The inequality holds because you'rer splitting the integral/sum into two parts.

You throw one of those away and you obtain a lower bound on the other.

It's actually a very crude bound that pertains to all distributions hence it's not close in many situations.