Actually, no, I don't think that's entirely it. The intention to vote has almost always favored Hillary, outside of a handful of days.
The issue is how they structured their poll. Only half of their poll is people that have actually voted before. They then asked these people to self-identify who they voted for in 2012. (This is terrible, because people rarely report this correctly when asked, and there's a larger likelihood you report voting for the person who won.) The other half of the sample is made up of new voters or people who have never, ever voted before. Which, is just stupid. So, they started out with a shit sample, and since they poll the same people, the shit sample is going to stay shit. That's why they have a 5 point bias, which, you know, I think makes them kinda worthless, except if you want to watch trends a bit. But even that is fucked, because there's no way Trump can go from 2 points to 22 points in two days among Black voters.