Ahh, one of the "classic" discussions of the CMMI. The SEI has a couple of
presentations on it at their site that you can Google, but they tend to talk
in generalities.

The first thing to understand is that you can't do either staged or
continuous in any sort of "pure" form. If you use the staged approach,
you'll still have to pick and choose which PA's you work on first, because
it usually makes no sense to try to address them all at the same time. If
you try to use the continuous representation to work on only one or two
PA's, you'll quickly find that you have to put at least part of some other
PA's in place (like PP, PPQA) in order to satisfy the Generic Goals and get
real value from the PA. So when you choose a representation, you're choosing
more of a mindset, than actually eliminating the other "approach".

Generally for lower maturity level organizations, I recommend the staged
representation. This gives you guidance as to which PA's have been found to
be most important by other players in the industry. The continuous model
gives you no guidance as to *which* PA will be most helpful to you. Of
course, you should have done problem identification first, which should help
you choose PA's, but I find that much confusion and wasted time can arise
arguing over which PA's in the continuous model should be examined, when
it's clear that the ML2 practices are what's needed.

Here again, I'm speaking from the original experience/intent with the
SW-CMM. The PA's were placed at maturity levels roughly in keeping with the
level at which the problem appeared during organizational assessments. In
more detail, early assessment teams found that ML1 organizations complained
about similar things (poor plans, infeasible commitments, no change control,
etc.), while ML2 organizations complained about different things (too long
to start up a project, can't move people between projects, always have to
start with blank page, etc.). When ML2's were asked "why don't you complain
about those things the ML1's complain about?" they responded with things
like "Oh, we prevent that from happening by doing X, Y and Z". Turns out
that there were a lot of similarities among the X, Y and Z's, and those are
what are captured in the model. So what's in the model are the attributes of
solutions to problems typical of the next lower maturity level.

One of the interesting things was that people at Level X normally didn't
complain about problems associated with Level X+2 practices. E.g., ML1's
didn't complain about problems associated with the lack of standards,
organizational training and re-use. That's because they had bigger problems!
The ML 1 problems, addressed by ML2 practices. This is something that we
have found true in all our assessments. People complain about what's in
their face, and don't complain about things that are 2 or more levels above
them. And it usually provides the biggest ROI to address those issues first.
Also, "lower" levels provide the (usually) necessary
infrastructure/discipline to perform higher level processes with any kind of
value. For example, you can have a standard, formal process for peer
reviews, but the standardization and formality (the Level 3-ness of it)
won't buy you much if you don't have the ability to do RM, PP and PMC!

I've seen several organizations get totally screwed up by tackling an ML3
process area without first addressing the bulk of ML2. In particular,
premature attempts to standardize can undermine a whole improvement program.
I saw a multi-billion dollar financial organization cancel their entire CMM
program after 4 years due to this kind of misunderstanding (and doing
"improvement" as compliance, rather than solving problems).

For higher maturity level companies, I think the continuous representation
is very useful. It allows you to look at what parts of your processes you
want to improve, to what level, to give you the biggest ROI for your
particular business environment. Essentially, the continuous representation
gives you more slack in terms of choosing what you want to concentrate on
first, what's most important to your particular business situation. In the
hands of an experienced person, the added flexibility can be quite powerful.
In inexperienced hands, that extra slack translates into just enough rope to
hang yourself!

From the point of view of dealing with senior management, the staged
approach has the advantage of being simpler to understand, simpler to
monitor progress, simpler to set goals. I also believe that, in *most*
lower-maturity situations, any extra power/value that might be achieved by
the additional nuances provided by the continuous representation are not
worth the additional risk. In some situations they are, but again, more at
the higher levels.

With ML1's I am a firm believer in the staged approach. This is because it's
usually critical to getting them started that they see the complete
"paradigm" of a ML2 organization. Tackling things piecemeal, at least in
introducing the program, makes it *much* harder to deal with the cynicism
that often accompanies such a program in ML1's. When teaching 3-day CMMI
courses to ML1 organizations, I find that there's a point somewhere on the
second day where the light goes on. Students have seen the flavour of each
level, and have seen RM, PP, PMC and PPQA of ML2. Up to this point, they
have followed everything intellectually, often looking at each practice with
the hidden idea of "yes, but even if you did that, it wouldn't work because
of this over here, or that over there." It's only after they've seen how the
whole "commitment process" works, that they start to say "Hey, I can see how
that would cover it from different angles. This could actually work, and I
can see how it could work in *this* organization!" That steady-state
paradigm of how the different parts of ML2 reinforce each other, is the key
ingredient in getting buy-in from the affected people, and getting them to
see how their problems would be affected and how the practices should be
tailored to their situation.

With the continuous representation (at least in training), it's much, much
harder to see this dynamic, mutually-reinforcing quality of the PA's, and
how certain problems are addressed from several different PA's
simultaneously. This affects buy-in and also raises the risk that the
connection to the organization's actual problems will be weakened or lost,
leading to "compliance" to the model, rather than value-creating
problem-solving.

One other note, on why PA's appear at their particular maturity level. In
the original SW-CMM, as I said above, it was mostly based on what showed up
during assessments of organizations at different levels. When the SW-CMM was
incorporated into the CMMI, this rationale was not the only one used. For
example, Measurement and Analysis became a ML2 PA primarily because the
early high-maturity level organizations were asked what they would do
differently if they were to do it all over again. Several of them noted that
they had wasted improvement time by not having their measures carefully
thought through at lower maturity levels, causing them to have to re-do much
of their data capture. 20-20 hindsight, but some of us believe it's not a
sufficient justification for putting it at ML2 and having it compete for
organizational attention and resources with more immediately important PA's
like planning. Also, the systems engineering model that was incorporated
into the CMMI was a continuous representation model, and the systems
engineers have their own priorities (cf. "Data Management").

Sorry for the lengthy reply, but the topic touches on a lot of issues (even
more than I've covered here!).

David Constant
Partner, Process Inc
(613) 722-8707
www.ProcessInc.com

Date: Mon, 09 Feb 2004 08:23:43 -0000
From: "penichet" <penichet_chafa@...>
Subject: CMMI Representations, which one is the better?


Hello everyone,

I'm trying to indentify the advantages and the differences between
both representations, and some times I think the Continous is better
than the Staged, but then I'n not so sure. A lot of people says that
it all depends on what your company is used to, but if you take
appart just for a moment the legacy in your orgnization should't it
be an easy way to choose between the two representations ???

Is there a list or a set of questions that you can apply to your
organization that can help you choose a representation ???

I'll appriciate any comments or ideas. Thanks