"User is a woman" would, in this hypothetical example, be one factor in your larger weighting algorithm. A weighting factor that is, might I add, quite easy to implement compared to detecting male performing arts majors with a specialty in costume design.
Because you don't know in advance. It's entirely why advertisers offer so many different bits of information on you, to optimise this.
> I would think you'd be better of targetting people who bought some sort of eye makeup if you wanted to sell mascara, and target people who bought pads or tampons if you were selling pads.
And people do this, if they have this information. Of course, there's also the side that you don't want to advertise to someone you strongly expect to buy, because it's the conversions from no-sale to sale you care about.