تجزیه و تحلیل طراحی جریان کاری رسمی با استفاده از مدل سازی جریان اطلاعات

Workflow design has become a critical function in enterprise information management. However, only scant research attention has been paid to formal workflow design methodologies. As a result, existing design methods in business process management remain a manual and experiential effort and result in inefficiency in design tasks and potential errors in workflow models. Considering that there are hundreds and thousands of business processes in organizations worldwide, overcoming this deficiency will have an enormous technical and economic impact on enterprise information management. In this paper, we investigate the possibility of incorporating formal analytics into workflow design, thus alleviating the intellectual challenge faced by business analysts when creating workflow models. The workflow design analytics we propose helps construct a workflow model based on information about the relevant activities and their associated data. In addition, our workflow design approach also helps determine whether the given information is sufficient for generating a workflow model and ensures the avoidance of certain workflow anomalies. The significance of our study is to enable the transformation of workflow design from a manual and experiential effort into a more systematic and rigorous approach.

Over the last decade during the drive towards e-business, many organizations have realized the importance of Internet-based business process automation as they face the challenge of obtaining competitiveness in the global marketplace. To achieve high levels of business process efficiency, many organizations resort to workflow management systems to integrate business activities that span multiple functional units, such as human resources, marketing, and manufacturing [31] and [33]. As a prerequisite for effective workflow management, a workflow model is required to specify the execution sequence of activities needed to support certain business functions, such as producing a product or offering a service. Thus, it is critical to design correct workflow models efficiently according to specific business requirements. The existing workflow design approaches, such as the participative approach widely used in practice, mainly focus on collecting business requirements rather than providing a rigorous procedure for generating workflow models. The lack of formal design approaches often causes design problems. Recent empirical studies have shown an error rate of 10%–20% from over 2000 process models used in different industry practice, which include the widely used SAP reference model, the models used in a process reengineering project in the service sector in Germany, the models documented in the financial industry in Austria, the models used by consulting firms, and the models from IBM projects [20], [21] and [22]. With the empirical evidence of the design problems, the importance of formal design methods starts to be recognized. The process of designing workflow models can be partitioned into three stages. The first stage is business analysis consisting of design activities to understand business objectives, system environment, domain knowledge, and process management rules [37] and [38]. The second stage is process analysis that deals with the issues of activity identification and dataflow specification [14] and [32]. Activities can be identified by workers and managers in various business units since they generally know what they do on the daily basis. Input and output data for each task, i.e. dataflow, can be identified from documents, forms, and databases used in the business process [15], [27] and [35]. The third stage, namely model construction to derive the correct activity sequencing, is an intellectually challenging step. First, no clear guidelines exist to help determine whether a given set of activities and related data are sufficient for designing a workflow. Moreover, due to the human cognitive limit, great effort is required to figure out all interrelations of different steps in a complex process with no guarantee for the resulting model to be error-free [21] and [22]. These challenges are compounded by the fact that a workflow can involve multiple business units and different organizations located in different geographic locations. The common wisdom for dealing with potential errors in the stage of model construction has been to confirm the workflow models through simulation and verification before deploying them in practice [2], [7] and [17]. However, it is a fundamental principle in software engineering that design errors should be prevented as early as possible [21] and [23]. The later the errors are identified, the more cost and effort are needed to fix the errors. This principle also holds for designing and automating business processes [5]. Consequently, we believe that once a set of activities and their input and output data have been identified in the stage of process analysis, it is possible to derive a workflow model through analyzing the data dependencies among activities, namely dataflow analysis, so as to avoid certain potential design errors in the first place. Of course, this assumes the knowledge of all the activities and their input and output data can be obtained in the stage of process analysis given that the necessary information can be identified from interviews, discussion sessions, and documents by using the existing design methods [13], [15], [27] and [35]. This simplification makes it manageable to develop a first-cut workflow design method. The scope of our research is that starting from a dataflow to derive a corresponding acyclic workflow consisting of five basic constructs [1], [2] and [28]. A dataflow is specified with a set of business activities, the related routing constraints defined on the given business activities, and input and output data items for each business activity. In this paper, we focus on determining relations between business activities from the dataflow and then developing a control flow consistent with the dataflow. The basic idea is to decide how activities should be sequenced in a workflow through examining the dataflow in a process, i.e., the transformation from input data to output data via a sequence of activities. First, we define certain dataflow properties, including well-connectedness, completeness, and conciseness, to verify if a dataflow specification provides correct and sufficient information for constructing a workflow model. Moreover, we develop design principles for deriving potential activity execution steps, referred to as “activity relations”, from data dependency analysis. Third, we develop an analytical method to enable the creation of complete workflow models from activity relations. Fourth, we simplify workflow design by applying the concept of inline block and decoupling the issue of model correctness from the issue of workflow efficiency. Comparing with the existing workflow design methods that involve largely manual and experiential efforts, our method converts the design of complex workflow models from a task that depends on the knowledge and experience of workflow developers to a task that is supported with a systematic and mathematically rigorous approach. Therefore, we refer to such an analytical approach as workflow design analytics. The impact of workflow design analytics is significant. First, workflow design analytics builds the foundation for a formal approach to deal with the complexity of workflow model construction, thus alleviating the intellectual challenge faced by business analysts. Second, analytical workflow design can help verify the sufficiency of the given process information, including activities and associated data, for workflow model construction. Moreover, our approach can help eliminate certain potential workflow flaws caused by violation of data dependencies [30] and inappropriate model structures [7]. This feature of our approach will reduce the costs of generating error-free workflow models. As our research goal is to enable the transformation of workflow design from a manual and experiential effort into a more systematic and rigorous approach, the contribution of our research goes well beyond the computational formalism presented in this paper. Successful application of our approach will have considerable economic implications for companies with complex business processes. The remainder of this paper is organized as follows. Section 2 reviews the relevant literature. Section 3 presents the data dependency concepts needed for workflow design. Section 4 defines the principles for deriving basic workflow structures from analyzing data dependencies. Section 5 demonstrates how to generate complex workflow models step by step. Finally, Section 6 concludes the paper by outlining our contributions and future directions.

Designing workflow models is a critical step in business process management. However, existing workflow design methods remain a largely manual and experiential effort, and the quality of workflow design depends mainly on the skills of the workflow developers, thus resulting in inefficiency in workflow design tasks and potential errors in workflow models. In this paper, we have demonstrated the feasibility of injecting analytics into workflow design by mathematically representing data and activity dependencies. We have shown that workflow design analytics is nontrivial because of the complex interplay of data, activity, and routing constraints. In our research, we succeeded in formulating a workflow model via an two-phase procedure by means of several novel techniques, including the concept of activity relations in workflow modeling, inline blocks analysis for design simplification, and decoupling model correctness from model efficiency when creating the control flow. Our research is significant because it is the first workflow design approach that incorporates a formal procedure starting with dataflow specification. Moreover, certain potential workflow errors can be avoided at the time of workflow design, thus alleviating the burden of workflow verification to a great extent. Our approach is applicable when a dataflow specification is well-connected, complete, concise, and acyclic. We limit our focus on the five basic workflow constructs, i.e., sequence, XOR-Split, XOR-Join, AND-Split, and AND-Join [36]; other workflow patterns [28] may be added as extension to our research. Thus, we do not claim completeness in this paper. Since advanced workflow constructs such as the OR vertex (either Join or Split) can be simulated by a combination of AND and XOR [7], the design approach we propose is extendable to handle more advanced workflow constructs. In order to design workflows with loops, one approach is to first identify the main path and the feedback path [8]. We are currently in the process of investigating design mechanisms for cyclic workflows with OR-nodes. It is worth noting that there are a few factors that can constrain the potential benefits of our workflow design approach. First, our approach relies on the quality of dataflow specification. Even though our approach can be used to determine if a set of activities along with their input and output data is sufficient for generating a workflow model, we do not address how to collect dataflow relevant information, how to determine if a dataflow specification truthfully represents the data transformation of a process, and how to deal with incomplete information of activities and dataflow. Therefore, when the dataflow is not well-documented, the usage of our approach may be restricted. Second, when other considerations, such as resource limitations and cost optimization, also play an significantly important role in determining the structure of a workflow model, those considerations may confine the use of the proposed method and require some modification with the workflow model resulting from our approach. In future research, we intend to extend our work mainly in two directions. First, we will extend our initial research results to more complex situations and develop practical guidelines on workflow design. Second, in order to evaluate the effectiveness of the workflow design analytics, it is important to either conduct laboratory experiments or examine the proposed method though real-world applications.