods of assessing business processes (Van der Aalst, 2007). Workflow mining algorithms allow detailed analysis to be carried out, taking into account four perspectives, each dedicated to discovering different aspects of process-related knowledge: Control Flow Perspective, Organizational Perspective, Case-Related Information Perspective as well as Conformance Checking Perspective (see Business Data Collection and Process Analysis section). Results provided feed analysis stage (Check), process simulations (BPS_2) and process definitions improvement (Act) to achieve improved workflow definitions. 3 Case Study 3.1 Business Domain Description The case study involves the workflow system which supports execution of business processes in the telecommunications organization in Poland. The investigated company provides mobile telecommunications services (voice, data transfer, internet) across the whole country and its internal structure is divided into four regions: Maritime, Mountainous, East and West. The analyzed case study was performed in one of them (Maritime Region). The investigated database comes from the workflow Action Request System (ARS) provided by BMC Software company from Houston (USA, Texas). The workflow ARS tool has been collecting instances of business processes since 2002. Because of big volume of recorded data in database, the event log investigated in our research reflects one-year activity. The instances concern the following business areas: planning, building and operating mobile telecommunication network and our case study concerns an event log which reflects three types of workflow processes: 1. Parameters Changing: setting telecommunication devices parameters, e.g.: transmitted radio frequency, radio transmission power, number of available voice channels; 2. Order Advice: transmission network modifications like radio link tuning, light pipe building and testing, equipment software updates, redirecting antennas; 3. Planed Work of Base Transceiver Stations : activities that require commercial service interruption to perform necessary modifications. Investigated processes relate to planning, operating and maintaining telecommunication services and physical in- 158 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 frastructure. All of them have been originated by personnel responsible for planning the telecommunication network and then executed by teams responsible for network operation and maintenance. Workflow instances flow between two departments: Planning Department and Executing Department. Workflows can also take place between teams within the department. Table 1 introduces data regarding teams of employees, which are involved in planning, building and maintaining telecommunication network. Relevant workflows can proceed according to different scenarios: employees of GPST1 can take decisions regarding the need for extension/modification of the telecommunication network. Then, depending on the set of tasks, projects can be expanded (by GPST 2) with more data, or can be passed directly to Executing Department, where projects are run and implemented. Both the employees involved in the preparation of projects and in their implementation can take advantage of specialized tools, which are usually purchased from suppliers of telecommunications equipment. Employees also use IT, measurement equipment and service cars to perform site surveys. During the flow of work, the employee uses workflow system to obtain information, documents and work description to perform assigned tasks. After completing contracted work, people forward process instance to another contractor. If the worker performs the last step of the process, then the process initiator checks the implementation of project and after this completes a particular instance of the process in workflow system. 3.2 Research Description Our research aims to elaborate guidelines which can help to answer the question how close to reality is the "as-is" computer simulation model. To develop our approach we use process mining - which enables discovering knowledge about processes, evaluating discovered information and powering the simulation model with data. We extend reality assessment by observations based on business process simulations (BPS2 included in extended PDCA cycle). After performing simulations (BPS_2) we indicate some additional techniques that help compare simulation process model with the real business process execution. The idea of business process management (BPM) provides continuous improvement of business processes, but this cycle requires useful simulations on models which capture the real business processes. In this paper, we skipped details regarding process mining algorithms, focusing on process mining results that can help assess the reality of discovered data. We also do not discuss the details of simulation algorithms. Both topics have been addressed in previous work (Gawin, 2009; Gawin and Mar-cinkowski, 2013; Marcinkowski and Gawin, 2014). Figure 2 overviews line of research. Research process involves a recurrent cycle, which, like the Deming wheel, provides for the continuous monitoring and improvement of business processes in companies. Workflow Management System is based on theoretical models of business processes and coordinates the execution of these processes in the enterprise. Process instances are recorded in the database and contain attributes reflecting the actual information regarding performed activities. These data include identifiers of personnel involved, durations of activities, invoked external applications and options selected in decision-making that influence the further course of the process instance. The database distinguishes subsequent instances by a unique number allocated by the workflow system on process instance start. Department Technical team Abbrev. Responsibilities Planning Department Design team 1 GPST_1 Estimating the deployment of telecommunications equipment in the field (base stations locations) and telecommunication equipment parameterization (base stations transmitting capacity, radio signals frequency etc.) Design team 2 GPST_2 Estimating the deployment of transmission telecommunication equipment in the field, (locations of radio lines, fiber optic), transmission parametrization (radio link transmission power, optical line attenuation etc.) and setting core parameters (e.g. for Mobile Switching Exchange) Executing Department Maintenance team 1 GUST_1 Implementing telecommunication projects (especially BTS part) Maintenance team 2 GUST_2 Implementing telecommunication projects (especially transmission part) Maintenance team 3 GUST_3 Implementing telecommunication projects (especially core part) Table 1: Technical departments and teams involved in the execution of business processes 159 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 Figure 2. Line of research Table 2: Data from -workflow database Discovered data Data type Data source Additional information Graphical model of the business process Processed data Workflow mining -Control Flow Perspective Based on hierarchically stored phases of each process instance, workflow mining algorithms reconstruct graphical process model Organizational structure Processed data Workflow mining - Organizational Perspective Workflow mining algorithms analyze database and discover who and how many times performed process activities Time-related parameters Raw data Basic statistic Basic statistic provides information how long every activity and every process instance has been historically executed; min., max and average statistics are available as well Number of business process occurrences within the chosen time period Raw data Basic statistic Number of occurrences comes directly from SQL query, limited to a specific period of time Probabilities for outgoing workflows from decision points Raw data Basic statistic Probabilities comes as a mathematical quotient of all running instances with respect to those affected by the selected path 160 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 Research begins with the Data exploration stage. At this stage process instances are explored from workflow management database. Then, we analyze them with process mining techniques which come from Eindhoven University of Technology. For this purpose, ProM (Process Miner) IT tool is used. Subsequently, we analyze mining results in different perspectives and assess how close to reality the discovered information is. Based on mining results, we construct an "as-is" simulation model for selected business process. Without any model transformation, we execute the "as-is" simulations to verify whether results might be helpful in assessing the reality of computer model. Business process simulation models allow to perform some process scenarios that enable us to predict the future behavior in a company. Simulation results can enable introducing changes in business process definitions (Figure 2: Business process simulation and optimization and then The proposal to modify business process models). In our research, further simulations involving modifying business process parameters were performed, yet remain out of the current paper's scope. 3.3 Data Collection Methods Simulation of business processes requires a process model interrelated with additional models (e.g. organizational structure) as well as powered with data regarding durations, costs, path probabilities etc. Most models and parameters can be discovered from the WfMS database, but some of them come from the "people knowledge" about the organization. A set of data to be collected depends on simulation tool, which shall execute "as-is" simulation model. Market investigation and evaluations led to selecting ADONIS Process Management tool provided by BOC. Research involved collecting data both from WfMS and from process participants. 3.3.1 Data exploration from WfMS database The workflow system records all events in the event log, from which the complete history of the process can be reconstructed. Seeking to achieve "as-is" simulation model for one of the available telecommunications company processes, the authors identify data and knowledge from database singled out in Table 2. All aforementioned data can be discovered WfMS database with ProM tool. Before performing analysis, data should be imported from database and transformed into the acceptable by ProM file format (see Figure 3). Required information regarding instances are transferred to MXML format using four intermediate Microsoft Access tables. MXML file contains many process instances and its attributes and enables performing workflow mining analysis to reconstruct the process model, as well as the organizational model of organization (Gunther and Van der Aalst, 2006). Raw data comes as ProM Basic Statistic and do not require any additional actions. Processed data are subjected to process mining algorithms, what leads to constructing business process simulation model. 3.3.2 Data exploration from personnel knowledge BPS 2 specificity (see Figure 1) requires additional information that cannot be collected from WfMS. Some of them are mandatory (as mandatory inputs for simulation algorithms), and some help to understand people's behavior and business rules in the company. Table 3 includes information that should be obtained from process participants along with techniques that may be used to obtain the information. Workflcw system database Intermediate Microsoft Access tables Figure 3: Simulation data preparation 161 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 Table 3: Information regarding business process and company provided by staff Information Collection technique Business process description from employees' point of view observations, interviews Official business process model and business activity description official documentation analysis, interviews Actual organizational structure official documentation analysis Availability of staff individual management interviews Number of hours per working day individual management interviews / official documentation analysis Number of working days per year individual management interviews / official documentation analysis Official documentation provides general overview of business processes. Additional documents include information regarding organizational structure, including "tree view" of departments, teams and employees along with full names and organizational roles. Generally, observation enables business analysts to access information that are not provided by any class of IT in a company. Elimination of intermediate links in the process of data collection which may contribute to increases in the probability of misinterpretations may be regarded as an advantage of observation. Research carried out involved combining diverse range of observation types (see Gray, 2013). Authors participated in business processes and observed rather non-controlled employees' behavior. Additionally, not all employees were aware of research conducted. Knowledge obtained with the technique enabled supplementing workflow management system with data regarding phone calls, mails and meetings, leading to executing process instances as best as possible. Research process involved performing interviews with management staff. Interviews are a practical alternative to observation as a method of collecting data without the use of IT support. They involve approaching respondents with more or less formal questions within a particular issue area. Interviews boiled down to the reciprocal flow of information and may be carried out using different procedures (see Sztumski, 2010). Researchers were allowed to perform face-to-face, unstructured interviews with management staff - rather groups than individual. It contributed to good understanding of actual business process instances in telecommunication company, ability to interpret research results properly and identify potential areas for optimization. 4 Business Data Collection and Process Analysis 4.1 Control Flow Perspective To achieve transparent results, the event log from the investigated WfMS database has been limited to a single business process - Order Advice. Because of a considerable volume of process instances stored, an additional filter was included to investigate cases limited to a single year. The initial workflow mining analysis - control flow perspective of a process - establishes interdependencies among activities. The goal of mining the perspective is to provide a visual, diagram-oriented presentation of all possible process instances historically executed. From a business point of view, managers responsible for workflow processes in an organization can review reconstructed diagrams and answer certain questions: How are the cases actually being executed? Are there any parallel executions? Are there any loops? What is the process model that summarizes the flow followed by cases in the log? ProM supports various plugins to mine the control flow perspective of process models. From available algorithms, three techniques were tested: Fuzzy Miner (Gunther and Van der Aalst, 2007), Alpha (Medeiros et al., 2004) and Heuristic Miner (Van der Aalst et al., 2007). Finally the latter one was selected due to the fact Heuristic Miner algorithm can deal with noise and incompleteness of event log. Additionally, the algorithm has options to focus on the main process paths instead of attempting to model the complete details of the behavior reported in the event log as well as wide parametrization capabilities. Heuristic net for Order Advice process (Figure 4) visualizes the reconstructed model that includes all the 71 instances (superposed). Every rectangle denotes a process activity and incoming/outgoing transitions indicate flow of work. Activities are assigned activities descriptions, identifiers of organizational groups responsible for execution, 162 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 Figure 4: Heuristic diagram of business process Order Advice. 163 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 activity status changes as well as the number of times the activity was performed. Both WfMS documentation and employees' statements confirmed that workflow system did not support parallel workflow executions, so multiple outgoing transitions from a single activity denote alternative flows. From the business point of view, lack of parallel paths caused numerous organizational issues during the process execution. To synchronize some parallel activities, employees used mails, phone conversations and meetings - what caused numerous relevant data to be processed outside the WfMS. The first decimal value assigned to transitions between activities indicates certainty level regarding existence of direct dependency between two activities. The value is always between -1 (not really sure) and 1 (completely sure). The second value provides information regarding number of flows through the transition between activities that were identified. The Heuristic Miner algorithm also enables testing the discovered process model. The result can be assess with the continuous semantics fitness (CSF) parameter. Based on discovered model, CSF algorithm parses actual process instance from the WfMS. Should any instance not precisely fit the model, algorithm indicates an error and continues processing. Errors correspond to Heuristic Miner discovery issues, especially in case the event log recorded noise, uncompleted instances and complex flows. The maximum value of CSF is 1 and indicates that model discovered from database 100% fits to instances from event log. In our case, the result of the parsing provided CSF value of 0,964 -Heuristic Miner coped well with the events log and almost perfectly mapped the instances from database into the visual model. 4.2 Organizational Perspective The process data can also be examined from an organizational perspective. It was the ProM Social Network Miner algorithm (Van der Aalst, Reijers and Song, 2005) that was found the most attractive for discovery of social network. The algorithm enables employee-oriented analysis; managers can observe who is mostly transferring work to whom, as well as who mostly begins, who ends the instances and how the work flows among performers. The organizational perspective provides also additional research that classifies people in terms of activities performed and organizational units. Organizational Miner (Song and Van der Aalst, 2008) is responsible for discovering and automatically grouping people carrying out similar tasks. Mined groups should coincide with the real organizational units from telecommunication company. Table 4 includes discovered organizational structure and activity assignment for Order Advice business process. Table 4 data indicates that in a single calendar year 19 people participated in 71 instances of business process. Employees executed a total of 897 process steps arranged into 14 coherent activities (ACT numbers from 1 to 14). Total number of process steps, in which employees were engaged and the percentage of employees' engagement in the execution of process steps were included as well. For the purposes of this study, the real names of the process participants were encrypted in ORIG.nr form. Based on the interviews with the staff, we have confirmed that the Organizational Miner algorithm properly organized employees into groups (based on similar tasks) as well as accurately assigned employees to activities - the matching was flawless. 4.3 Other Data Collection 4.3.1 Basic simulation inputs The ADONIS Process Management tool selected to support research process provides the following types of simulations: times and costs, accounting analysis, path analysis, capacity analysis as well as workload analysis (incorporating both steady state view and fixed time period) (more BOC Group, 2013). Each simulation algorithm needs some basic set of input data that allows initiate simulation scenarios. Table 5 summarizes the parameters required to carry out the simulations. The suggested technique of collecting values for the individual parameters was specified in each case; the upper part of the table lists the inputs that can be extracted from the workflow system database, while the lower part contains the data to be collected using other, non-IT techniques for gathering information about processes. Activity execution time, during which the current activity is executed, can be provided for every process step from ProM tool as basic statistic information. Alternatively, the metric might be extracted from Case-Related Information Perspective. Research is based on average activity execution times form ProM basic statistic across 71 process instances. Similarly, number of instances within the chosen time period can be reached in both ways. Also probabilities of initiating individual outflows of decision points come from ProM tool and enable assigning transition probabilities to all connectors leaving decision points (XOR logic). The sum of all transitions probabilities of the connectors equals 1. Cost parameters are considered optional for simulation and collecting them can be challenging in practice. Number of hours per working day serves in combination with the value in the field number of working days per year, in which instance can be executed. Process and staff calendars aren't taken into account in the capacity analysis algorithm that was used in research. Availability of staff specifies whether a performer works full-time or part-time in the business process. This parameter takes the value between 0 and 100% - the latter indicates full engagement in 164 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 Table 4: Organizational structure and task assignment for business process Order Advice No. Employee identity Organizational group identified by the algorithm Performed activity identifier Actual organizational group Participation in the process Quantity % 1 ORIG.0 minedGroupO 1,2,10 GPST_1 289 32.22% 2 ORIG.2 minedGroup4 8,4,9 GPST_2 156 17.39% 3 ORIG.3 minedGroup4 8,4,9 GPST_2 80 8.92% 4 ORIG.1 minedGroupO 1,2,3,10 GPST_1 69 7.69% 5 ORIG.7 minedGroupO 1,2,3 GPST_1 68 7.58% 6 ORIG.9 minedGroup4 8,4,9 GPST_2 62 6.91% 7 ORIG.4 minedGroup5 7 SYSTEM 53 5.91% 8 ORIG.5 minedGroupO 2,3 GPST_1 40 4.46% 9 ORIG.8 minedGroupO 2,3 GPST_1 20 2.23% 10 ORIG.11 minedGroup2 13,14 GUST_2 12 1.34% 11 ORIG.6 minedGroup3 5,6 GUST_3 12 1.34% 12 ORIG.14 minedGroup4 8,4 GPST_2 8 0.89% 13 ORIG.10 minedGroupl 11,12 GUST_1 4 0.45% 14 ORIG.12 minedGroupl 11,12 GUST_1 4 0.45% 15 ORIG.13 minedGroup3 5,6 GUST_3 4 0.45% 16 ORIG.15 minedGroupl 11,12 GUST_1 4 0.45% 17 ORIG.16 minedGroupO 2,3 GPST_1 4 0.45% 18 ORIG.17 minedGroupl 11,12 GUST_1 4 0.45% 19 ORIG.18 minedGroupO 2,3 GPST_1 4 0.45% Sum - 897 100% a single business process without any out-of-the-process activities. Capacity analysis algorithm enables simulation business processes while taking into account the corresponding working environment. That leads to the observation of workloads within the current organizational structure. Also the total execution time of all instances can be observed. Should the simulation model be based on reliable process behavior - as well as accurate organizational structure and other set of parameters (see Table 5) - then the simulation results reflect business scenarios that occurred in the past. So, results of capacity analysis can be taken into account while assessing the reality of "as-is" business process simulation model. 4.3.2 Algorithm parametrization Tuning-in the capacity analysis algorithm to meet research goals requires providing additional statistic from databases as well as some information from process participants (see Table 6). First parameter, Number of instances within the chosen time period, can be fed directly from heuristic diagram of business process, as it is assigned to initiating activity during mining process (see Figure 4). Special consideration should be given to Availability of staff, as assessing the value of the parameter proven a challenging task for management being interviewed. Consensus was met at 20% level. Regarding the Number of simulation, increasing the value of parameter improves the accuracy of the results while increasing the simulating device's CPU load. Value of 1000 allowed running simulations without significant delays. Participant-related information for simulation model are summarized in Tables 7 and 8. Please note that some activities in Table 7 - ACT_2, ACT 8, ACT12 - have no human resources assigned. This is due to incorrect WfMS deployment in the investigated company; the aforementioned activities represent periods of time when instances were awaiting next process participant (scheduled to perform succeeding activities in the process instance). Nowa- 165 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 Table 5: Data inputs for various simulation algorithms Algorithm Time and costs Analytical evaluation Path analysis Capacity analysis Workload analysis Technique suggested for collecting data Inputs Steady state Fixed time period Business process model X X X X X X Process mining (Control Flow Perspective) Working environment model - - - X X X Process mining (Organizational Perspective) Activity execution time X X X X X X Process mining (Case-Related Information Perspective) or basic statistic Number of instances within the chosen time period - - - X - - Process mining or basic statistic Probabilities of initiating individual outflows of decision points X X X X X X Process mining (Case-Related Information Perspective) Cost indicators (per activity) X (opt) X (opt) X (opt) X (opt) X (opt) X (opt) Interview / Observation Cost indicators (per performer) - - - X (opt) X (opt) X (opt) Interview / Observation Number of hours per working day X X X X - - Interview / Observation Number of working days per year X X X X - - Interview / Observation Process instance initiation calendars - - - - X X Interview / Observation Performers' calendars - - - - X X Interview / Observation Availability of staff - - - X - - Interview / Observation Number of simulations X - X X X X Scientist decision Table 6: Additional parametrization Parameter Value Number of instances within the chosen time period 71/year Number of hours per working day 8 Number of working days per year 260 Availability of staff 20% (for every employee) Number of simulations 1000 166 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 days, WfMS do not allow for designing waiting periods as separate activities - the system should monitor and store time between activities as transition attributes to estimate bottlenecks in the process instead. Total number of employees in the group comes from general knowledge of the process and determines the maximum number of resources of a given organizational unit (which theoretically can be involved in the implementation of business process execution). Number of people involved in the process is determined based on organizational perspective, while Avg. execution time may be established in process mining research as well as ProM basic statistics. Table 8 provides more detailed information from WfMS database that refer to the historical performance of the business process execution. Data include number of times each participant took part in the execution of a process step (see organizational perspective). Both summarized metrics and summaries excluding activities with no resources assigned are provided. Some differences between values provided in Tables 4 and 8 can be observed. While both refer to the same phenomena - number of tasks performed by employees - task assignment in Table 4 captures how many times employees recorded beginning/ending individual tasks by clicking start/stop buttons in the WfMS. Data in Table 8 reflects how many times employees actually participated in activities. Some activities involved one-click records (ACT1, ACT_7, ACT10) because this activities relate to the initiating/concluding the whole process instance. Remaining activities involved two-click records - at the beginning and at the end of the single activity. Process simulation algorithm accepts only the actual participations in the process. Aside from personnel, in many cases company IT is also engaged in performing certain activities. In telecommunication company, it was WfMS (codename ORIG.4) that performed ACT_7 as many as 53 times - the latter being a codename for closing the process instance (see Table 8). Cause for that was violating some business rules by employees from GPST_1 team - after completing the process, the person who initiated instance should verify whether ordered work was done correctly. After verification, the same person should close the processes instance. Since instances were rarely close manually, WfMS automatically concluded them after two weeks. 4.3.3 Construction of simulation model ADONIS process model. Activities ACT_2, ACT8, ACT12 representing waiting times for undertaking next activity in the process are modeled as notes indicating the place of their placement. The aforementioned activities are correctly implemented as additional time parameters for adjacent activities. So, the execution time of ACT_2 is entered as resting time for ACT1, while ACT 8 ACT12 are represented respectively as waiting and resting time for ACT11. Teams involved in the business process are modeled as pools. In accordance with Tables 4 and 8, each process activity was assigned dedicated resources as non-notational properties, which simulation algorithm can use for simulating process activities executions. 5 Business Process Simulation to Achieve "as-is" State After preparing a business process simulation model, the simulation was performed with capacity analysis algorithm. Simulation has been initiated as 71 process instances in accordance with explored data. Output generated by the simulation tool for aforementioned number of instances included total execution time of 01:071:21:30:18. Thus, business process simulation results may be interpreted as follows: executing the process 71 times requires a total of 1 year, 71 days, 21 hours, 30 minutes and 18 seconds. The result is an approximation of the historical executions of this process. To verify simulation results, actual initiation time of the first instance - that took place at 10:39, 20XX-02-12 - was explored from WfMS database. The end of the last instance has been recorded next year -on 3:15, 20XY-02-12. Based on real-life data from WfMS database it can be concluded that the actual delivery time for 71 process instances was 365 days. So, the difference between the simulation result and information from WfMS database is approximately 70 days. Table 9 outlines ADONIS simulation results that relate to the employees' involvement in the execution of each activity. The comparative analysis of simulation results (Table 9) against actual historical data (Table 8) allows to address the question whether simulation results of the business process coincide with the results that have been developed historically in the real-life implementation of the process. After collecting information regarding Order Advice business process, target simulation model is to be built. The control flow perspective provided a business process model which was adopted for ADONIS. Additional data - from organizational perspective/ProM basic statistics/personnel knowledge - enabled constructing fully parameterized simulation model and tuning it in to perform capacity analysis. Figure 5 illustrates heuristic net that was redrawn to 167 Activity Proj. preparation (radio) Project assign, (radio) Proj. execution (radio) Proj. preparation (trans.) Proj. execution (system) Proj.end (system) Proj. closed (wfm) Proj. assign, (trans.) Proj.end (trans.) Proj. closed (radio) Proj. execution (trans.) Proj. assign, (trans.) Proj. execution (object) Proj. end (object) Identifier ACT 1 ACT 2 ACT 3 ACT 4 ACT 5 ACT 6 ACT 7 ACT 8 ACT 9 ACT 10 ACT 11 ACT 12 ACT 13 ACT 14 Responsible organizational GPST1 - GPST1 GPST2 GUST3 GUST3 WfMS - GPST2 GPST1 GUST1 - GUST2 GUST2 group Total number of people in the 7 - 7 6 9 9 1 - 6 7 6 - 13 13 group Number of people involved 3 - 6 4 2 2 1 - 3 2 4 - 1 1 in process Avg. execution 00:00:02:36 10:03:38:49 00:00:09:19 03:09:25:37 00:00:23:25 09:12:06:19 00:00:00:00 02:00:14:00 13:07:22:22 00:00:00:00 00:00:26:22 104:03:03:5 00:00:00:48 14:17:09:10 dd:hh:mm:ss Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 Table 8: Number of activities performed by process participants as process-mined from organizational perspective Sum_1 without ACT: 2, 8, 12 Ti CO i> i> o m - - iL i> m LZ CS - - - - CS m m X rt m Sum_1 2 m rf m OZ O X t- o rf fTi Tf T X rt 1 H O A m m ACT_10 i> - X ACT_14 m o\ 1 H % m i> CS ACT_13 m ACT_11 - - - - Tf H O A o-, m o CS m CS r- ACT_3 o m - - rTf H O A m - Tf ACT_5 m - Tf 2 H C A - - - - Tf ACT_8 m m CS 2 1 H O A CO CO i> o m - - o ACT_1 CO ^o CS - r- Availability of staff (%) o (N o CS o CS o iN o CS o CS o CS o CS o CS o CS O CS o CS o CS o CS o CS o CS o CS O CS o CS Employees ORIG.0 ORIG.1 ORIG.7 ORIG.5 ORIG.8 ORIG.18 ORIG.16 ORIG.2 ORIG.3 ORIG.9 ORIG.14 ORIG.10 ORIG.12 ORIG.15 ORIG.17 ORIG.11 ORIG.6 ORIG.13 ORIG.4 Sum 2 169 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 i ! ■ J Hi tf As 1* Vs t 1 2 19 Si o 1' ■ s i i u C * , 1 - 5 § L W • [ •4 0 -0 1 ho « S § I-JH y J » L IS fl o < r-i » * W it * 0—1 il ■— o s? C o-l 2 h Figure 5. ADONIS model of Order Advice business process 170 Organizacija, Volume 48 Special Theme: Simulation Based Decision Making Number 3, August 2015 Table 9: Process simulation results - quantitative participation of employees in the execution of individual activities Capacity 0.004479 0.007439 0.007569 0.003733 0.002613 0.002986 0.00112 7.309601 7.336265 7.175558 1.607065 0.001056 0.002113 0.931486 0.608405 0.204052 o Total working time 00:000:01:51:48 00:000:03:05:41 00:000:03:08:56 00:000:01:33:10 00:000:01:05:13 00:000:01:14:32 00:000:00:27:57 01:120:00:47:39 01:121:03:53:10 01:113:01:01:56 00:083:04:32:20 00:000:00:26:22 00:000:00:52:44 00:048:03:29:54 00:031:05:05:47 00:010:04:53:09 00:000:00:00:00 Sum 3 m o m m o ao m g o TT ae m o - VO m m m m ACT_7 »n »n m m ACT_10 r- VO ACT_14 m m 1 H O A