An analytics engine / machine learning tool has to begin somewhere. Why not start on a small(er) data set you are passionate about (and that has immediate benefit to your business model and user base), that you can record and measure more easily, and that is not generally obtainable by many others, before unleashing it on the public markets where people will be less tolerant of inaccuracies.

Mighty oaks begin as small acorns.