Turnout is key.
Turnout can never be predicted by polls since people often tell pollsters they will turn out when they don’t do it.
This flaw is ridiculously easy for pollsters to discover.
All they need to do (in a state, for example) is ask voters whether they voted in the 2016 presidential election.
Then compare the polled turnout numbers to the actual turnout numbers.
You can bet the polled “turnout” is a lot higher than the actual turnout.
Not if their jobs are dependent upon them not discovering it. :)