What is Minimum detectable effect?

Sample size scales with the inverse square of it, so halving the effect you want to detect quadruples the visitors you need. That relationship is the real answer to "why does my test need so many people".