Last week I was discussing an A/B test proposal. Everything was well laid out; clear hypothesis, MDE calculated and experiment planned. The only problem: the variation hardly differed from control. The actual change was minimal. So minimal I don't expect people to notice. Now I could be wrong and that is why we are testing this of course. But my recommendation was to create a more bold version based on the same hypothesis. Take the variations you're testing to the extreme and don't be (too) cautious. You want to learn from customer behavior, but to be able to do that you need to be able to separate signal from the noise. Try something bold.