ارزیابی طرح های فرآیند جریان کار با استفاده از انسجام و اتصال متریک

Building on the similarities between software programs and workflow processes, this paper proposes a heuristic that offers guidance for the creation and evaluation of process designs in administrative settings. Designers can use this heuristic to select from several alternatives the process design that is strongly cohesive and weakly coupled. It is argued that such a design will result in fewer errors during information exchanges and in more understandable activity descriptions. The paper includes an application of the heuristic in an industrial workflow setting, which supports its feasibility and practical value. The paper also presents the freely available CoCoFlow tool that implements the heuristic and its associated metrics.

In 2006, one of the world’s largest asset managing companies redesigned its business process to produce the annual reports for its investment funds. On the basis of a meticulous analysis of the essential information processing function of the process, a new process design was developed using the method of Product-Based Workflow Design (PBWD) [45]. The new design reshuffles the order of the original steps and exploits the opportunities of an automated, XML-based message exchange between investment managers, accountants, print shop, etc. It is anticipated that this new way of working will cut the cycle time of the process by half. Not only does this imply a better service to the company’s customers, e.g. corporate investors and pension funds, but also it facilitates compliance with deadlines from the financial authorities. This case illustrates that the combination of process redesign and the application of IT are viable today, even though Business Process Reengineering (BPR) was introduced over a decade ago [25]. Particularly in administrative settings, where the focus is on information processing (e.g. the evaluation of a damage claim, the issuing of a building permit, or the handling of a mortgage application), various redesign opportunities exist [39]. For example, by using electronic documents and workflow technology it is relatively easy to change the routing of a file along the various decision-making steps in a process. When doing this properly, the average execution cost of the overall process can be minimized [3]. However, coping with such design freedom may also be problematic. The main issue that we focus on in this paper is the proper size of the individual activities (or tasks) in a process. This design choice is known as the issue of process granularity [18]. Badly chosen sizes of activities in a process may negatively affect its performance when being executed or enlarge the maintenance burden of the process model in case of updates. Small activities, on the one hand, may increase the number of hand-offs between activities leading to an increase of errors [42] and [49]. Large activities, on the other hand, may become unworkable to be executed well by humans [18] and [42]. This paper addresses the problem of activity design in the domain of information-intensive processes, typically found in the service industry. This issue is particularly relevant within the setting of BPR projects [26] where also the application is being considered of workflow management technology [5]. In the remainder of this paper, we shall refer to such processes as workflow processes or simply as workflows. We will present cohesion and coupling metrics that focus on the content of activities, i.e. their operations. By using the proposed set of metrics, it can be quantitatively expressed to what extent operations “belong” to each other within one activity or, in other words, how cohesive such an activity is. In addition, it is important to measure to what extent the various activities are dependent on each other or, in other words, how much they are coupled. The inspiration for the proposed metrics comes from software engineering, where an old design aphorism is to strive for strong cohesion, and loose coupling. We build on work by Selby and Basili [50] and Xenos et al. [59], who defined coupling and cohesion metrics for software programs. The proposed solution based on comparable metrics fits conceptually with the PBWD method, as applied in the case we mentioned in the opening. PBWD is a BPR method which is adopted and actively applied by one of the Big Four consultancy firms [45]. For a more detailed application of the method in the financial services, see ref. [41]. In this paper, findings from our earlier work [43] and [46] are integrated and extended in several directions. A first version of the workflow cohesion metric was introduced in ref. [43], but it lacked some facilities. For example, it could not handle conditional alternatives to achieve the same output (XOR-construct), which is a construct very common in workflow processes [6]. Secondly, we introduced in ref. [46] an additional coupling metric, but the integrated design heuristic did not yet consider resource issues that in real life impose constraints on good designs. Thirdly, in our experiments we noticed that the manual application of the metrics can be time-consuming and may easily lead to human errors. Therefore, this paper introduces the freely available CoCoFlow tool that can be used for the automatic computation of the metrics. Finally, we applied the integrated design heuristic to the case of a Dutch governmental agency to demonstrate its feasibility and practical value. The structure of this paper is as follows. In Section 2, we introduce a motivating example that clarifies the goal of our research. Section 3 describes the cohesion and coupling metrics and the heuristic for selecting a favorable design among various alternatives. The fourth section contains an application in an industrial setting, which is followed by a description of the CoCoFlow tool in Section 5. In Section 6, we elaborate on related work in the fields of software engineering and workflow to position our contribution. The paper ends with a discussion and prospects for future work.

Our claim in this paper is that the proposed cohesion and coupling metrics will help designers in the creation of workflow models that have a better execution quality and that are easier to understand by people. Understandability in its turn will cause a better maintainability of the workflow model. Our overall claim is based on the following assertions: • The heuristic leads to intuitive outcomes. In our earlier experiments, i.e. the IBG toy example (Section 2), and an industrial case (Section 4), design alternatives are selected that incorporate activities which are neither too large nor too small. Such a balance is necessary because a great number of small activities will lead to a tightly coupled process with many error-prone hand-offs, while activities that are too large will lack cohesiveness and cannot be understood or performed well by humans. Involving a small group of experienced workflow designers in our research, we already showed that the cohesion metric conforms with expert decisions (see Section 6). • Similar heuristics in software engineering are effective. Cohesion and coupling are good indicators for software quality with respect to the reduction of errors and maintainability. Considering workflow processes that are targeted at information processing and show similarities to software programs (see Section 3), similar results may be expected. • The application of the heuristic in realistic workflow design settings is feasible. The computation of the cohesion and coupling metrics is supported by the CoCoFlow tool, which will limit the risk of human mistakes that can occur when manually determining these values (see Section 5). The focus on the information processing function is also compatible with the industry-strength workflow design method PBWD [2], [42] and [45]. Further support for the viability of the metrics may be found in earlier suggestions by workflow researchers to use coupling and cohesion metrics for workflow model decomposition [7] and [8]. As we have argued in Section 6, we believe that our presented work delivers concrete design guidance, which is often absent in related work. Clearly, broader empirical evidence that support the effectiveness of the presented heuristic needs to be gathered. This will require extensive fieldwork, because the results of the workflow design with respect to its execution quality will only be measurable after the implementation of the design. In a typical workflow project, this period may span several months or even years. For the understandability of the workflow models, it is feasible to set up experiments which are not attached to actual projects. We are thinking about conducting experiments in the spirit of [48], where respondents will be asked to evaluate alternative designs under laboratory conditions. We have used software engineering metrics as a source of inspiration, but it should be noted that software engineers use metrics for a much wider range of properties than we have considered here. In this paper, we have aimed for a modest goal with respect to workflow process design, focusing only on execution quality and maintainability issues. These properties have led to the most tangible results in the software engineering domain. A comprehensive quality concept for a workflow design should perhaps encompass additional aspects such as reliability, security, interoperability, etc. [24]. One of the drawbacks of the presented heuristic is its elaborate mechanics. In contrast to the metrics as proposed in ref. [13], it is not likely that the metrics will acquire popularity on the basis of their simplicity (as, for instance, the “number of lines code” metric in software engineering). The development of the CoCoFlow tool can be expected to counter this effect for some part. Fortunately, in settings where the PBWD method is applied, the step towards the application of the proposed heuristic is rather small because the fine-grained analysis of information elements and operations has already taken place. PBWD has been adopted as one of the standard innovation methods by the management consultancy firm Deloitte and is frequently applied. From a scientific perspective, the greatest challenge lies in the fact that the heuristic does not suggest any clustering or ordering of information elements itself. An extension of the heuristic so that it can efficiently generate (semi-) optimal activity definitions itself would generate a quantum leap in this domain. Finally, we would like to note that this approach is not limited to the application domain of Petri nets. This method can be used together with any process modelling language (such as EPC, IDEF, or Data Flow Diagrams) as long as the product structure is properly described. However, in the cases which do not have a suitable product description or in which the product cannot be described as a network, other quality metrics in the workflow domain (Section 6.2) may still be very useful for design guidance.