Glossary · Testing · also guardrail, do-no-harm metric
What is a guardrail metric?
A guardrail metric is a measure a test must not harm, monitored alongside the metric the test is trying to move, so that a variant that lifts its target while damaging something else is stopped. Typical ecommerce guardrails are revenue per session, checkout errors, page speed and margin, and a breach ends the test whatever the primary result says.
- Reviewed 6 October 2026
In more detail
A test that optimises one number will happily trade away another. A discount lifts conversion and cuts margin; a pop-up lifts email sign-ups and slows the page; a new layout lifts add-to-cart and breaks checkout on one browser. Guardrails are the list of things the test is not allowed to make worse.
Guardrails are checked on every read, and the threshold for stopping is set before the test starts. They are deliberately one-sided: a guardrail looks only for harm, so a breach ends the test even when the primary metric is winning.
Data-quality checks such as sample ratio mismatch belong on the list too. They do not measure the business, but a breach means no other number from the test can be trusted.
Key points
- Watches for harm, not for wins.
- Set before the test starts, checked on every read.
- A breach stops the test regardless of the primary metric.
How Liftable relates
Sources
- Kohavi, Tang and Xu, Trustworthy Online Controlled Experiments (Cambridge University Press, 2020). Standard reference on online testing practice, including guardrail metrics, sample ratio mismatch and the cost of peeking.
Related terms
Guides that go deeper
Last reviewed against the product on 6 October 2026.
Scan your store.
See where your store is losing revenue and what to fix first. It is free, takes about a minute, and needs no install.