This analysis from last year is, I think, a great look into whether OGAE is reliable as a predictor of Eurovision results:
https://www.escxtra.com/2017/04/21/...eath-fan-favourite-bias-worthwhile-indicator/
Essentially, it shows that the poll isn't a perfect metric, but is certainly better than random guesses would be, and appears to be mirroring the final results more closely as the years go on.
As for why it's not perfect... well, there are a bunch of reasons, and I think the biggest ones have been mentioned here already. Combining those with some of my own ideas, I've done up a list below in roughly descending order of what I think has the biggest impact:
1. There really aren't any metrics at this point – OGAE polls, other polls, betting odds, whatever – that can properly account for staging, camera angles, live vocals on the night and all that sort of thing. And we've seen before that those can have a huge impact, with a good performance elevating a mediocre song into a top 5 spot or a bad performance leading a great song to NQ.
2. There's a lot that can happen, even outside of staging, in the month or so after OGAE votes close. Hype, media narratives and momentum from the mainstream media tend to get up to top speed in the couple of weeks before the contest, and these can be huge, at least in determining the winner – Conchita, Jamala and Salvador all seem to have benefited hugely from this.
3. OGAE members (and Eurovision fans in general) have different tastes to the general public, and more time to listen to the songs. While the fandom absolutely has a range of tastes, the average preferences are still quite different to the wider public at times, and the tastes of Eurovision fans can also be influenced by their extra knowledge of what's going on (returning singers, backstories, the meanings of lyrics, national final results etc.)
4. The scoring systems are different. In OGAE votes and most other fan polls, you've got 58 points to spread across your favourites, meaning you have a lot more room to reward stuff outside your top few favourites. This means that fan polls can favour the more widely-liked, agreeable entries – something coming 6th on every viewer's list would probably do great in an OGAE poll, but if viewers on the night only vote for their top 5, it won't be impactful enough to actually get those votes that matter.
5. Along the lines of the above point, OGAE polls don't have any juries, and the juries account for 50% of the final result in May. With jury criteria (generally seeming to focus on chartworthiness, vocal ability etc.) sometimes being quite different to televoter standards, that can put a gap between any televote-like preview poll and the actual scores on the night.
With all those factors – and probably more that I can't think of right now – I'd say that metrics like fan polls and odds actually do a surprisingly good job of predicting the final results!