In my primary dayjob such projects are usual. And they are really the mess, and the experience is worth a book :) But I'll try to be short.
At first accept in your mind that huge projects can't be perfect when you are alone warrior strugling. This will save your spirit from depression and months of nightmares with ugly controls and prototypes (I had). But you can make it a lot better.
Then try to get as much information about subject of matter as possible. Read books, ask experts and so on. For some projects it took more than month of immersion. Parallely, try to understand how the application works, sometimes even non-task-based documentation can give info about application tasks. Try to read between lines — why it was implemented? which goals it helps to reach?.
Write every supposition. Try to organize all your thoughts to structure and discuss it with participants. Techsupport usually can tell you a lot — not in direct way (all feedback will require interpretation), but you may get info about user's requests and reclamations.
Usually huge projects suffer in taxonomy, integrity and local (concrete intertfaces) solutions. This is where you can do a lot even without information about priorities and conceptual disadvantages of the system. So recombine, unify all interfaces, make simple solutions, hide all complex and individual solutions. It is possible to make concrete form implementation and navigation model better and simpler.
Also it is very important to make good relations with the team. If they are good enough, you may ask sales (or techsupport) department to send to clients short questionaire with open questions about application.
One important feature of huge professional projects is that they are function-based, not scenario-based. It is normal — MS Word is used by almost all users— and it is not possible to find personas and scenarios for all groups of them. And MS Word is not bad (but not ideal). So most of your attention should be payed to make any task as easy and automative, as possible, and navigation through all functions — as obvious as possible. But it doesn't mean that there are no vital sceanrios — try to search for them and you'll find them. But don't try to describe every – it is unreal.
Pay attention to competitor projects if they exist. If there are no subject matter straight competitors — there always do exist unobvious analogues similar in complexity and approach.
So, in sum, be detective — research unobvious, make suppositions, simplify & unify — and iterate, iterate, iterate.