I like to think of startups as research projects. You start with a hypothesis, which is usually "there is sufficient market for <idea>". Then you do the experiments to prove the hypothesis, which is the actual startup. How you go about proving the hypothesis is just as important as the hypothesis you start with. In the world of science, if your experiments have flaws they won't be accepted as proof of your hypothesis. In the startup world, flaws don't necessarily discredit the hypothesis, but then fail to prove it when you go bust.