Polls, not the ultimate results are the randomness in the equation. Since you're (obviously) not polling 100% of the group, your results will always have a sampling margin of error.

A few other entries into the system for randomness I can think of:

Voter's propensity to change their vote/stay home in the face of a negative October surprise for their candidate, or a positive revelation about the opposing candidate.

In the case of the USA, the structure of the electoral college can make for situations where one candidate has many more paths to victory than the other. More paths, more probable they'll win.