Software Hell is a Crowded Place
I’ve been thinking a lot about management fads lately, and ran into this 2005 article by Nick Carr, titled “Does Not Compute”. Here’s the part that caught my eye:
“A look at the private sector reveals that software debacles are routine. And the more ambitious the project, the higher the odds of disappointment. It may not be much consolation to taxpayers, but the F.B.I. has a lot of company. Software hell is a very crowded place.”
Carr continues by describing two examples of failed projects: a massive systems integration effort at Ford Motor Company, and a overzealous business intelligence initiative embarked upon by McDonald’s. Both projects were cancelled when the price tags got too big: $200M for Ford, $170M for McDonald’s. The catch is that failure is good, because when we fail we at least know one solution path that’s not workable – we just need to 1) understand that it doesn’t have to be expensive, and 2) have more courage to allow ourselves and our colleagues to fail without getting depressed or thinking our coworkers are idiots. This is often expressed as “fail early, fail often“. (But note that the assumption is that you persist, and as a result of the learning experience, ultimately meet your goals.)
Without an effective team culture, rational managers, healthy relationships with stakeholders, and capable programmers dedicated to continually improving their skills, all roads can lead to software hell. The process of getting there – which is hellish in and of itself – is the famed death march. This is where a software-related project, doomed to fail, sucks up more time, people, resources, and emotional energy at an ever increasing rate until the eventual cataclysm.
Carr also cites The Standish Report, which in 1994, asserted that only 16% of projects were completed on time, and budget, and meeting specifications. By 2003 the percentage had grown to 34% in a new survey. Other projects that were still completed ran, on average, 50 percent over budget. (And this is for the survey respondents who were actually telling the truth. I know a few people who wouldn’t admit that their project was quite so grossly over budget.)
One way to solve this problem is by focusing on sufficiency and continuous learning, starting the blueprint for a system based on these questions:
- What features represent the bare minimum we need to run this system?
- What are the really critical success factors?
- What do we know about our specifications now? What do we not know?
- What do we know about ourselves now? What do we want to learn more about?
Software development is a learning process. It’s a process of learning about the problem we need to solve, the problem domain, and ourselves – our interests and capabilities. It’s a process of recognizing what parts of building the solution we’re really good at, and what parts we’re not so good at. Let’s start small, and grow bigger as we form stronger relationships with the systems that we are developing. Having a $170M appetite sure didn’t get McDonald’s anywhere, at least in this case.
Many of the questions posed in blueprint questions are addressed by Agile or Lean development processes.
> * What features represent the bare minimum we >need to run this system?
Features in the form of vertical slices of customer valued functionality. In the form, “User X can do Y to achieve X.”
> * What are the really critical success factors?
This is addressed by writing acceptance tests for the feature set.
> * What do we know about our specifications now? >What do we not know?
> * What do we know about ourselves now? What do >we want to learn more about?
Daily Stand-up and the Retrospective.
Pingback: Software Quality Digest - 2009-02-11 | No bug left behind