1. You need a good cube. If you want it to last for awhile, get a DIY with cubesmith tiles; those will last forever if you apply them on correctly. Currently, I'm using a type D DIY from cube4you (driving to the owner's apartment = awesome) with Cubesmith stickers (orange switched out for purple).
2. Don't blame the method! Erik Akkersdijk has done sub-20 (or so I've heard) with only a beginner LBL solution ("beginner LBL" is quite redundant, though). However, I suggest you switch to Petrus or Fridrich. When learning those, make sure you try to intuitively come up with stuff to ease the learning process. That being said, don't stick with your first intuitive solutions: even though you know an algorithm to solve a certain case, you should look for the optimal solution.
3. Getting faster sometimes requires slowing down. This is a trap that some fall into. They get the idea that turning faster undoubtedly gets them faster times. While I'm not going to lie and say that it doesn't, it sometimes works against you. Ideally, there is some point in which you have a high turning speed while maintaining good look-ahead, and it doesn't occur at the ends of the spectrum (<1tps or >7 tps).
4. Lookahead first, then turning speed. Generally speaking, you'll do better on the left side of the optimal point than on the right side (see above).
5. Practice! Some people think that learning Fridrich will instantly get them sub-20 times or whatever crap. Remember, all of us practiced for a couple hundred hours at least to get to the fast speed we're at.
