I went through this same question last year with a couple of accounts, before I trusted my own conclusion. Strip out the noise and how much the interface pushes a subscription shapes the first fortnight more than anything. The same signal appeared fairly clearly when I compared like for like.