The Conference on Advanced Information Systems Engineering (CAiSE) is one of the world’s leading academic conferences in the area of information systems and process modeling. This year’s CAiSE took place in Montpellier, France from 16 – 20 June and it featured quite a number of presentations concerning BPMN and “good” process modeling.

Not only could I experience that many BPM researchers share a common interest: following football games. I could also enjoy quite a number of inspiring presentations. In this article I point to some I found particularly interesting.

Towards Systematic Usage of Labels and Icons in Business Process Models

In this presentation Jan Mendling presented some interesting insights into labelling of activities in process models. Together with his colleague Jan Recker from the Queensland University of Technology in Australia he empirically evaluated what linguistic structures are most commonly used as activity labels in process models and what verbs occur most often. Although many process modeling teachers out there preach the “verb + noun” labeling style, e.g. “create invoice”, reality seems to look different. However, you always have to consider the source of the process models you are actually evaluating: In this case the study was largely based on the SAP Reference Model that contains around 600 EPCs. The fact that most process experts having created these models most likely speak German as native language might have influenced the actual labeling.

But even more interesting were the investigations of verbs used and a strategy to take advantage of it. The presenter’s idea was to come up with a core set of verbs and automatically assign icons to activities accordingly. E.g. if you have an “approve order” activity, your modeling tool would automatically add an approve-icon to that activity. That way the intuitive understanding of the model should be strengthened. Using semantic relationships as captured in WordNet the corresponding “core verb” for a verb used can be found. While first experiments have already shown that proper icons can indeed increase understandability, badly chosen icons can in turn consufe modelers. How would e.g. an icon for approval look like? Another interesting insight was, that it is not always easy to find a corresponding core verb and icon for every verb used in a model: The most frequent verb used in the set of models under investigation was “to process”. However, the same technique could be used to guide modelers actually use meaningful verbs.

The paper can be accessed here.

How Much Language Is Enough? Theoretical and Practical Use of BPMN

Michael zur Mühlen, a researcher at the Stevens Institute of Technology in the USA, presented a piece of work that already led to lively discussions before the conference. Together with Jan Recker from the Queensland University of Technology he investigated what BPMN constructs are actually used “out there”. The models under investigation stemmed from students in BPMN seminars, consultants and from “the wild”, i.e. found on the web through Google.

The main finding was that only a small subset of constructs are frequently used, e.g. tasks, control flow, pools and lanes and exclusive data-based gateways. On the other hand there were quite a few constructs that were never used at all.

Another interesting finding was that the usage profiles of constructs differ considerably. E.g. the consultants’ models often contain pools and lanes and exclusive data-based gateways as they tend to focus more on hand-overs between different organizational units and decision responsibilities.

As a take-away message Michael zur Mühlen concluded for BPMN training that indeed a wide range of constructs should be taught. However, different subsets of constructs might need to be defined for tool vendors as only few cover the full language. Defined subsets would then enable certification of tool vendors along different modeling purposes.

More information on the discussion on “How much BPMN do we need?” can be found here. The paper can be accessed here.

Modeling Service Choreographies with BPMN and BPEL4Chor

At CAiSE I was not just watching the other guys present their newest findings but also contributed myself. In the session on enterprise systems interoperability I gave a presentation about a more technical dimension of BPMN, summarizing joint work that we did with Frank Leymann’s group in Stuttgart. The main focus was on how well BPMN and BPEL work in interplay for supporting the definition of service choreographies. Choreographies can be seen as interaction contracts between different business partners that want to interact through electronic message exchanges. BPMN seems to be naturally suited for this purpose as it explicitly distinguishes control flow (within one business partner) and message flow (between different business partners). In the presentation I focused on how BPMN supports the different choreography design stages and where extensions to the language were necessary.

For overcoming the limitation that currently there is no commonly agreed upon interchange format for BPMN, we resorted to BPEL4Chor, an extension for abstract BPEL. As part of that work, we contributed to the BPMN 2 BPEL story by extending the most state-of-the-art mappings available and implementing it in our web-based modeling environment Oryx (see

The paper in full length can be found here.

Activity- vs. data-centric modeling

Several presentations and even a keynote were devoted to the topic of activity- vs. data-centric modeling of business processes. While classical activity-centric modeling languages such as BPMN focus on activities and how they causally relate, data-centric modeling approaches rather focus on the business entities involved, what state transitions they can go through and how state transitions of different business entities relate to one another. Using this approach a better modularity of models should be achieved and actual execution is more straightforward.

While this approach seems to be promising for certain areas of application, especially when going from conceptual business process models to actually executable models, activity-centric process modeling is still highly relevant in my opinion (and in the opinion of many other researchers). Only here, the end-to-end nature of business processes becomes visible. Only here, the value creation from input to output can be traced intuitively. You might still have a look at e.g. the paper by Santhosh Kumaran, Rong Liu and Frederick Y. Wu, an inspiring read, indeed. You can access it from SpringerLink.

Further presentations

You might also want to check out “On a Quest for Good Process Models: The Cross-Connectivity Metric” (link) where a novel metric for process models was presented, “Measuring Similarity between Business Process Models” (link) covering the question of similarity between process models, “Open Source Workflow: A Viable Direction for BPM?” (link) discussing workflow pattern support in state-of-the-art open source BPM systems or Mike Papazoglou’s keynote (link) on service and process evolution. Furthermore, there was an entire workshop called “Business Process Modeling, Development, and Support” (link).