Experiments

The Swift SDK provides lightweight client-side A/B testing. Variant assignments are persisted in Keychain and automatically included on all events, allowing you to segment metric and funnel analytics by experiment group.

Assigning a Variant

Call Owl.getVariant() with an experiment name and a list of options. On the first call for a given experiment, a random variant is selected and persisted. All subsequent calls return the stored variant -- the options parameter is ignored after the initial assignment.

let variant = Owl.getVariant("checkout-redesign", options: ["control", "variant-a", "variant-b"])

switch variant {
case "variant-a":
    showNewCheckoutFlow()
case "variant-b":
    showMinimalCheckoutFlow()
default:
    showOriginalCheckoutFlow()
}

Method Signature

@discardableResult
public static func getVariant(_ name: String, options: [String]) -> String

The return value is marked @discardableResult, so you can call it without using the result if you only need the assignment to be recorded on events.

Force-Setting a Variant

Use Owl.setExperiment() to override the assignment for a specific experiment. This is useful when variant assignment comes from a server-side feature flag system rather than random client-side selection.

Owl.setExperiment("checkout-redesign", variant: "variant-a")

Clearing Experiments

Remove all experiment assignments from memory and Keychain:

Owl.clearExperiments()

This is useful during development and testing, or when you need to reset experiment state for a user.

Persistence

Experiment assignments are stored in the device Keychain under the service com.owlmetry.experiments. This means assignments survive app restarts and even app reinstalls, providing a consistent experience for each device.

Auto-Inclusion on Events

All events emitted by the SDK automatically include an experiments field containing the current set of assignments. You do not need to pass experiment data manually -- it is attached to every info(), debug(), warn(), error(), track(), startOperation(), recordMetric(), and all automatic lifecycle events.

This enables segmenting any data by variant without additional instrumentation.

Example: A/B Testing a Feature

struct PaywallView: View {
    let variant = Owl.getVariant("paywall-design", options: ["control", "minimal", "social-proof"])

    var body: some View {
        Group {
            switch variant {
            case "minimal":
                MinimalPaywall()
            case "social-proof":
                SocialProofPaywall()
            default:
                OriginalPaywall()
            }
        }
        .owlScreen("Paywall")
    }
}

With this setup:

1. Each user is randomly assigned to a variant on first view. The assignment is persisted in Keychain.
2. Every event from this user (screen views, funnel steps, metrics) includes "paywall-design": "minimal" (or whichever variant was assigned) in its experiments field.
3. You can query funnel conversion or metric performance grouped by the paywall-design experiment to compare variants.

Combining with Funnels

Experiment assignments are especially useful with funnels. Track funnel steps as normal and query results with experiment grouping to measure which variant produces higher conversion rates.

let variant = Owl.getVariant("onboarding-flow", options: ["control", "streamlined"])

// All track() calls automatically carry the experiment assignment
Owl.track("welcome-screen")
Owl.track("create-account")
Owl.track("complete-profile")