Secondary Metrics

A vs B gives every metric on an experiment one of three roles: primary, secondary, or guardrail. Your primary metric is the one metric your experiment is judged on. A secondary metric is one you track for context, not for a decision. It never decides who wins.

The three roles, in one place

Every metric you attach to an experiment plays one of these parts:

• Primary: the one metric the decision rests on. Every experiment has exactly one.
• Secondary: tracked for context. Read alongside the primary, but it never decides the result.
• Guardrail: a safety check on a metric you protect, not one you try to improve, like page load time or refund rate. If a guardrail gets meaningfully worse, A vs B blocks a "Ship" verdict, whatever the primary metric says.

This page covers secondary metrics. If you want a metric that stops a launch when it gets worse, that is a job for the guardrail role, not secondary. See Guardrail Metrics for how that works.

What a secondary metric is for

A variation is one specific version being tested: control, or a challenger. When a variation crosses your decision threshold on the primary metric, A vs B declares it the winner.

A secondary metric runs alongside the primary metric. A vs B measures it and shows it in full, but it never enters that decision. Think of it as a health check on your primary result. It answers a simple question: "We won on the primary metric, but did anything else move?"

Reading secondary metrics on the Results page

Secondary metrics appear in sections below the primary metric section on the Results page. Click a section to open it. Each row inside shows:

• Visitor counts.
• Conversions (or events, for a continuous metric).
• The conversion rate (or value per visitor).
• Lift versus control.
• A significance column, which depends on your stats engine.

The Bayesian engine is A vs B's stats engine that reports results as a probability of beating control. On Bayesian, that column shows exactly that: a probability to beat control. On Frequentist, it shows a p-value instead: the odds of seeing a result this extreme if the variation changed nothing. On Sequential, it shows a Safe-to-stop or Inconclusive call. On Bayesian experiments the column is headed Prob. to Beat Control; on Frequentist and Sequential it is headed Significance. The primary metric table and the revenue cards use the same heading, so every table on the page agrees.

A metric with revenue data also shows Revenue and average order value columns.

Good ways to use secondary metrics

Funnel metrics

A common setup: make a bottom-of-funnel metric your primary, such as "Purchase completed". Add upper-funnel metrics as secondary, such as "Add to cart click" and "Checkout page view". This shows you whether a win on purchases came from the whole funnel improving, or from one step alone.

Example: purchases improve 15% and add-to-cart clicks improve 18%. That pattern suggests visitors' intent to buy rose across the board. Now imagine purchases improve 15% but add-to-cart clicks stay flat: something about checkout itself, not the product page, likely drove the change.

Revenue alongside conversions

Say your primary metric is a conversion event with no revenue attached, like a click or a page view. Add a custom event that does carry revenue as a secondary metric. That gives you a rough revenue view too.

For example: primary = "Checkout page visit", secondary = "Purchase completed with revenue". You get a fast-moving primary metric and a revenue picture, in one experiment.

Adding a metric and setting its role

Metrics are attached in Step 4 (Metrics) of the experiment builder:

1. Search the metric list and select the one you want. The first metric you attach becomes your primary right away.
2. Every metric you attach after that starts as secondary.
3. Under each attached metric, a role picker lets you change it to Primary, Secondary, or Guardrail. Choosing Guardrail reveals a safety-margin field.
4. To make a different metric primary later, drag it to the top of the list, or promote it from the role picker. A vs B always keeps exactly one primary and steps the old one down on its own.

You can attach as many secondary metrics as you like. A focused list (3 to 5 metrics) is easier to read than a long one.