A lot of people, especially in the press, think/claim the polls were wrong because they want to think the polls were wrong. In reality, they were actually quite accurate. A lot of people just refused to believe them. They also shifted rapidly in the last couple weeks thanks to Jim Comey.

A few aggregators were way off because they made the freshman-stats mistake of ignoring correlations between the different state polls. This was the issue with Sam Wang (who should probably get out of the business after giving Hillary >99% odds), and to a lesser extent Upshot (who I think gave Trump 1 in 10 odds on election day, which was presumably an underestimate though still very compatible with Trump winning). I hope now that we have more and more data scientists out there that news organizations will hire them and be smarter with their analysis.