ELO does not work for a team game. Period. They are still attempting to avoid using personal performance stats because they started from ELO as a design basis.
Google this: “Elo vs. Regression to the Mean: A Theoretical Comparison”
They may have tried to introduce some additional design elements and a little personal performance but their whole theory is wrong from the get-go. That’s why matchmaker is broken. Bad theory. That’s all.
There is a solution though: Two different methods of ranking and a slider between them. At low ranks you use simple personal metrics and almost ignore win/loss but at medium to high ranks you switch back to the current system that is ELO-like and mainly ignores personal in-game performance.
They are too afraid that actual individual personal performance stats will reward bad and selfish behaviors. Sure. Okay. Fine. …and True.
…but teamplay doesn’t even exist at low ranks while massive chaos from smurfs, children, newbies, drunks, and permanently low-skill players does. Personal performance ranking stats is the only thing that will tone down the chaos.
Biggest importance and biggest difference at low rank is personal basic skill. It flips to team coordination, timing, game knowledge, and other subtle things at med-to-high rank. The MMR system must do this too!