Calculating Required Sample Size

Before launching an A/B test, your team asks: 'How many users do we need per variant?' What four inputs does a standard power analysis need to answer that question, and what happens to the required sample size if you want to detect a smaller effect?