Literate analysis, literate modeling
Recently, I've begun using R2wd and may also use Sweave as the application of literate programming ideas to data analysis. In that field, the concept is also known as reproducible research. In my experience so far, that approach does help organize the analysis process, and it does seem to justify the "reproducible research" moniker. Between the two programs, they can turn a reproducible analysis into your favorite publication format: PDF, ODT, HTML, or DOC(X). (Any scripting approach can help make analyses reproducible; R2wd and Sweave integrate the analysis with the communication of the results.)
The same principle applies to simulation modeling. That's why I've advocated literate modeling in Introducing literate modeling and Literate modeling and neatness.
In a heavily GUIfied, point-and-click world, I realize this may go a bit against the grain. Perhaps it helps to admit that Emacs is my computer tool of choice and ESS is my favorite interface to R.
Have you tried anything like this? What have you learned?