The issue is when there is overdetermination - fixing a single cause resolves 80% of the current cause of failures, but the process fails at another point which becomes the new bottleneck. Say, most people are failing at a task due to stress(A). But even if you solve that, then most people can start failing due to another cause, B, say laziness. Fixing A is not "a little bit" and any solution will have to face A(in order to have success rate > 20%) but it is not sufficient to fix stress alone.
You can of course agglomerate A & B into a single cause C but for that to be meaningful C has to be the 'root' of A and B, otherwise the solution for C also just becomes an agglomeration of the solutions for A and B.