Hi:

Which to use Data Flow Diagrams (i.e., Gains & Sarson) or BPMN?

Key Principle:  Larger scale systems, at higher levels of abstraction, have no definable sequence.   Therefore, sequence based techniques such as BPMN can not be used.  This is where non-sequence  based techniques (data flow diagrams)  come into play.

Look at the BABOK.  A requirements spec specifies the behavior of a system.  The BABOK is really a requirements spec; it specifies the behavior of a system of (one or more) BA's performing typical BA tasks (processes).  Ever wonder why the BABOK is organized around inpu/process/output diagrams and not flow of control diagrams like BPMN?  The reason:  There is no set sequence to BA processes.

If, in your modeling efforts, you find that nailing down an integrated  sequence of operations becomes impossible, keep this in mind.

Tony

Hi:

How deep to go?   Excellent question!    You want to be Agile.   You want to proceed in as top down a fashion as possible, with a focus not such much on the stand alone requirements, but upon the interrelationships between the requirements (i.e., the data flows).  Especially, for larger scale projects, if you proceed bottom up, you will "drown" in a ocean of detail.

How many levels down.  You would be surprised.   With just a couple of levels of decomposition - but not shortchanging on data flow interrelationships - you will be way ahead of 99.9% of all requirements analysis efforts.   For the lower levels, just like they recommend for User Stories, you can let the developers hash such out in their conversations.   What the developers do not understand is the bigger-picture, and especially the higher level interrelationships.

Questions to ask:  Again, the focus is on the ESSENTIAL interrelationships.  Let the developers handle the nuances.  Think input/process/output in developing your questions.

Estimating:  I demopose the system down into equal size chunks, estimate for one chunk, then multiple that out.  Time for a chunk, depends on how far you decompose down.  But, if you do equal size partitioning, this grately facilitiates estimating, especially vs the way such is typically done.   It is so much more easier to estimate for a set of equal size chunks than it is for a big blob.

Tony

 Any updates on the same?