How to Engineer for Explainability: When a Black-Box is Just not Enough