How do you think as a data scientist when you are identifying a problem and working it into a prediction?

Easiest way to start is seeing what people do manually that is tedious in a work place. For example, people use to do all sorts of tedious tasks like hand counting items on the internet/etc, or repetitively filter fields on excel because of free text.

If you can build and train an NER or a text classifier for example, you can save people a hell load of time because it just saves genuine elbow grease.