Distinguishing Fact from Fantasy
A common mental model relative to software reliability is ...
"Yes, we could design highly reliable software. The problem
is we can't afford it. It will take too long and cost too much.
Our most important priority is to get the product to market. If
we don't get our product to market quickly, we won't make sales
and profit; and, after all, profit is what counts the most."
Based on each of our own personal experiences, this really does
make sense. In every personal project we've ever undertaken, we've
discovered that making a higher quality product requires more
care and therefore more time. The problem is that extending our
personal experience to an organizational level is often not valid.
In fact, it is quite common that the system-level consequences
of our actions are extremely counter intuitive.
What if designing reliable software is actually less expensive
at an organizational level than designing lower quality software?
What if, as asserted by Putnam & Myers in "Measures for
Excellence", "When productivity improves, errors seem
to decline, or, as others put it, when more emphasis is put on
quality, productivity increases."?
If producing higher quality software does result in higher overall
organizational productivity, this would change the entire approach
our organizations take in producing software. As a consequence,
what would be the effect if we were to consider an alternative
Such a model might be: "Yes, we design highly reliable software
because we can't afford to not do so. It takes too long and it
costs too much to correct defects once they're in the product.
If our engineers aren't fixing problems in response to customer
problems, they can be creating new products and features. After
all, our most important priority is to make sales and maintain
customer satisfaction to produce the highest possible profits."
The end result desired based on both of these world views is the
same. How we endeavor to accomplish this result, and our success
in doing so, depends on our mental model of the software development
Systems Thinking and Software Project Management
The system of interactions in a software development project
is a complex web which we must understand to avoid the unintended
and counter intuitive consequences that that cause project overruns
and even project failure.
The frightening reality is that many software development projects
get caught in what's known as the "Impossible Region."
Causal loop diagrams of the complex web of effects of such things
as, for example, overtime and rapid staffing effects on project
quality, schedule and cost show that projects are not just "problems
to be solved," but "messes" in the truest sense
of the word. We greatly underestimate the magnitude of the unintended
consequences of such interactions.
This initial representation indicates that at the Work Remaining
with regard to the current project schedule increases it will
tend to produce Schedule Pressure. This Schedule Pressure
with tend to promote Overtime to reduce the Work Remaining
with regard to the current project schedule. This structure represents
a balancing loop where Overtime is used to counteract the
The following structure points out a couple unintended consequences
of increasing Overtime to combat the Work Remaining
While the Overtime is intended to counteract the Work
Remaining it has a couple additional influences. If Overtime
increases sufficiently it will begin to depress Morale,
which will subsequently influence Productivity to decrease.
The decrease in Productivity will then tend to increase
the Overtime required. This structure represents a viscous
reinforcing loop moving opposite to the direction desired.
At the same time an increases in Overtime influence Morale
to decline it is promoting an increase in Fatigue. The
increase in Fatigue will then tend to reduce Productivity
further increasing the Overtime required. What we have
is another viscous reinforcing loop moving opposite to the direction
If these unintended consequences weren't enough annoyance, the
following structure points out a couple additional unintended
While Morale and Fatigue are influencing Productivity
declines, Morale and Fatigue also have the nasty
habit of influencing Quality to decline. The decline in
Quality is often not immediately realized but enters into
the structure as Undiscovered Rework. This Undiscovered
Rework is work that needs to be done, we just don't know about
it yet as it's undiscovered. As Undiscovered Rework increases
it will tend to influence us to find more of the rework thus increasing
the Known Rework. An increase in Known Rework just
serves to increase the amount of Work Remaining which increases
Schedule Pressure, and subsequently increasing the need
for Overtime. Here we are again, with two viscous reinforcing
loops taking us in exactly the direction we don't want to go.
A short time ago we addressed the manner in which decreases in
Morale and increases in Fatigue tend to cause Quality
to decline. The following structure alludes to another influence
resulting from a decline in Quality.
As Quality declines, and once the decline is realized,
there will be an increased focus placed on Quality. This
is represented in the above diagram by Quality Pressure.
As Quality Pressure increases it will tend to increase
the resultant Quality. Thus we finally have a balancing
loop which moves something in a desired direction.
Yet, don't get too comfortable for the following structure provides
an insight into an unintended consequence of Quality Pressure
which isn't quite so beneficial.
Once a decline in Quality is realized there is an increased
emphasis on Quality, i.e. Quality Pressure. This
increase in Quality Pressure will serve to improve Quality
as shown in Figure 4. Yet, an increase in Quality Pressure
also serves to decrease Productivity because there is more
of an emphasis on getting it right than getting it done. So this
decline in Productivity serves to promote more Overtime
which increases Fatigue and decreases Morale. The
end result being a tendency for Quality to decline. Thus
we have two more viscous reinforcing loops which simply indicates
that the more Quality Pressure applied the more will be
needed. Doesn't seem to make much sense does it? Welcome to the
dysfunctional reality of organizational life!
When we merge the structures in Figures 2, 3, 4, and 5 with Figure
1 we end up with Figure 6. The current elaboration of the understanding.
Before you let yourself become overwhelmed by the complexity
of this diagram you had best fasten your seat belt as we're only
about half way there.
Overtime has this real nasty habit of costing more than
regular time so there are some implications of increasing Overtime.
An increase in Overtime brings with it an increase in
Overtime Cost. As Overtime Cost increases there
is an increased emphasis on cost which shows up as Cost Pressure.
The Cost Pressure is interpreted by the management of project
in such a way that it shows up as additional Schedule Pressure.
This increased Schedule Pressure then leads to even more
Overtime. Here we have but one more viscous reinforcing
loop in which actions influence the overall effect to be just
the opposite of what is desired.
Overtime and Overtime Cost have a couple more influences.
Prolonged Overtime has a tendency to lead to Burnout
which means Hiring must occur to replace or augment resources.
Yet Hiring only serves to increase Cost Pressure
also, creating another viscous reinforcing loop.
Also, in an attempt to minimize Overtime Costs additional
resources are hired. And, because of the time delays involved,
Hiring only serves to increase Cost Pressure. We
there for have another viscous reinforcing loop driving Cost
Pressure to increase Schedule Pressure leading to more
Overtime. Does it sound like things are going down hill
Now as Fredrick Brooks stated in "The Mythical Man Month"
more than 20 years ago, "Adding additional resources to a
late software project only makes it later," has a very solid
foundation. What follows are some of the unintended consequences
Hiring serves to increase the Percent New Staff
which tends to increase Attrition Rate which simply servers
to require more Hiring. You guessed it, another viscous
As the Percent New Staff increases it tends to produce
Supervisor Strain. As Supervisor Strain increases
it influences a decline in Productivity and an increase
in Overtime and we're back to the same part of the model
presented in Figure 8. Yes, but another influence which is part
of two viscous reinforcing loops. Are you beginning to feel there
is no hope in sight?
Percent of New Staff has another influence just as miserable as
described in the next figure.
As Percent New Staff increases it decreases the Average
Skill Level of the resource pool. This has a tendency to decrease
Quality which feeds right into the viscous reinforcing
loops described in Figure 3 and Figure 5.
Now when we combine the implications developed in Figures 7 thru
11 with Figure 6 we have a nightmare even I'm not happy looking
Schedule Pressure has a couple additional influences that should
Schedule Pressure serves to increase Overtime
thus reducing the Work Remaining and finally decreasing
the Schedule Pressure. This balancing loop is supported
by a virtuous reinforcing loop as Schedule Pressure tends
to increase Productivity. This increase in Productivity
then tends to decrease Overtime increasing the Work
Remaining. This increase in Work Remaining then supports
the continued Schedule Pressure.
Schedule Pressure also has an effect on Quality.
Schedule Pressure serves to influence Quality
to decline. This decline in Quality results in an increase
in Quality Pressure which serves to decrease Productivity
resulting in an increase in Overtime. The increase in Overtime
then serves to reduce the Work Remaining. This is a balancing
loop such that an increase in Schedule Pressure tends to
reduce Schedule Pressure. The decrease in Quality
due to the increase in Schedule Pressure serves to increase
the Undiscovered Rework thus increasing Known Rework
and the Work Remaining. The increase in Work Remaining
influences an increase in Schedule Pressure. This is a
viscous reinforcing loop where an increase in Schedule Pressure
tends to influence additional Schedule Pressure.
Now, combining the structures in Figure 13 and 14 with Figure
12 we have:
If this is reality is it any wonder we have such difficulty
getting projects done on time and within budget?
Our standard approaches for managing and controlling projects
(including reviews, work breakdown structures with earned value-based
tracking, and PERT/CPM and Gantt scheduling) are not adequate
to understand, and guide us to prevent, problems caused by these
dynamics. Using them is like driving by looking in the rearview
For example, projects often get a bad start due to underestimating
the effort and the time required. Project underestimates often
end up causing seemingly never-ending difficulty and would cause
Mr. Rogers to ask, "Can you say Death Spiral?" Underestimates
can put projects in what might be called the "Dead Meat"
region where they are subjected to large and simultaneous quality,
schedule, and cost pressures.
This region is larger than one might think because the effort
required on a project goes as the cube of the code size and the
inverse fourth power of the development time (see "Measures
for Excellence" by Putnam and Myers, 1992). Seemingly minor
underestimates in code size and/or duration-required can cause
a major underestimate in the effort required.
While managers have little control over projects, they do have
great influence in avoiding the unintended and counter intuitive
consequences that that cause projects to falter. Systems thinking
can help managers, engineers and programmers understand the dynamics
of project system, their part in the system, and the varieties
of policy feedback that cause project performance problems.
Such a systems perspective sheds light on what doesn't work, and
on what does work, in managing software projects. For example,
demanding excessive overtime and hiring personnel too rapidly
definitely don't work because they have an adverse impact on quality
and productivity -- and ultimately on project schedule and cost.
Among the things that work are to
- do excellent planning including product specs, project plans
and test plans before starting development,
- guard band schedule beyond the minimum development time because,
for example, a 15% schedule guard band saves ~50% in required
- identify independent, parallel development opportunities
because two decoupled sub-projects take about one quarter the
manpower of one large project of the same size,
- test as soon as possible to avoid the effect of defects on
downstream code, and
- before the project starts, identify optional functions that
can be worked on later, or dropped, if the project gets in trouble.
Special thanks to Bob Powell, Ph.D., MBA, for contributing
the conceptual foundation for this article. Bob is a consultant
in continuous improvement and learning organizations based in
Colorado Springs. Contact him at (719) 599-0977 or at email@example.com
theWay of Systems
Copyright © 2004 Gene Bellinger