There's an important but confusing distinction that is the key to unlocking part of this apparent mystery. Namely, we need to be very careful about what we mean by "consequentialist justifications." This phrase (taken outside of any context) is ambiguous between:
- Justifying something based on the consequences it led to.
- Justifying something based on the consequences we expect it would lead to.
To make this distinction a little more practical, let's say some kids are playing baseball and one of them smashes a window. If "consequentialist justifications" means the former, then if the kids discover the person in the house was having a heart attack and save that person's life, then hitting the ball such that it smashed the window is a good thing, because it led to good consequences. If "consequentialist justifications" means the latter, then while it's good that we were able to save someone's life, it's wrong to play baseball in places/ways where you're likely to smash other people's windows cetera paribus.
The above distinction is a real problem for consequentialist views, because presumably we neither want good/bad to just be luck (in the purest form of 1) in which case we can have no moral rules nor do we want it to make it wholly depend on our expectations -- because we know we are not infallible about future events.
Kant wholly rejects anything like the former, but he also rejects the latter and the passage you're quoting doesn't require him to accept the latter due to some pretty fancy footwork.
That footwork is as follows. Kant thinks we can figure out what the categorical imperative would have us do in several ways. One way is that we can imagine what happens if everyone does something. In this case, we are looking at "consequences" but we're not looking at actual consequences; instead, we are using reason to consider logical consequences, and our primary concern is logical outcomes. So we ask "could we consistently will a world where everyone is smashing the windows of other's homes?" Presumably, the answer is no. Thus, for Kant, it's wrong for kids to play baseball in this way regardless of what might follow from them doing so, i.e. regardless of the actual consequences.
A second way that Kant thinks we can work out the details of what we should do is the "formula of humanity" which says that we must always treat rationality in ourselves and others as an end and never merely as a means. Here, the test is "would I in my action be treating another person as a mere means?" Again, the question is not about actual consequences but about the logical arrangement.
Turning to the specific text, we need to keep the above distinction in mind and remember what Kant is trying to reject. This can be evaluated either in the formula of humanity manner or in the universal law manner. Starting with the formula of humanity, the task to do things that sustain rationality is obligated not because the specific act sustains humanity in that person but because the obligation exists before the specific act. That's a bit abstract, but the basic point is this: I need to help people not because in the moment I see that helping them would extend rationality, but because logically I am obligated to extend rationality and this requires me to act in ways that I would conceivably imagine extend rationality.
One interesting and bizarre consequence is that this avoidance of focus on actual consequences -- and even on locally expected consequences -- leads Kant to say we should be honest to people we have every reason to believe are dishonest (Cf. On a Supposed Right to Lie ...).
Why does Kant think this? To get to the bottom of it, we need to go back to Critique of Pure Reason and the third antinomy. As a rough sketch, Kant accepts Newtonian physics and much of Hume's ideas about where that leaves us in terms of things in the world. He joins to this a belief that there's a mind that operates outside of all of this (like the rationalists) and imposes order on this world through putting it under forms and categories. Outcome: everything in the world is deterministic, but the rational self is not in the world. So despite everything I see or evidence I have, I really can't know what another person will do.
tl;dr for Kant, we don't pick (moral) actions because of consequences even though actions do have consequences. We can look at the logical consequences of actions to determine whether their outcomes would be immoral but this is not an evaluation of particular actions. Thus, to help people does have the consequence of assisting them, but we don't determine the morality of helping based on whether or not it in fact assisted them (and we have to be very careful about what sort of consequences we mean when we are saying we should do what we expect would help them).
