Modern IT systems are complicated. Sometimes it is very complicated to describe in words conditions that lead to failures. It is even harder to describe these conditions using formal language of triggers.
No doubts, triggers are very good and efficient in simple cases. But from time to time problems may sneak below radars. Let's say, all triggers were in OK state, no notifications were sent, but some kind of mysterious failure occurred and we want to prevent it in the future.
As said, defining complex problem situation with triggers is not trivial. In this situation user should have a possibility to tell Zabbix that problem took place at time moment X (and probably specify which parameters are most likely responsible for the failure). Zabbix should be able to memorize such problem states and perform on-the-fly comparison of current situation with memorized problem states.
This is not a replacement for usual triggers. This should be a complement. Semi-automatic, self-educating system.
Details of a possible implementation will follow in the comments.