Yeah, Bayes is useless without training, and users give up after awhile if a significant reduction in spam does not occur.
As a result I’ve turned off Bayes entirely and am focusing on tuning SA (via available plugins and tweaking rule scoring from the defaults). I’ll revisit Bayes once I’m confident SA is reasonably tuned with respect to default config.
Personally my own email accounts are 10-15 years old, but they’ve never been publicly displayed (i.e. not discoverable by scrapers). I get 1 or 2 spams a day, if that. Some clients on the other hand have heavily trafficked sites where the primary email address was exposed early on and is/has been in spammer databases for many years. These users just get hammered with spam, despite SA blocking the majority of it.
Right now we’re down to the nonsense messages with subject lines like, “Trump’s Plan to help Americans Earn More” with random garbled message body text, or 1 liners like “Hi, how are you? click here to [some scammer site]”. It’s the volume that’s the problem, users don’t care that the majority is getting blocked when the trickle is itself a daily time consuming process of deleting spam from one’s inbox.