Apart from the matchmaking, I believe there is a fundamental issue with the GA: it's practically (and in theory) stale.
For the entire time since it had been rebuilt, I (as a f2p player) was moving up and down within certain limits (Aurodium 2-3 and up to Kyber 4 – and always up and down again between these). But there is absolutely no chance to get out of your boundaries except you're spending money to get ahead of other f2p players, or relying on external factors, such other players abandoning the game.
Just hypothetically, if every player in the game is f2p and everyone grows their roster at the same speed, this means, your opponents (so everyone else in your range) are growing the same way as you do. And the players who are way ahead of you do so too. The only difference between real-life and this hypothetical scenario is that people have different rosters, different mods, different defense/offense teams, more or less GLs, luck, being active/inactive and so on. These variables affect your GA results.
However: Once you either go up the ranks, you eventually reach a threshold where your opponents are just simply way more powerful (such as having 3-4 more GLs than you do, have twice the GP, etc), or you've been pushed back far enough to be able to beat your opponents again.
In general, you stay within these boundaries (under the assumption that in average your opponents grow in the same speed as you do.
On top of that, the higher leagues (or divisions) yield higher crystal income, increasing the gap even more (the higher leagues you can grow quicker). I'm not complaining about the crystal income. I'm just tired, seeing that it's impossible for me to escape these boundaries (except by going p2w, or – conversely – being inactive to drop down the ranks).
I remember CG promised (after they introduced the new GA) to constantly monitor this and add improvements. Would be great to see CG addressing this in the future. Once possible change to improve this would be to finally improve the matchmaking (which had been promised all along), such as better consideration of GL count, GP difference etc.