When to use AI Optimize

Traditional tests and manual personalizations give you direct control over how experiences are delivered.

AI Optimize doesn’t replace either one. It adds an adaptive layer that uses machine learning to learn from visitor behavior and available signals, then changes which variations visitors see over time.

[IMAGE: Testing and Personalization shown as two paths, with AI Optimize layered across both.]

Use AI Optimize when you:

• Have several ideas or combinations to explore
• Want delivery to adapt as performance changes
• Want to reduce exposure to weaker-performing variations
• Have lower traffic and may struggle to reach significance with a traditional test
• Don’t need to stop and declare one permanent winner

The optimization type still matters: you’re still testing or personalizing. AI Optimize changes how variations are delivered.

What you’ll configure

AI Optimize still starts with familiar building blocks:

• Base: the original experience
• Variations: the ideas you want AI to explore
• Target goal: the outcome AI should optimize toward
• Audience, optional: limits who is eligible for the optimization
• AI Optimize: controls adaptive delivery

[IMAGE: An AI-optimized setup showing base, several variations, target goal, audience, and AI Optimize enabled.]

The biggest difference is what happens after launch: instead of keeping delivery fixed, AI Optimize continues adjusting based on what it learns.

AI Optimize with testing

A traditional test uses the traffic split you configure to compare variations and determine which performs best overall.

AI Optimize has a different purpose: continuously adapt variation delivery to maximize conversions as visitor behavior changes.

Imagine you’re exploring three homepage headlines:

• A benefit-focused headline
• A feature-focused headline
• A proof-focused headline

At first, AI needs to explore the different options. As it gathers data, it can adjust delivery based on which variations are more likely to drive the target goal for different visitors.

That means weaker-performing variations can receive less exposure while stronger-performing variations receive more. That allocation can continue changing as AI learns.

Traditional test

Which variation performs best overall?

AI-optimized test

Which variation is most likely to work for this visitor right now?

Key takeaway

The goal isn’t to arrive at one permanent winner. It’s to keep adapting variation delivery over time to improve performance.

AI Optimize with personalization

Manual personalization depends on rules you define:

Enterprise audience → Enterprise variation

With AI Optimize, you don’t need to manually define every audience-to-variation match. Instead, AI learns which variations are more likely to work for different visitors based on the signals available to it.

[IMAGE: Different eligible visitors being routed toward different variations by AI Optimize.]

The goal is still personalization: different visitors may respond best to different experiences. What changes is who decides which variation they receive.

The audience still determines who is eligible for the optimization. AI Optimize determines which variation an eligible visitor sees.

Build and launch with AI Optimize

The setup should feel familiar because you’ve already built a traditional test and a manual personalization. You’ll:

1. Create your variations. Give AI meaningful options to explore.
2. Choose the target goal. AI Optimize uses this outcome to decide what to optimize toward.
3. Set an audience, if needed. Limit the optimization to a particular group of visitors.
4. Enable AI Optimize.
5. QA each variation. AI can only optimize what you give it, so every variation still needs to work as intended.
6. Launch.

Now that you know what changes when AI Optimize takes over variation delivery, use the walkthrough below to practice building an AI-optimized experience from start to finish.

[ARCADE: Create an AI-optimized experience, add multiple variations, choose a target goal, set an audience if needed, enable AI Optimize, and launch.]

Give AI useful options

AI Optimize works best when the variations give it meaningful differences to learn from. If every variation says essentially the same thing, there isn’t much for the system to explore.

Give it ideas that reflect real hypotheses, such as:

• Feature-focused vs. benefit-focused messaging
• Short vs. detailed copy
• Different value propositions
• Different imagery
• Different calls to action

The same principle from earlier still applies: start from a real opportunity and give each variation a reason to exist.

Best practice:

Give AI something worth learning from

AI Optimize can adapt which experiences visitors see, but it can’t make weak or nearly identical ideas meaningful. Start with variations grounded in real hypotheses about what could influence visitor behavior.

Ready to continue?

You now know how AI Optimize changes variation delivery for both tests and personalizations, what you still configure yourself, and how to give AI useful options to explore.

Next, you’ll learn how to interpret the results of traditional tests, manual personalizations, and AI-optimized experiences — because each one asks you to look at performance a little differently.