Back to glossary

Experiment Design

The structured process of creating controlled tests to validate product hypotheses. Good experiment design specifies the hypothesis, metrics, sample size, duration, success criteria, and potential confounds before the experiment begins.

Rigorous experiment design prevents teams from running tests that produce ambiguous or misleading results. Every experiment should start with a clearly stated hypothesis in the format: We believe that [change] will cause [effect] for [users] because [rationale]. The design should specify primary and secondary metrics, guardrail metrics that must not degrade, the minimum detectable effect, required sample size, and how long the experiment needs to run.

For AI-powered features, experiment design requires extra care because AI behavior can vary across user segments in unexpected ways. A recommendation engine might perform well for power users but confuse new users. Growth teams should design experiments that examine segment-level effects, not just aggregate results. Additionally, AI experiments often need longer run times because users may need time to build trust and adjust behavior. Pre-registration of hypotheses and success criteria prevents the temptation to cherry-pick favorable metrics after seeing results, which is especially important when AI features affect multiple metrics simultaneously.

Related Terms