Subject matter experts should create and oversee effective validation processes addressing bias-related challenges including noisy labelling (for example, mislabeled samples in training data), use of proxy variables, and performing system tests under optimal conditions unrepresentative of real-world deployment context.