Routing a game for a speedrun

1. Determine what is unskippable within the game.
2. Assume everything in the game is not required, even those that are required.
3. Starting from the end, work backwards to see what pre-requisites for certain tricks, timesavers and progressions are.
4. Once you have a minimum path to beat the game, time everything that seems reasonable as a detour from that path. If you need 4 collectibles out of 6 in a certain room, you'd time entrance to exit with 0 collectibles, and then each collectible as a detour from the main path and from each other. From that data, you can piece together what the minimum detour time is. Do note that this approach also applies to revisiting areas: you'd simply time the detour from a different path.
5. Do a test run. Does anything feel slow? Did something not seem necessary? If there's items or experience, was there anywhere that having more of those would save more time than the detour to collect them?
6. Tweak the route based on the observations from step 5.
7. Go back to step 5 until the route has no clear large time losses.
8. If there's another completely different route (say, skipping a sidequest vs. completing the sidequest for a benefit) go back to step 4 for this route.
9. Repeat step 8 until you've exhausted all reasonable ideas.
10. Retime the best different routes. You're probably better at the game now, redoing the earlier routes with the same skill level as the later routes makes it less likely that you'll miss something.
11. You now have a route. Don't assume this route is optimal. Collaborations help. New tricks will almost always have an impact on the route, sometimes well beyond (both before and after) the small area where it was found. For example, if a new strategy for a boss in an RPG is found allowing victory at a much lower level, you have to take into account the time saved by not leveling beforehand and the time lost by being a lower level for the rest of the game. One small improvement can drastically change the best route, allowing other tradeoffs.
12. Consistency. Sometimes the best route won't be faster most of the time. Depending on how far into the run it is (early is usually worth resetting over the luck, late isn't worth going for except as a last ditch effort) slower but safer alternatives are worth using in runs. This routing process will also discover most of these alternate routes.

Yes, this process takes a while and there will sometimes be an obvious "best path" (or only a few worth considering). You get a feel for what works better the more times you've gone through the process & played the game you're routing.