It's possible to sample the overall vote count of a state from a small number of ballots in every precinct. There's a high margin of error, but if the results are decisive enough, it's possible to call a date at 2% official count.
Not sure if it's allowed in the US, but if you can get data from counting locations before they're officially reported, then it's very easy to get an accurate sample well before the official numbers are clear. Exit polls help with analysis as well.
Video about the math behind it: