Casual Conversation

One thing that I expected to be absolutely amazing in 2024 from online vendors was product recommendations.

That vendor, assuming you use a single, persistent account to do purchasing, has a full list of your purchase history. They may well also have browsing data.

And so, given all that data to mine and analyze, one of the few places where I actually have tried to see what a vendor can do in terms of analyzing my preferences...has been really unimpressive.

I'm mostly thinking of Amazon and Steam, since they're the online vendors that I use the most; Steam in particular has a considerable amount of data it can gather, including video game playtime.

Yet even though Amazon grabs some eyeball space on every page to try to recommend products, I have rarely been recommended anything I actually want to buy on Amazon. Occasionally, sure, but virtually everything I get is via plain old searching. And the most-successful recommendation approach Amazon uses, by far, is just asking me whether I want to purchase more of something that I've purchased in the past. I'll grant that maybe there's subtlety there that I can't appreciate from the outside, like computing frequency at which a given "repurchase" recommendation happens or taking into account past average purchase frequency, but it doesn't seem like the most-sophisticated form of recommendation.

Granted, I normally make it a point to limit Amazon's data-gathering. I browse logged out, make a list of what I want to buy, clear browser state, and log in only long enough to make a purchase. That probably makes it harder for Amazon to associate me with my browsing behavior. But it does know what I actually buy. And it has a pretty substantial history there.

And for Steam, Valve knows what games I play, how long I've played them for, and assuming that there's any mining based on game achievements, even -- at least as an abstract concept that would permit for correlating preference across video games -- what I do in those games. Like, players who get "evil path" achievements in one game maybe prefer video games with "evil" routes, stuff like that. But I have browsed Steam's discovery queue zillions of times, and while I've probably found a game or two on there, the success rate of its recommendations is abysmally low. Probably the most-useful recommendations system on Steam is the "similar games" section when viewing information about a game. But I'm pretty sure that most games I find on Steam that I actually like are just by using user ratings and searching for tags. While, Steam's scoring is opaque, and it's possible that they're using some degree of input, I don't think that it's making use of information about me there. I wouldn't be surprised if it's nothing more than ranking games based on their player review score, which...isn't much more than things like MetaCritic and similar have done. I've occasionally had luck looking for games that have very high hours played, with the idea that people wouldn't play a game a lot if they didn't like it. That makes some use of aggregate data about users, but not about me.

Most video games that I get on Steam that I like are games that I've discovered somewhere other than on Steam, often looking for human "roundup" articles comparing collections of similar video games and giving a brief blurb about pros and cons. That's not new technology.

That comes as a very great surprise to me, when one considers the enormous amount of effort and resources that goes into harvesting and mining data about people. Now, okay, a lot of that is for ads. And advertising isn't exactly the same thing as doing good product recommendation. An advertisement is trying to effectively get someone to buy a product regardless of whether they'll ultimately like it or not, whereas a product recommendation -- at least in the ideal, user-focused sense -- is trying to find products that people will like. But there has to be a substantial amount of overlap between the two. Advertisers don't want to waste money advertising to people who won't buy their product, so trying to find people who are interested in their product is a major part of advertising.

I haven't used any systems that log my music-playing and make recommendations; I'd rather keep my privacy there. Perhaps if I did, that area would be more-successful.

But by and large, it's an area that I'm very surprised is not more successful than it is. It's a "flying cars and jetpacks" thing, something that I'd always vaguely expected of the future, but which never seemed to really arrive. Product recommendation systems never really got to the point of anticipating my needs very effectively, even where they have what I'd consider a fair amount of data to work with.

What's your experience? Does it differ from my own? Do you find that product recommendations from vendors are really useful, pretty much hit the nail on the head for what you want? How do you "find" products? Am I missing something, maybe like merchants on Amazon or publishers on Steam trying to game the recommendations system one way or another, and poisoning its inputs?