Yes, you are right, it is much simpler when you have just a single number to "optimize" across population. And the correlations are strong indeed. But if you want to explain more variance, you need to reach for better tools.
We are all well familiar with people's various simplified models of the world. They tend to itch hackers, because they work well enough to not be automatically rejected by their users, yet hackers know, and sometimes even have proofs, that the models are ultimately wrong. The same thing happens with IQ. We all know it is mostly bullshit, but the truth is that it does work as a rough predictor of performance. It does explain some of the variance, not all of it.
We are all well familiar with people's various simplified models of the world. They tend to itch hackers, because they work well enough to not be automatically rejected by their users, yet hackers know, and sometimes even have proofs, that the models are ultimately wrong. The same thing happens with IQ. We all know it is mostly bullshit, but the truth is that it does work as a rough predictor of performance. It does explain some of the variance, not all of it.