A/B Experiments

OwlMetry supports lightweight client-side A/B experiments. You define experiment names and variant options, and the SDK handles random assignment, persistence, and attaching assignments to every event automatically.

Add an OwlMetry A/B experiment to this project.
Run `owlmetry skills` to find the SDK skill file.

- Use getVariant("experiment-name", options) to randomly assign
  users to a variant on first call (persisted automatically).
- Render different UI or behavior based on the returned variant.
- All events are auto-tagged with experiment assignments — query
  funnel or metric data segmented by variant to compare results.
- No server setup needed — experiments are defined in app code.

How It Works

Experiments in OwlMetry are client-side only. There is no server-side experiment configuration or central experiment registry. You define experiments in your app code, and the SDK manages variant assignment and persistence locally.

Every event emitted by an SDK automatically includes an experiments field containing the user's current variant assignments:

{
  "experiments": {
    "checkout-redesign": "variant-b",
    "pricing-page": "control"
  }
}

This means you can segment any data -- events, metrics, funnels -- by experiment variant without additional instrumentation.

Variant Assignment

getVariant(name, options)

Returns a variant for the named experiment. On the first call, a variant is randomly selected from the provided options and persisted. Subsequent calls with the same experiment name return the previously assigned variant, regardless of what options are passed.

getVariant("checkout-redesign", ["control", "variant-a", "variant-b"])
// → "variant-b" (randomly assigned on first call, same value on all future calls)

Assignment is random with equal probability across all options.

setExperiment(name, variant)

Force-sets a variant for an experiment, overwriting any previous assignment. Use this when variant assignment comes from a server or external system rather than random local selection.

setExperiment("checkout-redesign", "variant-a")

clearExperiments()

Removes all persisted experiment assignments. The next call to getVariant() will perform a fresh random assignment.

Persistence

Experiment assignments are persisted locally so that a user always sees the same variant across app sessions:

Persistence survives app restarts and reinstalls (on iOS, Keychain data persists across installs). Assignments are only cleared when you explicitly call clearExperiments().

Events and Experiments

All events emitted by the SDK automatically include the current experiment assignments in the experiments field. No additional code is needed. This applies to:

• Regular events (via log() / event())
• Metric events (via startOperation() / recordMetric())
• Funnel events (via track())

Funnel Segmentation

Experiments integrate directly with funnel analytics. You can:

• Filter a funnel query to a specific variant: ?experiment=checkout-redesign:variant-b
• Group funnel results by variant: ?group_by=experiment:checkout-redesign

This lets you compare conversion rates between control and variant groups without building a separate analysis pipeline.

SDK Guides

For implementation details, see:

• Swift SDK experiments
• Node SDK experiments