From my experience building very expensive industrial machines, the gain is more like 2x-4x not 20x, and that assumes that the game of telephone in the engineering pipeline actually gets it right.

In a recent HN discussion, someone suggested that step one of emulating successful tech companies (Netflix was used as an example) was to fire half your workforce and pay the remaining half double. That logic meshes with my experiences; the machines made you more efficient, but you needed more intelligence to use them than the tools they replaced.