Personalization versus Customization

A lot of things in online retail have changed in the last some years. The biggest is that we have been told that the online buyer now doesn't like take-it-or-leave-it experiences. I'm not sure anyone's tested it out (e.g. do users buy from suboptimal websites if, say, the prices are slightly lower there?), but let's say it makes logical sense. Another equally big thing that has changed is our ability to create multiple experiences on the same website, without which the first insight is useless. The third thing to keep in mind as background is what I may have said before in the Escalator Problem post, most shopping sites are built by techies, not retailers. Techies love complexity. Multiple experiences, multiple skins on the same backbone is exciting and cool for techies.

As a result of all the above and more, the hue and cry for Personalization.

There's one small school for customization and another big school for personalization, to be clearer, and the second school considers vanilla customization-ability uncool. Custom flows are about giving the user a choice of experience that the user explicitly makes (e.g. you want to see more deals? more lifestyle or tech? when you log in?) while Personalized flows second-guess the user (e.g. we know that you have bought / browsed T-shirts, so here's more T-shirts when you log in). It is cooler to not ask the user but surprise him / her with what we know without being told.

I'm not sure what the best approach is. Personalization could be cooler but could just be techie-cool and not user-cool. I believe customization is safer, not sure if that's the best. Here are a few things to remember whenever, if ever, we as businesses make this call.

a) Users may actually like being asked questions when they walk in. Maybe it's a nice thing to ask them if they like seeing deals on furniture better than music instruments.
b) We know some people mind being second-guessed. We all know the Target example, what happens when your analytics team figures out someone's daughter is pregnant without that someone knowing about it.
c) There could be parts of the user's browsing or buying that (s)he doesn't want recorded for a number of reasons you can imagine
d) Preferences change, with time, with occasion and so on. You may know the buyer's history, but that may not be a good predictor of his / her present state of mind. For that matter, even people change over time.
e) There could be things the user is done with. Maybe (s)he's already purchased the dream double-bed from your site or elsewhere, and now your snowing him / her with latest double-beds is not just useless but harmful. The next time the same person will buy a double-bed is either ten years later or never.
f) There is a distinct possibility, and probability of multiple-user-ids e.g. the son using his Dad's id or the wife using her husband's - maybe it is faster than creating a new id or maybe that someone's received a coupon the other person wants to use. In this case you'd be personalizing for an average of the husband and the wife, a person who doesn't exist, and not impressing them both.
g) Personalization needs data (if not Big-Data, the buzzier, fuzzier word) and if a lot of your users are new users without a lot of data history, you'll have a cold start problem. By the time you figure out the person's preferences, you may have already pissed him off with irrelevant suggestions

I'm sure a lot of statisticians and coders are working on these angles as we speak, but we have buyers walking in now. What should we do? Any views?