Business Process Model Similarity as a Proxy for Group Consensus

Abstract:

Consensus is an important measure for the success of any business process modeling effort. Although intensively studied in the general literature on group processes, consensus has hardly been considered in business process modeling and never seriously measured. We define consensus as the level of agreement of group members’ views on the process and introduce business process similarity as a proxy. We validate the measure by comparing it to an existing self-reported measure of consensus.

Introduction:

The literature on business process modeling is vast and the importance of measuring the success of process modeling projects and sessions has been widely recognized [1-6]. But prevalent success measures for individual modeling sessions primarily involve some form of model quality measure [7-10]. While it is undisputed that the quality of a business process model is relevant to modeling success it is not the only and perhaps not even the most important success factor.

The reason for this is twofold: the process model itself is a social construction, and its purpose is again to support some social process, e.g. a change project or system development project. In other words: the model documents the results of one social process (modeling) and serves as a point of departure for another one. If the model were to be processed by a computer its quality would be of prime importance to ensure correct interpretation by the machine.

But the results that are documented in the model are primarily the mutual knowledge that has been developed in the modeling session, the conflicts that had to be solved on the way, and the consensus that has been achieved among the group members as a result. It is precisely this consensus that is a prerequisite for people’s commitment to the ensuing change project, for example. Often a poor model with high consensus goes further than a good model with little consensus. Hence consensus is a major result that needs to be achieved in business process modeling sessions much like in many other forms of group work.

But while there is considerable research on consensus in other areas [11-13] the topic received little attention in business process modeling with researchers barely mentioning the issue [14-17] and, to the best of our knowledge, not researching it in a systematic way, let alone measuring consensus.

The purpose of this paper is to develop such a measure. To do so we first define the concept of consensus in the next section, Group consensus in process modeling. For this purpose we rely on cognitive theories of modeling. Based on the cognitive concept of a view and the model as its externalization we can interpret consensus as “view agreement” and hence as “model similarity”.

The section Business process model similarity therefore introduces a measure for the latter. A proper evaluation of a new measure typically establishes validity by comparison to an existing measure of the same concept.

The section Other group consensus measures therefore introduces an established measure for group consensus. The actual validation of the new measure was done in field experiments. The set-up of these experiments is described in the section Comparing model similarity and consensus in field experiments.

The section Data analysis reports on the analysis of the data that we collected in the experiments. The results and implications of this analysis are treated in the section Discussion. The paper concludes with a summary of the findings and an outlook on future work.

Conclusion & Outlook:

We have suggested the use of business process model similarity as a proxy for group consensus in business process modeling. We have shown that this proxy is a reliable measure of group consensus as compared to an established self-reported consensus measure. The proxy is also a more objective and accurate measure and has therefore the potential to outperform self-reported measures because it eliminates the group thinking bias usually associated with self-reported measures of group performance. It does also not rely on the existence of a group model and can therefore be used at any stage in the modeling process.

The scenarios for using this proxy are manifold and the Discussion section has shown some of them: evaluation of modeling session performance, modeling project management, conflict detection, and so on. But the real advantage lies in integrating this measure into new methods for collaborative business process modeling where the progress towards a consensus model can actually be measured and controlled.

This will allow us to organize modeling in such a way that the steps in the method really lead to an improvement in consensus so that the success of modeling sessions can actually be planned. Because of the collaborative nature of these sessions, the individual group members, or perhaps small teams, will generate model proposals anyway so that additional drawing of models (as in our experiments) will no longer be necessary.

A continuous assessment of the status quo can therefore be made at any time in the modeling project as the computation of model similarity can be done automatically once the views of individuals or teams are available as models. An evaluation of the similarity between views on the business process can hence be the driver for the whole modeling effort.

This opens up possibilities for developing a new range of consensus-driven, or consensus-oriented business process modeling methods and supporting tools for the creation, maintenance, review, revision, and integration of model proposals and their convergence to a consensus model for the group.

Beyond this the further evolution of the model after implementation can also be supported in a consensus-oriented and decentralized way.

While all this is still hypothetical the assessment and control of consensus in business process modeling is a relevant issue already today. We believe that the measurement of consensus in an objective way is an important key to solving the consensus-related issues in process modeling.