Models can be wrong–in subtle ways undetectable to even trained analysts. There are many paths to obvious mistakes that experienced analysts make often (ask Dr. John Elder of Elder Research). For the unscrupulous, there are ways to make so-so or even bad models look good.
This is a major challenge for Analytics Executives. Outside consultants have a conflict of interest here which is why you must have a ‘second opinion’ about their work if you don’t know their scruples. It’s also a challenge for your own employees. With the science still emerging commercially (in non-science firms) and many new practitioners coming into the workplace, ethics and responsibility in letting the data say what it wants and nothing more (even if that means nothing at all) is a culture you need to support.
Data scientists also have a social responsibility. Models impact people, both your employees and your customers. They impact your investors and stockholders. They impact society in the ways whole industries use them (think banking). Remember the quote from George E.P. Box: “Essentially, all models are wrong, but some are useful.”
I highly recommend reading the posts from Rachel Schutt on this subject (October 4, 2012 · by Rachel Schutt · in Ethics and Humanity, Models) from Columbia University.
You must have an independent judgement about the ethics of your analytics practice.