with data science, statistics and machine learning, the debugging cycles are much longer and interpreting data and picking estimator functions, etc. is an artform that requires attention to detail and "deep thought". Running a solution/query could take thousands of dollars in cloud compute cycles, so you'd better be right.