Guardrail Metric

What it means

A guardrail metric is a measure you watch alongside an experiment’s primary metric to protect against unintended damage. Where the primary metric answers did this change win, the guardrail answers did it break anything important. Typical guardrails include page load time, error rate, crash rate, checkout completion, and unsubscribe rate. A variant that lifts clicks but tanks a guardrail is usually not worth shipping.

Why it matters

Optimizing a single number invites tunnel vision. A flashier call to action might raise sign-ups while quietly slowing the page or driving refunds up, and a team looking only at sign-ups would never notice. Guardrail metrics encode the constraints you are unwilling to sacrifice, so the experimentation platform can flag regressions automatically and stop a harmful rollout before it spreads. They turn win rate into net value.

In practice

Suppose you test a heavier hero image that lifts engagement. You set page load time and bounce rate as guardrails. The platform alerts you when load time crosses a threshold, and you either optimize the asset or abandon the variant. Experimentation tools like Statsig, PostHog, and Optimizely let you attach guardrail metrics to any test and will warn when a variant moves one the wrong way. Our guide to running an A/B test covers choosing them before you launch.