Some kind of system that detects missing data, data of the wrong datatype, detects data that is duplicate, detects data that does not "match" other data in the same column etc in some input datastructure. It could then visualize this in an easy manner on a dashboard, together with a way of sorting the datastructure to only show specific "potential errors".

Would make model creation and validation so much easier. This could then be used as an input to a potential risk-model validation system further down the road.