A/B Testing

Attach an experiment to any flag to measure which variant actually drives more conversions.

How experiments work

An experiment links a flag's variants to a goal event you care about — for example, "signed up" or "clicked checkout". The SDK records which variant each user sees (an impression) and which users later complete the goal (a conversion). Vexil calculates the conversion rate per variant and surfaces it in the dashboard.

Experiment lifecycle

Creating an experiment

1. 1
   Open a flag detail page

   Click on the flag you want to experiment on.
2. 2
   Go to the A/B Tests tab

   You'll see any existing experiments for this flag.
3. 3
   Click "New experiment"

   The creation form opens.
4. 4
   Set a name and goal event

   Give the experiment a name and enter a goal event — a short string like signup or purchase. This is what you\'ll track in your code.
5. 5
   Choose an environment

   Experiments are scoped to one environment. Use Staging to test before running in Production.
6. 6
   Define variants

   Each variant has a value (the flag value it represents) and a label like "Control" or "Variant A".
7. 7
   Click "Start"

   The experiment begins collecting data immediately.

Tracking in code

After evaluating the flag, tell Vexil which variant the user saw. Then, when they complete the goal, record the conversion:

// 1. Evaluate the flag and record the impression
const variant = await ff.evaluate<string>('hero-cta', { userId })
await ff.impression('hero-cta-experiment-id', {
  userId,
  variantValue: variant,
})

// 2. Later, when the user completes the goal
await ff.track('signup', { userId })