I believe the GP's point was that if you follow the "remove anything that users rarely use in telemetry data" philosophy then you end up removing the restore feature of the backup software, defeating the entire purpose of backup software in the first place.
In other words, just because a feature is used rarely doesn't mean it isn't important to keep around. So much of what's wrong with software these days can be attributed to this "A/B test everything, the data never lies" approach to design.
