Beating an 8-Year Undefeated Paywall : With Claude Code and Superwall
Data-informed decision making
DEFINE → PLAN → COLLABORATE → LAUNCH
Partenerd w/ PM & Engineering
Role
Product Designer Product Manager
Team
Product Manager
Front-end Engineer
Back-end Engineer
Timeline
6 mo June 2025
Original paywall
Let me share some backgound:
This is our original paywall, and for 8 years, it beat every "better" version we threw at it.
Cleaner layouts, sharper copy, more polished design. None of it worked, across 60 million users and 17 languages.
That bugged me when I joined. I figured if I couldn't beat it, I could at least find out why it's working.
So instead of trying to out-design it, I ran 11+ AI-designed experiments across 6 phases — with zero engineering dependency — until one finally won
Result: a 36% lift in ARPU, and a paywall that's still undefeated — just no longer unbeaten by us.
Problem of the
Original paywall:
Based on Mixpanel,
Less than 1% of users subscribed during onboarding. Not because the product was weak, but because the paywall wasn't working.
And every attempt to fix it required a full developer sprint — by the time a test shipped, the hypothesis was already stale.
We needed a faster way to learn.
The real problem wasn't the paywall. It was our inability to iterate on it at the speed the problem deserved.
2. Problem Statement
We need to find why design alternatives can beat the paywall that has 1% subscription rate- with limited time and resources.
3-1. Research w/Claude
AI platform to speed up the procedure
Claude code prompts
Prompt 1: Recommend AI platforms to speed up software/Product simple paywall UI engineering iteration time.
Our team (CTO, dev head and I) had a meeting
To choose the platform that would be best for us, had a rounds of meeting with Revenuecat and Superwall team, and choose superwall based on their customer service and price
3-1. Research lead us to chose Superwall for our testing platform.
We don't needed engineering work once the SDK and backend logic is synced.
3-2. Research w/Claude
Competitive analysis
Claude code prompts
Prompt 2 : Competitive analysis
Run a competitive analysis on paywalls: Surface a pattern on winning paywalls with screenshot themselves and compare their winning factor with ours
Competitive analysis surfaced a clear pattern: competitors led with emotional urgency and kept paywalls tightly focused. Ours had social proof and a feature list — but it was long, dense, and asked users to process too much before making any decision.
Mixpanel told a sharper story: 99% of users didn't subscribe during onboarding.
But 16–23% of those users did convert within 24 hours — meaning the intent was real. It just wasn't being captured in the moment. And 75% of users who reached the CTA abandoned without tapping.
KEY INSIGHT
We didn't have a demand problem. We had a timing and friction problem. Users wanted to subscribe — the paywall just wasn't making it easy or urgent enough to act right now.
4. Ideation & Strategy
Before fixing the paywall, we fixed how we could learn about it.
The research pointed clearly toward experimentation. But our engineering bottleneck meant we couldn't test fast enough to make experimentation meaningful. Every test took a full sprint. By the time results came in, we'd already moved on.
With the bottleneck removed, I designed a six-phase roadmap that moved from surface-level variables to structural ones — ordered deliberately by cost of change. Each phase isolated a single element so every result pointed somewhere clear.
5. Design & Iteration
Five phases of narrowing. One phase that broke through. 🎉
Phases 1 through 5 weren't failures — they were eliminations. Each test ruled out a class of assumption and pushed us toward the one that actually mattered.
Color and imagery moved metrics in small increments. Restructuring the paywall into a multi-step flow improved completion rates, but conversion still plateaued. Copy and social proof tightened the experience. Every test contributed — but none broke through.
By Phase 5, a pattern was becoming clear: we'd been optimizing what we showed users. We hadn't yet tested how much control we gave them over the decision.
KEY INSIGHT
Every phase that didn't move conversion was teaching us something: the answer wasn't in the surface. It was in the structure of choice itself.
Design your own trial
Phase 6 tested a different assumption entirely.
Instead of presenting the "best" option, we gave users a choice: a free 7-day trial, or a $5.99 30-day paid trial. We called it Build Your Own Trial (BYOT). The hypothesis was that autonomy — the act of choosing — would create a different psychological relationship with the subscription.
Users who chose their trial converted more — and churned less.
THE UNLOCK 🔓
When users chose their own trial, they weren't just selecting a price point. They were making a decision — and people commit differently to decisions they made themselves. Autonomy reduced cognitive friction and increased felt ownership. That's what we'd been missing.
6. Result
Every metric moved. The system kept learning.🎉
ARPU increased 36%, driven by users self-selecting into the plan that fit them. Paywall conversion improved from 2.5 % to 22.5 %. Cancellation rate remained same after introducing the "design your own trial" flow.
The system kept learning. Each phase sharpened the next question.
7. Takeaways
Fix the learning system before fixing the problem.
Superwall didn't just speed up testing — it changed what kind of designer I could be on this team. Removing the engineering bottleneck meant every week became a data point. The real output of this project wasn't a better paywall. It was a team that could now learn faster than the competition.
Order experiments by cost of change, not by excitement.
Moving from color → layout → copy → social proof → pricing → autonomy wasn't arbitrary. It was deliberate sequencing: cheap, fast tests first to eliminate surface explanations, high-commitment structural tests last. This is how you protect a team's time while still being thorough.
Autonomy converts better than optimization.
The BYOT result reframed how I think about conversion. We'd spent five phases making our recommendation clearer and more compelling. The breakthrough came when we stopped recommending and started offering a choice. Users who design their own decision commit to it differently — and churn less as a result.