A practical way to approach approach is to decide what success looks like before you change anything. Otherwise it's easy to keep tweaking without knowing whether you're actually improving it. I would trust a repeatable result more than one impressive run.
