There are a lot of problems with polling to predict elections. You can miss by getting unrepresentative samples, for example. In the US, pollsters try to use phone interviews. They are restricted to landlines, which gets them a non-representative sample - they get more older, less technically comfortable people. More likely to get people who don't work (home during the day).
The pollsters claim to be able to account for these issues, but the proof is in the election, where everyone actually votes (rather than just a sample).