Early stage startups tend to constantly "redesign" everything where in fact evolving your design to the desired spot is a more educated approach.
when you're going from "black and white wireframe" to "something userfacing" for example, you probably don't have to care; but with an established userbase that is USED to the product, you should care more.
if your product is transactional and there aren't good alternatives (people come to site X to send money to another person and this is the only one with an API) you might not need to care, vs "people spend time as a diversion on my site and explore photos and status updates of friends", it might be important to care more.
100% agreement that evolving design to the desired spot is more educated -- I think that statement is indisputable, as it is essentially saying "if you can incrementally split test results with lots of different versions, you can learn more".
With a large "total redesign" you just get one fat multivariate test that can have conflicting results and you can't isolate the winning changes.
I was just stating that the time penalty for evolution might be too large for extremely small teams. its not just a design, its design, implentation, and testing the differences -- which takes a lot more time than just implementing the design itself. without testing results, incremental design is only good for "easing the user in" and you lose all the benefits of the "more educated approach" you're referring to (which, in my mind is the valuable part... users are forgiving, you can shock UI from time to time, but information gained can be priceless)