A/B testing and measuring -- are you kidding me? That's how you go from 35% conversion to 36% conversion, not from 2 daily active users to 1e7 daily active users. Refinement only works if you're exploring a space that has productive alternatives in low dimensions, which is almost never the case. No amount of "progressive refinement" is going to turn Farmville into Candy Crush or Candy Crush into Fortnite. You have to look no further than those examples to see what I mean -- "Oh, look, Zygna has figured out a formula to hook users" followed by "Zygna can't repeat its success".
A developer of traditional games might think, "How can I make this game better in an aesthetic, or an educational sense?" Or, "How can I make this game a better FPS than the ones before?"
A developer of another game might think, "I want to make money, let me survey existing games and read psych literature and create the next billion dollar game. I don't care what form the game takes, or what players get out of it."
The claim is not that game developers for each game are creating an addictive game from scratch by A-B testing, but rather exactly what you said, that they're AB testing for smaller improvements in eyeball-retention, spending, and addictiveness, because all they care about are those metrics which drive profits (or valuation which can be converted into profits).
That a set of conscious design choices, on its own, isn't sufficient for extreme addictiveness doesn't change the fact that those conscious design choices are necessary, and those choices are being examined, analyzed, and documented by other game devs and by psychologists. So, of course you can take all the known lessons and apply them to a brand new game concept and fail horribly, but if your game is a success on other grounds and you mined other games and the literature for addictive ideas, you are much more likely to have a billion dollar super-addictive game than you would have been 20 or 30 years ago.
Zynga had repeated success, untill market saturated and other developers caught up. That does not make these tactics not working, that just makes them so common that they are not such a huge advantage anymore.
I'd say, until people got bored with the same single local optimum game that all Zynga games converged to despite differnt skins.
It's also how you go from unique and successful entertainment product to something completely unusable and boring is just few hundred A/B iterations.
Enjoyment is not the space where gradient descent works particularily well.
Amazon is absolutely horrible expeirience for me. It's hard for me to point e-commerece site I used that's worse than amazon.
