The most well-regarded poll-based predictor, 538, gave Clinton a much lower than 90% chance and pointed out the key error of those predictors giving that high of a chance -- they used models which assumed, contrary to evidence, that deviations from polling results in different states in the same Presidential election are independent, when in fact they are strongly correlated.
seriously. fox news. cnn. msnbc. EVERYONE said it.
538 said it was 66/33 or 75/25 or something similar (i forget, but it wasn't 90/10 in favor of Clinton) and they got roasted over it when they don't even control any of the polls.