I usually have a feeling of what I want, and then look at my list of player motivations (25 of them) to see what tools that I should pick from. I also think about what kind of uncertainties (from a list I made) that would be useful for that particular game.
Sometimes I just pick player motivations at random to combine them and wonder how that game should look like. This have made me create some games I had to learn to enjoy because I didn’t know if they would be fun while I was creating them.
Or I just rip a mechanic, and wonder what kind of setting it would fit to.
… but normally, it’s a combination of these three above, or I just go another way.
It’s important, for me, to not create enough from the start so the players can bring their own ideas into the playtest, and then I just iterate while adding more components. I constantly go back to my design thoughts, to either steer the game mechanics back to it or realize that what I got now is better than I imagined at the start.