The polls were right, you were wrong
For everyone hating on polls and polling and saying that they got everything wrong, here is, word for word, what Nate Silver wrote three days out from the election:
[Clinton]’s probably thinking: I’m certainly not thrilled about how the last week or two have gone, and particularly about my campaign’s inability to redirect the focus back to Trump’s many vulnerabilities after the FBI news broke. But that’s water under the bridge now. I can imagine a few ways that I might lose: If African-Americans don’t turn out in large numbers, if there’s a large turnout for Trump among white non-college voters, or if some whites with college degrees — traditionally a Republican-leaning group — come home to Trump. If one of those three things happens, I should be fine. If all three of them do, I’m probably toast. If it’s two out of three, that’s where we could be headed for a very long night and a possible popular vote-Electoral College split.
In the end, each of these conditions happened and Trump won:
1) African-Americans turned out modestly less than they have in recent elections (voter suppression likely had something to do with this).
2) Non-college educated voters do appear to have turned out for Trump. [1]
3) Whites with college degrees didn't so much "come home to Trump," as they simply decided to stay home and not vote at all. The net effect of this, however, was that this bloc broke for Trump.
In other words, the writing was on the wall. If Clinton's campaign had more poll savvy people working for them, they might have been able to prioritize these blocs in their GOTV.
Or maybe they did have these people, and tried to do something, but it was too little too late.
Either way, the polls and data provided plenty of useful information. Enough that one of the world's most savvy poll readers was able to tell the story of Tuesday's election shocker three days out.
Just because you have the evidence doesn't mean you know what to do with it. The fact that you don't know what to do with the evidence doesn't make the evidence wrong.
[1] I'm having a relatively easy time finding information on relative voting patterns or important socioeconomic and demographics groups, but not on absolute patterns that capture turnouts. Overall turnout levels, however, support this story.