BMG had the least accurate of all the polls on the day before the General Election, and published EU membership predictions
varying from +3/-10/+8 in a short period of time, so a single result from them is not really sufficient evidence.
However, their own write-up on this poll is interesting, and indicates that people not voting in the referendum were split 50:50 a year ago and are now 80:20 remain, (whilst at the same time the graph there also suggests it's not just previously undecided people causing the difference).