Minimum Detectable Effect (MDE)

The smallest true improvement that your experiment is realistically able to spot. Decide it before you start: “I only care about catching a lift of at least 1%; anything tinier isn't worth chasing.”

The MDE drives everything about your sample size — wanting to detect a 0.1% change needs vastly more users than detecting a 5% change, because tiny effects hide easily inside random noise. Beginners often set an unrealistically small MDE and then wonder why their test would need a billion users; pick an effect size that is both detectable and worth the effort.