I usually begin with a basic idea for a setting, which I then expand into a
description of a few pages, mainly the general framework without any de-
tails.
When this is finished, I hand it to the players and ask them what they think
of it, what should be added or deleted, and what kind of characters could
have what kind of adventures in this setting.
If the players like the setting, this input from the players is then used to
complete the setting, now with the background details I do not want the
players to know, usually a description of about ten to twenty pages.
In most cases I also post the difficult parts of a setting on a forum to ask
for some more interesting ideas, especially ones that are different enough
from my usual "style" to surprise the players.
If necessary I then write the modifications of the rules required by the
setting, including for example altered or new careers, different technolo-
gies and thelike, and give the players these informations, too.
Once the players have created their characters and we have made sure
that they fit into the setting and can work together as a group, we play an
introductory adventure to "implant" the characters into the setting and to
test the modified parts of the rules.
This introductory adventure is the only pre-planned adventure of the cam-
paign, afterwards the setting becomes a "sandbox", and the players are
free to let their characters do whatever they like.
This approach still requires some preparations, so at the end of each ses-
sion the players inform me about their characters' plans for the next ses-
sion. This allows me to design the necessary NPCs, locations and all that.
My "philosophy" during the game is quite simple: The better detailed the
setting, the less rules and dice rolls are needed, because many decisions
can be made by following the "inner logic" of the setting, for example the
motivations of the NPCs and such.
Dice are only used when there is no obvious plausible development of a
situation and when the outcome is important for the story, otherwise the
rule is: "What is most likely is what happens".