Depending on the data you collect, targeting by user - unfortunately - works. If the granularity is not one user, it will be a hundred. If not, a thousand, and so on. I've seen apps run ads targeting a total of 5 cohorts(together holding a hundred million users), and I've seen companies run ads targeting 100s of cohorts with the same number of users. They all work better than no targeting at all.

However what you're saying isn't completely wrong. I've also seen user targeting become a self-fulfilling prophecy. What happens is that it's championed by a high level executive as the panacea for improving revenue, implemented, and seen to not work. Now, as we all now, the C*O is Always Correct, so everything else around it is modified until the user-level targeting A/B test shows positive results. Usually this ends up in the product being tortured into an unusable mess.