ML in production is an ever growing part of the final decision making. That’s why - just like every decision made by humans - a decision made by a model has to be traceable and accountable, such that people can verify and reproduce the outcomes.