As with most future-telling, it is an imprecise science. Using data from the U.S. Election Project for the years 2010 and 2014, though, we can get some sense to the extent to which early voting is helping or hurting candidates in several states.
So look at North Carolina, for example. The 2014 vote (the wider bar) has far more Democratic votes than in 2010 -- and is a greater percentage of the vote total, too. (It's up 1.3 percentage points.) In Oregon, where a possibly close Senate race has turned out to not be terribly close, turnout (as of Oct. 30) trails 2010 turnout (as of Nov. 2) by quite a bit. In Florida, the percentage of the early vote from Republicans is down quite a bit, even though the number of Republican voters is up. It's just not up as much as the Democratic vote.
What's particularly important to keep in mind is that several races (like Democrat Bruce Braley in Iowa) are banking on voter turnout to make the difference. Early voting allows campaigns to spread turnout efforts out over weeks, theoretically meaning that we should see dramatic (or at least noticeable) increases in early voting in those places for those campaigns. We don't see that in Iowa.
But Georgia is an interesting (and important, given the close Senate contest) state to consider. The state offers early vote numbers by race which provides some indicator of likely vote patterns, given how strongly black voters in the state back the Democratic Senate candidate, Michelle Nunn. We've seen the percentage of the black vote climb, particularly after Sunday voting last week.
But also, black voters make up a bigger percentage of those voters that the U.S. Election Project can ID by race and as having not voted in 2010 than those who are known to have voted in the last midterm. The large block of voters that is identified as "unknown" has a number of newly registered voters, who could be anyone (but also draws new attention to the battle over new registrations in the state).
There's a more concrete indicator of good news for Democrats. The Atlanta Journal-Constitution's pollster provided the paper with its analysis of the early vote among white voters. The composition of the early vote in 2014 among white voters looks a lot more like 2012 than 2010 -- down as a percentage of returned ballots, a trend that suggests a more presidential-election-cycle turnout.
"Suggesting." The Journal-Constitution's pollster is quick to point out that "the percentage of early voting vs. overall final voters has been fluid: In one year the percentage increased (2008), in one year it fell below early voting (2010) and in one year it remained the same (2012)." So the entrails of the early vote may look like we can make predictions, but given that they're a subset of the total and, to some extent, data captured in amber at the moment of the election that the vote is cast, we recommend strongly against preparing any acceptance or concession speeches based on the graphs above.