I start a paper draft very early in the process.
Sometimes it is started before even a single bit of data has been collected.
It starts, often, with some literature that I am reading that starts to gel an idea. So I'll jot down the author/date and some words related to my thinking at the moment. Could be a full manuscript ready sentence, sometimes just a few words.
At this point I don't even insert "introduction" and "discussion" headings because I'm not sure where it is going. As time goes on there will be a tipping point where I take an hour to put in the structure.
Title page, headings, maybe some cut and paste methods that we'll be modifying later.
I didn't use to do this, but I have gotten better about writing up figures as they roll off the assembly line. Even before I know the end analysis, etc. So maybe I waste a little time if I have to redo analysis with more groups or something and reconfigure the graph.
I've found that it helps me later to know what we have and what we don't have.
So now I might actually start a draft around a key figure that I really like. Stare at that graph in the Word file and the ideas start coming.
The key for me is to trigger early on just getting some words down on the paper in approximation of what I am thinking. At the moment.
Thoughts often change. I write many times more words in the drafts than will ever appear anywhere in print.
This helps me to think. To see.