It'll probably help to enumerate all the premises that polling begins with and question whether those premises are in fact the case.
If your definition of "work" is that a poll accurately predicts the outcome of the election, take a step back. Polls don't do that.
I'm not just referring to inaccuracies that are introduced by non-representedness of the sample, although that's clearly a factor as well.
Polls invite respondents to indicate how they would vote "if an election were held today". But the election isn't being held today, and the respondents know it. Polls are an invitation for respondents to send a message to politicians before the poll, to influence the politicians' offerings. Respondents know that, and use the opportunity that way.
So what are polls? Polls are their own little universe, with their own characteristics, dynamics, and effects. What they aren't is a crystal ball.
Predictions don’t count for much unless you check that the expectations line up with reality.