Experiences with approximating questions in Microsoft’s manufacturing big-data groups