I suspect you missed the point of the article, which was that the telemetry measured the wrong level of abstraction in users' tasks.