I am the CEO of Process Master and the vast vast majority of users are only using 20% of the available stencil

However, BA’s and process specialists are using it all - which is the beauty of BPMN

So using BPMN for the production of maps and documentation, only needs a fraction of the stencil - and it expands naturally to 100% if you are using it for automation or indept BPI

There is a video of all this on http://www.ProcessMaster.com

Cheers

Alan

The third seems fine on the face of it but consider an analogy. Would the technicians designing and building an airbag system or anti-lock braking system for a car gloss over it because it’s only ever used in 0.1% of car journeys? Obviously not. To have a complete car, which functions as you would like it to in all circumstances, you need all those systems. So saying we don’t need a complete modelling language based on a statistical analysis of diagrams may not be valid.

Using those data to say that the OMG should not spend a lot of time understanding and defining the most complicated and intricate, if least used, elements of a modelling notation is, in my view, a mistake.

Thanks for your comments.

As for your first remark, I have a gut feeling (but don’t have the data to prove or disprove it) that when you move closer to the execution level that there is a crossover point where implementing process semantics relates closely to programming. If you look at the development environments of most BPM vendors these days they are all integrated in Eclipse, giving you easy access to the source code of the adapters / data structures / programs / rule sets etc. you may need to tap into in order to make your process work. While you can create a very detailed specification using a more comprehensive subset of BPMN, the necessary overhead created by the graphical representation of a programming concept may just be too much. For example: In a BPMS I know well you can specify multiple timer events per activity. You could - say - increase the priority of the activity after 2 hours, reassign the activity to somebody else after 4 hours, and notify the manager if it has not been completed after 2 days. Now, when the activity gets reassigned to a different person that holds the same role as the first assignee, do we model this as a different activity in a different swimlane? And how do we represent the notification of the manager (who may reassign the task, but may choose not to work on it)? Details like this today are hidden in the property pages of BPMS. It is an open question, whether we should make them visible in our process models.

As for the second remark, we tried a cluster analysis on the models themselves, to see whether we could find commonalities in BPMN usage between models that were created for a similar purpose. Unfortunately, the models did not cluster at all - the reason being that each model was created with a fairly distinct subset of BPMN. If we had a larger base to draw from we might be able to find multiple cores - I would like to think that several well defined sets of constructs will emerge over time.

Two remarks:
1) If BPMN is not only used for high level process representation, but also to define operational process execution logic, does that somehow impact the core set of BPMN elements? I mean if one is to run the process that was designed for real, are there additional elements that would become core, just because the level of precision required is different.

2) I think it would be very interesting to go deeper into the analysis of the use patterns of BPMN by different populations of users. For me BPMN is above all a candidate shared language between business users and IT folks: that’s rare enough to investigate what parts of the language are actually shared, shoud be shared and (possibly) can’t be shared.

Regards

When will people find out that simplicity adds more value than complexity?