Earlier quoted context omitted.
They realized they could make money selling high quality question/answer pairs to machine learning companies building dialog systems. As a result the more pairs they have, the more their dataset is worth. Buy questions and resell as QA pairs.
Do they actually do this as a meaningful part of their revenue stream or are you saying this in jest?
I didn't appreciate how hard it was to come up with good QA pairs for a topic until IBM Watson acquired the company I was working for (Blekko) and got to see how things got made. I isn't something you can farm out to Mechanical Turk, and without AGI it isn't something that is easily auto generated from a knowledge graph. Ten thousand questions about the insurance business and vetted and ranked answers? Solid gold.