The bottom line, something we really learned from Yardsale, is no matter what you build, the perfect product, you'll end up discovering you built the wrong thing. The longer you spend on your beta, the more costly deep iterations become -- design assets, polish, etc. all are very expensive.
We started out with I think a 2 wk spec, which actually took 3 wks to build, and followed up by restricting every new feature to a 2 day spec, including big things like user-to-user payments. This changes how you think about these features. Suddenly you realize you don't need ACH integration, you can mail checks until you hit some scale where an intern can't handle it any more, etc.
PS - Shout out to Scribd, and the Parse founders, who've made an art of this kind of development.