Shift E-Commerce A/B Checkout Weights in Real Time (Kiponos Java SDK)

A/B tests are supposed to be scientific. Reality: variant B is crushing it at 10 AM, variant C is hurting conversion at 2 PM, and marketing wants to reweight the split before the weekend rush — without a deploy window.

Kiponos.io lets your Java checkout service read live experiment weights from memory on every session assignment.

Why not hard-code weights?

int bucket = hash(sessionId) % 100;
if (bucket < 50) return Variant.A;
if (bucket < 80) return Variant.B;
return Variant.C;

Those cutoffs (50, 80) usually live in config files. Changing 50→60→70 means:

Rebuild + redeploy during peak sales
Or per-request DB lookups (latency + load)

Live weight model

experiments/
  checkout_q2/
    variant_a_weight: 40
    variant_b_weight: 45
    variant_c_weight: 15
    enabled: true

@Service
public class CheckoutExperiment {
    private final Kiponos kiponos = Kiponos.createForCurrentTeam();

    public Variant assign(String sessionId) {
        var exp = kiponos.path("experiments", "checkout_q2");
        if (!exp.getBool("enabled")) return Variant.CONTROL;

        int a = exp.getInt("variant_a_weight");
        int b = exp.getInt("variant_b_weight");
        int c = exp.getInt("variant_c_weight");
        int roll = Math.floorMod(sessionId.hashCode(), a + b + c);

        if (roll < a) return Variant.A;
        if (roll < a + b) return Variant.B;
        return Variant.C;
    }
}

Product manager slides weights in Kiponos dashboard → new sessions immediately use updated distribution. Existing sessions can stick with assignment cookie — your choice.

Use cases

Scenario	Live tweak
Winner emerging	Shift 70% traffic to best variant
Bug in variant C	Set `variant_c_weight: 0` instantly
Flash sale	Disable experiment, route 100% to optimized flow
Geo test	Separate weight folders per region