Subscription Marketplace — From three static screens to a YC-backed company
Product
A credit-based marketplace for flexible access to premium tools — Grammarly, Adobe, Otter.ai, MasterClass, Crunchbase, and 150+ others — without long-term subscriptions. ClassPass for digital tools.
My role. Founding (and for most of Keye's life, only) designer. I designed every surface — web, mobile web, Chrome extension — wrote PRDs, ran research, and hired the two designers who replaced me.
Timeline. 2 years (2021–2022): beta to credit system to marketplace to extension to seed funding. After the founding team dispersed, the company pivoted into YC W2024 on the traction we built.
• User growth: 0 → 20K+ MAUs in under 12 months
• Engagement: +45% within 3 months
• Trial-to-paid conversion: +32% post monetization
• Catalog & partnerships: 150+ products; 35+ direct partnerships incl. Adobe, Grammarly, Otter.ai
• Partner subscriptions: +44% YoY
• Extension adoption: 60% of users within 3 months
• Login friction: −63% via auto-login
$1.5M pre-seed · $30K in grants and competition wins · Featured in Forbes, BulletPitch, UPenn Venture Lab · Accepted to YC W2024
Users created fake emails for free trials, shared passwords in friend groups, even paid for a VPN to unlock cheaper regional pricing. Premium tools like Grammarly ($180/year) and Crunchbase ($600) were priced for committed subscribers, not curious explorers.
"I keep creating new email accounts to get free trials. It's the only way to try things without committing."
ClassPass proved flexible multi-provider access worked; Apple One bundled someone else's services. Neither filled the gap between free trial and full subscription — that gap was Keye's opportunity.
My first task was the purchase flow. Founders had baked in mandatory consent and a feedback survey at checkout — completion dropped 30% there. I moved consent to signup and tested the survey in three placements: email (80% ignored), in-checkout (confused buyers), and a post-access popup, which worked because users had just used the product. Drop-off fell 20% in one sprint.
The hardest call was removing product descriptions after testing showed users scanning, not reading. I led with imagery and a side-by-side price comparison, pre-selecting Keye — some users then bought full subscriptions through us, a behavior we hadn't designed for but kept.
The most important design decision in Keye's history — not for how it looked, but for what it revealed.
Unlimited credits (MVP) removed all intentionality. Users activated HBO, Hulu, and Tubi the same day, exhausting access before people who actually wanted those products could get in.
A 200-credit cap changed behavior immediately — users now had a stake, and price became a discovery mechanic. When credits ran low, they explored cheaper products they'd never have clicked.
Monetization followed the evidence. We charged $20/month for 200 credits once the mental model was set — conversion rose 32% over beta. We reinforced it with a visible credit balance, onboarding pop-ups, and a How Keye Works page; support requests dropped 74%.
Users were copy-pasting credentials from Keye into partner sites. Partner integrations were supposed to fix it, but each required engineering investment partners wouldn't prioritize — then our technical co-founder left, taking backend capacity with him.
Watching Honey apply coupons at checkout, I saw it: live where the user already is. I built a Figma prototype that weekend — visit a partner site, extension detects it, auto-fills credentials. We shipped in ten days.
MVP was deliberately narrow — purchased products only, auto-fill only. Results: 60% installed within 3 months, login friction down 63%, repeat sessions up 41%. V2 added in-extension purchasing, earned by that data.
I came from architecture, where everything is a parametric component — change the object, every instance updates. I treated the product card the same way, evolving it from a long-description carousel to a compact, partner-neutral frame. Engineering kept asking: can this be a state of the existing component, not a new one? That question kept us reducing instead of adding.
The mature card supported eight states — browseable, active, disabled, Buy Again, Featured, Coming Soon, Leaving Soon — all derived from behavior, none speculative.
I rebuilt my handoff: annotated docs, precisely named assets, pre-answered edge cases, Saturday-morning overlap sessions. That discipline became the onboarding foundation for the designer I hired — when she took over my features, users couldn't tell where my work ended and hers began.
By year three, the founding team dispersed. The company pivoted into YC W2024 on financial analytics — the marketplace didn't survive, but the research, partnerships, and funding that made YC possible came from what we built.
Keye compressed the feedback loop between decision and consequence from months to days. It taught me to treat every decision as a hypothesis — the ones I confirmed without testing were the ones that broke in production.