If your ml model is a magic black box and there's zero visibility into the system, even internally, as to the signals that resulted in a decision, you've hired a monkey that can run a tensorflow tutorial, not actually someone that understands machine learning.