Led 3-person design team
Zero-to-one
Web 3
The challenge
Cronos was losing prediction market activity to competitors. Users had to leave the ecosystem entirely to place trades pulling volume and engagement outside the platform. The business objective was to prove that a native protocol could retain users within Cronos and generate real trading activity.
We were given 6 weeks. No existing user base, no IA to inherit, and a product category that most users had never encountered before.
Where I pushed back
The PM proposed lifting the IA directly from a competitor to compress the timeline. I argued we couldn't commit to a structure before understanding who Delphi's users were and what would make them stay.
I pushed for a two-week research extension and presented a clear discovery plan to the team. The original 6-week timeline moved to 8. That decision shaped the product's navigation, scope, and feature prioritisation from the ground up.
My role
Led two product designers across the full project and through post-MVP.
Owned: user research, IA, primary flows, and final design approval. Led: design direction, twice-weekly alignment, critique, and work distribution: core flows with me, execution and edge cases with the team.
The job wasn't just designing Delphi. It was making sure three designers shipped one coherent product.
How I ran the team
Across the project I ran two recurring design syncs per week, one to review progress, one to make decisions on open items and unblock the next sprint. During discovery we reviewed research findings together and formed hypotheses as a team rather than me handing down conclusions.
With the wider cross-functional team I set up a weekly design show and tell from the start, so Product and Engineering were never surprised by where design was heading. In the final weeks, as build ramped up, I moved this to twice a week — Mondays and Thursdays — and added a parallel engineering show and tell where developers walked through the build step by step.
That meant design could catch implementation drift early, before it compounded. Issues found in a show and tell take an hour to fix. Issues found in QA take a week.
Research approach
Competitive
analysis
User
interviews
Workshop
& IA
Prototyping
& testing
I looked at 12 existing prediction markets Polymarket, Kalshi, Manifold, and others comparing features, design, and real usage data.
Accessibility gap
Half weren't beginner-friendly. We owned accessibility.
Market creation gap
Only 3 of 12 let users create markets. We made it core.
Engagement, not reach
Polymarket: $607M from 437k users. Limitless: $38M from 600. Depth beats reach.
User interviews
I ran 7 interviews with a mix of newcomers and experienced users. Instead of asking what they wanted, I had each person pick their favourite protocol Polymarket, Manifold, whatever they actually used and walk me through how they used it, what they looked for, and where they got stuck.
Three patterns kept emerging.
Key insights
Placing a bet shouldn't require a trader's vocabulary
People join when the market looks active and trustworthy
Engagement shouldn't end the moment a bet is placed
User persona
Design principles
Why it matters: These insights shaped the MVP scope and structure, prioritising clarity, confidence, and early participation over feature depth.
Principle #1
Clarity before power
Make outcomes and next steps clear before introducing advanced controls.
Principle #2
Reveal complexity gradually
Only show advanced detail when users need it.
Principle #3
Keep jobs separate
Treat Discover, Portfolio, and Rewards as distinct tasks
Principle #4
Design for confidence
Use clear states, confirmations, and recovery paths to reduce anxiety.
IA workshop
(how I drove alignment)
Research time was tight, but alignment on product structure had to happen before a single screen was built. I organised a hands-on card sorting workshop with Product and Engineering (10 participants across both disciplines) to map user tasks, group them by intent, and agree the MVP navigation in the room.
The team walked out with shared ownership of the IA not just my recommendation of it.
Core design decision.
The most important design decision wasn't visual. Behind every market sits a conditional token framework, users lock real money as collateral, receive tokens representing possible outcomes, and those tokens settle at $1 if correct or $0 if wrong. None of that mechanism should reach the user.
I worked with engineering to land on a single metaphor: the price of YES is the probability of YES. A market at 38% means YES costs 38 cents and pays $1 if it wins. One concept, applied consistently across every screen.
With architecture and principles locked, I moved into wireframing and early validation to pressure-test the riskiest parts of the flow before visual design making sure users could understand the system, complete key actions, and recover from mistakes without extra explanation.
User test results
6 participants tested Delphi across 6 tasks.
Overall
sentiment
"I really like this UI a lot, very similar to stock trading app, not super crypto, it's very simple." — Participant 5
I established core UI foundations and component patterns, evolving them into a lightweight design system that supported consistency and rapid iteration across the MVP and future features.
Components were built to WCAG 2.1 Level AA as a baseline, contrast ratios, labelled inputs, and descriptive error states were defined at system level, not left to individual screens.
Design decisions
I worked with the Creative team to define the visual direction. The goal was an interface that felt credible and calm in a category prone to feeling chaotic. Financial-grade, not crypto-loud.
Design metrics
Delphi ran for six months before new leadership decided to absorb prediction markets into the core Cronos app, a strategic decision to scale the feature at platform level. These metrics reflect that full window.
54%
First-trade activation
up from 42% in month two. We introduced a welcome incentive for users who traded within their first 3 days.
43%
second-trade rate
driven by limited-time 2×/3× reward multiplier events, giving users a specific reason to return.
82%
of users who started a trade completed it
After changing the Max button from wallet balance to market maximum, the manual-adjustment drop-off disappeared from screen recordings.
4 min 20 sec
Median time to first trade - Improved alongside trade completion after the Max CTA change. Fewer steps at the moment a user was committing real money.
Learnings
Looking back, Rewards was the most nuanced UX problem in the product. The main flows were the right priority as they were the foundation everything else sat on, but in focusing on getting those right, Rewards didn't get the same depth of thinking it probably needed. Given the chance to do it again, I'd have carved out dedicated time for it earlier in the process rather than treating it as something we'd refine post-launch.
The Discord and screen recording plan was the right call given the constraints. But it shouldn't have been the plan. It should have been the fallback.
Post-launch
iterations
We couldn't run another round of usability testing before launch due to timeline constraints. Instead, we built our feedback loop into the post-launch phase monitoring Discord and reviewing screen recordings to identify where users were struggling and why.
Redefine 'Max'
Screen recordings showed users tapping Max, then immediately reducing the amount, a sign Max wasn't behaving as expected. The button was set to wallet balance, but markets have liquidity caps. A user with $1,000 in a $200-capacity market would hit Max, get constrained, then manually drag down to find the ceiling. We changed Max to reflect the maximum available in that market. Same button, different intent, one fewer step at the moment a user is committing real money.
Trade completion rate improved to 82%. Median time to first trade dropped to 4 min 20 sec.
Turning rewards into competition
Users earned points after each trade, tied to a future protocol token we couldn't announce, the legal status and utility were still being finalised. Discord showed frustration: people didn't understand what the points were worth or why they should care. We repositioned them as a leaderboard. Instead of a future promise, users had live rank, something to compete for now. Discord sentiment shifted.
From onboarding to first trade
42% of new users connected a wallet but never placed a trade. To bridge that gap we introduced a time-limited welcome incentive: a 1.2× reward booster, redeemable within the first 3 days. A specific reason to act now rather than later. First-trade activation moved from 42% to 54%.
Giving users a reason to return
To drive repeat behaviour we introduced limited-time multiplier events, selected weeks where every trade earned 2×, 3×, or 4× leaderboard points. A countdown timer created urgency. A specific reason to come back this week, not someday. 43% of first-time traders placed a second trade.
