I once did a machine learning project at Intel. The end result was that it was no better than simple statistics; but the statistics were easier to understand and explain.

I realized the machine learning project was a "solution in search of a problem," and left.