A proper growth experiment starts with a written hypothesis — if we do X, we expect metric Y to move by roughly Z, because of this specific reason — not just ‘let's try this and see’. Define the success threshold and the minimum run time before you launch, not after you see early results, since eyeballing results mid-flight is the single fastest way to fool yourself with noise.
Then actually decide: kill it, scale it, or iterate — and write down which, with the reasoning, so the next quarter isn't relitigating the same test. Teams that skip the documented decision step tend to re-run the same failed experiment under a new name a year later.