There are two ways to get a game design - Intelligent Design, and Evolution.
I've always believed in ID, an orderly and reliable process. You start with an idea, something you care about and want to run. Then you gather half a dozen game designers, go to the pub with a notebook, and open a tab. A couple of hours later, you have a stack of ideas, all of which are awesome and some of which are practical, legal, or safe.
You take the notebook and spend a couple of hours putting together an initial ruleset yourself. Then you grab one or two other designers - in a coffee shop this time, not a pub - and give serious consideration to the mechanics, the balance, the logistics, and the safety of the new design.
Then you run it, collect feedback sheets, tweak it, and run it again. As soon as you start averaging four stars on the sheets, you have a new format. Science.
If you're evolving a game design, you start with the last step - by playing Calvinball. We started by tossing a frisbee around, and then people started throwing in new rules. On a successful catch, you must bury your feet in the sand. On a miss you must dive. Two catches means stand on one leg. A missed throw means a down-up. An interception means a race to the waterline and back. Good rules last; there's no discussion, but if a rule is boring, people stop observing it. Water breaks reset all the rules. It's fast-moving but not chaotic.
Somewhere along the line we grab a grip ball from our bag of random equipment (very important to evolving game design), and after a couple of lucky rules we have a hit. We stop evolving the game and start playing, developing skills and strategies until it's time for ice cream.