If you have only a small sample size to experiment with then you need to run experiments that can cause dramatic lifts (not minor changes like moving a link, etc.) Because if you have a sample size, then you need to wait much longer to measure the significance of these results and as a startup, time is not on your side.
I think if you can't get statistical significance than the growth team model is probably too early. You mostly need to be in the "invest" stage and not the "test" stage. That's not to say you can't try to hack merely that measuring through multivariate tests is a waste of time and effort.
