Money is the main reason here. We're still running 2008 (not R2) on one environment, while 2012 on the other. I'm not going to lie when I say that I hate working with the 2008 server, mainly because the part of the company that use it historically had bad data management; wrong data types and thus bad data formatting. Thus I really miss simple things like TRY_CONVERT. As the application is a vendor database as well, when doing things like running aggregates I can't use quirky, as the indexes aren't in the right set up. Thus I often end up with Triangular Joins (puke). LEAD and LAG are also a massive loss.
We are looking at upgrading mind, however, we're going to move both environments onto one. This'll save us a lot, as instead of 2 sets of licences, we only need the 1. The 2008 server has little traffic it, but will be required indefinately so it makes sense to do a merge when it's not going to impose any performance problems hosting both front end application on the same server. We're most likely looking at 2017 as why wouldn't we (well the reasons Jeff gave). Earliest we'll likely have the server is going to be End of Q1 2018, however, more likely looking at Q2. Even if we got it early, by the time testing is done this means that we should be on about CU8 (so hopefully a lot of bugs will be ironed out).
There's a lot of new features that have come out since 2012, a lot with 2016 which should be quite interesting.
Excuse my typos and sometimes awful grammar. My fingers work faster than my brain does.