This
Politico article outlines the thinking within the Clinton camp prior to the election: they placed an unreasonable amount of faith in data and let that dictate their campaign strategy. We'll get to why that's a preposterous plan later and why polling numbers are usually skewed.
Huffington Post tweet the day before voting, based on polling numbers, placed Clinton's chances of victory at 98%. They also claimed that Clinton would get
323 Electoral Votes based on the data.
A
Princeton Election Consortium survey, again based on polls (where this entire argument started yesterday) nationwide, placed her probability of victory at 99%. This
Stanford projection, again based on polls nationwide, stated Clinton had a 99% chance of victory. The
New York Times Upshot, a supposed major source for accurate projections, suggested Clinton had a 91% chance of winning. Even supposed rock solid sources like
Five Thirty Eight had the Electoral Vote numbers completely in reverse with Trump projected to get only 235 to Clinton's 302. This
Reuters poll suggested Clinton had a ~90% chance of success and signalled something similar with Electoral Votes heavily in Clinton's favour, with the complete role-reversal on actual Election Day.
CNN allowed data from internet users to suggest Clinton had a 91% chance of winning last November (admittedly not a poll per se, but an example of restrictive data being distorted to suit an agenda).
My point about demographics never being equal so the waters are always muddied in polls, intentional or otherwise, still stands. For example,
CNN polls (flick to their 'Methodology' section on page 19 of the document to see the breakdown) regularly oversample Democrats in their polls, sometimes by ~1/3 compared to Republicans (in this case 32% identified as Democrat and 24% Republican).
Overall, and my point from the beginning, is that failed predictions based purely on numbers suggest the rush to exploit data may have outgrown the ability to recognise its limitations.