Simply based on the information in the link you provided, I think the answer is "the sooner the better". Maybe it's something that's decided upon before realizing changes need to be made. i.e.: making the entire company's website more user-friendly > Finding out the shopping cart isn't user-friendly at all and deciding something should be done.
If done during the requirements phase (or before) it would be much more successful than at other times, as efforts can be made to incorporate the vision into the development of a project from the start.
If you're mid-project and everyone decides it's necessary for the company's future, there's no reason not to switch the focus to that new vision. Going back and changing an ongoing project to align closer to the vision may not be the best or most feasible solution, but all future work can hopefully be aligned closer to the vision going forward.
Picking a vision at the end of a project to test the project against is probably the worst time to do it...obviously it's a bad idea to retroactively judge a project on requirements that were never previously discussed.