The problem with collaborative filtering systems, whether based on personal preferences or on past purchases, is that you end up in a recommendation bubble. Seeing the same items recommended over and over again.
Over time the things I want to buy change. Perhaps because I had a major life event - got married, had a child, turned 50 and had a mid-life crisis, etc. The systems can't predict the timing of those because people don't do them on a schedule.
It's an exploration vs exploitation trade-off. They are just predicting the things you are most likely to buy and suggesting those. But you also want to throw in things which you are less likely to look at, but if you do it gives them more information to make better suggestions. And then as you mention, account for the fact people are dynamic and change over time.
Over time the things I want to buy change. Perhaps because I had a major life event - got married, had a child, turned 50 and had a mid-life crisis, etc. The systems can't predict the timing of those because people don't do them on a schedule.