A/B testing compares two variants with real users. It's tested in almost every product-focused ML interview.
Design steps:
Define metric (conversion rate, revenue, engagement)
Calculate sample size (based on minimum detectable effect, α, power)
Randomize users to control/treatment
Run until sample size reached (don't peek!)
Analyze with appropriate statistical test
Interview question: "Design an A/B test for a new ranking algorithm."
Metric: click-through rate. Unit: user (not pageview). Duration: weeks minimum (weekly patterns). Watch for novelty effects. Consider guardrail metrics.