Why Re-work isn't a Bad Thing in all Situations.

Software Development has many characteristics of production. I'm thinking of the various steps that as a whole comprise the development process, in which artifacts are passed from one sub process to another. Requirements are passed on to design, code is passed on to test (or vice versa, in test driven-development), and so on.

Alistair Cockburn wrote an interesting article on what to do with the excess production capability of non-bottleneck sub-processes in a complex production process. In the Crosstalk issue of Jan 2009, he explores new strategies to optimizing a process as a whole. One idea is to consciously use re-work (you know, the bad thing you want to avoid like the plague) to improve the quality of affairs earlier rather than later.

In terms of what to do with excess capacity at a non-bottleneck station, there is a strategy different from sitting idle, doing the work of the bottleneck, or simplifying the work at the bottleneck; it is to use the excess capacity to rework the ideas to get them more stable so that less rework is needed later at the bottleneck station.

Mark-up is mine. You see the basic idea.
In conclusion he lists the following four ways of using work that otherwise would have spent idling around, waiting for a bottleneck to finish work:

* Have the workers who would idle do the work of the bottleneck sub-process.

Note: Although this sometimes is mandated by some of the Agile community, I have seldom seen workers in the software industry that can produce excellent quality results in more than one or two disciplines.

* Have them simplify the work at the bottleneck sub-process.

Note: An idea would be to help out the BAs by scanning through documents and giving pointers to relevant information.

* Have them rework material to reduce future rework required at the bottleneck sub-process.

Note: See the introducing quote by A. Cockburn. Idea: Improve documents that are going to be built upon "further downstream" to approach a quality level of less than 1 major defect per page (more than 20 per page is very common in documents that already have a 'QA passed' stamp).

* Have them create multiple alternatives for the bottleneck sub-process to choose from.

Note: An example would be to provide different designs to stakeholders. Say, GUI designs. This can almost take the form of concurrently developing solutions to one problem by different engineering teams, for the best solution should be evaluated and consciously chosen.