Reward-Driven Flags

How Versia's reward-driven flags learn what works

How Reward-Driven Flags Work

Reward-driven flags are regular feature flags with a superpower: they learn.

The Loop

  1. User evaluates a flag. Versia picks a variant.
  2. User interacts. Your app sends a reward if the user converts.
  3. Versia learns. Future evaluations favor variants that get more rewards.
  4. Repeat. Every interaction makes the flag smarter.

What You Need to Do

Define your variants

Create a flag with two or more options. A CTA with different text, a pricing page with different layouts, an onboarding flow with different steps.

Pick a reward

What does success look like? A click, a signup, a purchase? When it happens, send a reward event.

That’s it

Versia handles everything else. Traffic starts equal and automatically shifts toward whatever works best.

Rewards

A reward is a simple signal: “this variant worked for this user.”

// Pass the same context used during evaluation so Versia can personalize
const ctx = { targetingKey: 'user-123', plan: 'pro', device: 'mobile' };
client.track('conversion', ctx, { flagKey: 'banner-cta', value: 1.0 });
  • 1 = success
  • No event = neutral (handled automatically)
  • You can also send a dollar amount for revenue optimization

What If Different Users Like Different Things?

Versia handles that too. Pass attributes in your evaluation context, like device type, plan, country, or anything else:

const variant = await client.getStringValue('banner-cta', 'sign-up-free', {
  targetingKey: 'user-123',
  plan: 'pro',
  device: 'mobile',
  country: 'de',
});

Versia learns that mobile users prefer one variant while desktop users prefer another, or that free-plan users convert better with a different CTA than enterprise users. No rules to configure. It figures it out from the data.

When to Use Reward-Driven Flags

  • CTA optimization
  • Pricing page variants
  • Onboarding flows
  • Landing page experiments
  • Any decision with a measurable outcome

When to Use Regular Flags Instead

  • Feature rollouts (no measurable outcome)
  • Kill switches
  • Simple on/off toggles
← Flag EvaluationAnalytics →