Skip to content
Mar – May 2026 · Internship

Play Again, not Buy.
Would play again: 90%.

My impact
FiPet 1v1 Quiz Battle, home screen and battle loop
Role
Product design intern, lead on Quiz Battle
Team
1 PM, 2 designers, 10 engineers
Tools
Figma, Maze, React

10 of 10 interview participants called the app overwhelming. One said it felt like homework, which is the exact word a financial literacy app for kids cannot afford.

Overview

Launched with no usability testing and reviews that repeated the same complaints. I led design on a new 1v1 Quiz Battle, built the design system the team adopted, and ran the company’s first usability test.

Open the coded prototype
My impact

Players who bounced off the home screen now finish the game and ask to go again. The team kept the design system and the testing habit.

  1. 01FiPet shipped without a single test, and the reviews said so.
  2. 02So I ran the company’s first test: 65.9% misclicked before reaching the game.
  3. 03I rebuilt the flows, argued Play Again over Buy, and re-tested the same missions.
0%
would play again

9 of 10 full completions. This is the number that settled the Play Again versus Buy with coins argument.

0.0/5
fun rating

The first measurable engagement signal in the product’s history.

0
sessions recorded

The company had never run a usability test. This was the first, and the practice stayed after I left.

Evidence

The first task failed before anyone reached the game.

Task success was 100%, which looked fine until misclicks came in at 65.9%. The Start button sat below the fold, so people read the page as a buffer screen.

100%
task success
65.9%
misclick rate, task 1
228s
avg duration, task 1
90%
would play again

Naming a feature 1v1 is not enough if it plays solo.

Only 20% chose competing against a rival, the lowest of three options. They wanted to see what the rival answered, not a score at the end.

Who it is for

“It’s so confusing. I don’t even know what this app is for.”

Kids, ages 8 to 15 · First-time users · intercepted in public spaces
  • Something to tap within the first ten seconds
  • A clear signal that this is not homework
  • One consistent visual language, not five

Parents are the second audience and the decision-makers, but the failure happened before they mattered. Kids opened the app once, could not tell what to do, and never came back.

Solution 01

Play Again, not Buy with earned coins

The PM wanted the shop as the primary action. Conversion would never be higher. I argued the opposite, and the disagreement was about who the user actually is.

The argument

Kids this age are momentum-driven. The moment after winning is an emotional high, and a shop interrupts the loop the game just built.

Momentum, not transaction·8 to 15
What the position rested on
Verified
Flow Theory, Csikszentmihalyi 1990
Verified
children more flow-sensitive, PMC 2020
Directional
78% of parents, Commonwealth Bank 2023
What shipped

Round 1 later validated the call: 90% said they would play again.

Play Again as primary, shop reachable from home
Play Again · primaryHome · secondaryShop · from home
Why it landed

The PM had a defensible position, so mine had to be more defensible.

Specificity, not preference
three citations, not one opinion
Agreed and shippedRevisit post-launchwith real engagement data
Solution 02

A system four decisions wide, so the team could actually use it

One orange accent, Inter, SVG icons, an 8pt grid. Scoped small on purpose. The team adopted it for the broader redesign, which is the only measure that matters.

FiPet design system, one accent color, Inter, SVG line icons, 8pt grid
The fork

Same screen, two exits

The disagreement was one tap wide, and it decided whether the session continued or ended at the shop.

What the PM proposed
Match endsCoins awardedBuy with earned coinsShopSession ends
Conversion is highest here, and that is a real argument. It also spends the moment the game just built.
What shippedshipped
Match endsCoins awardedPlay AgainNext matchShop, from home
The shop stays one tap from home. What changes is which action gets the emotional peak.
Round 1 later measured the call: 90% said they would play again.
Exploration

Three home screens. Only one gave a reason to come back.

Two of them are things you watch. The third is something you do with someone else.

Daily lesson, social feed, and quiz battle home screens with the verdict on each
Only the third had a second player in it.
Fidelity

The argument was settled in greyscale

Button order is a hierarchy question, not a visual one. Deciding it in wireframe meant nobody in the room could argue that the orange was doing the persuading.

The win screen as a wireframe and as the final design, with the three decisions fixed before colour
90% said they would play again.
Mascot

Two mascots, two jobs

A pet cannot announce that its owner lost, so the referee had to be someone else.

The existing fox mascot beside the four Referee Owl variants
Drawn to read at 40px as well as full size.
Character system

One owl carries every system message

Four variants, held to that number so every state still reads from copy alone.

Referee Owl variants and the two screens where the owl appears
Held to that size on purpose.
Under the hood

Some questions a static prototype cannot answer

The obvious next step was another Figma iteration. I argued against it. Timer pressure, live scores, and answer-switching cannot be faked in a static prototype, and that is where the unknowns lived.

Verified
Only 20% chose competing against a rival as what they liked most
Round 1, multiple choice, lowest of three options
Rival answer moved to the reveal screen.
Verified
Participants asked to see what the rival answered
Round 1, open response
Added a waiting state so the match reads as turn-taking.
Directional
Requests for more game-like mechanics beyond the quiz
Round 1, open response, unprompted
Logged, not built. Widening the format before the core reads well would hide which change did the work.
Three changes shipped into the coded build. The fourth stayed on the list.
Open the coded prototype
Rebuild

The rebuild made the game deeper, not just testable

Once the battle loop ran on real state, gaps in the design became visible. Three additions came out of it: a power-up hand (Cut 2, Time+, Peek), a live timer ring, and a Mystery Box on the win screen. None existed in Figma. They only became designable once the prototype could play.

Power-up hand bottom sheet and Mystery Box reward screen from the React prototype
Cut 2, Time+, Peek. Agency under the timer, a way back when losing.
Read the full story

Why the first test mattered more than the feature

The app had shipped to the App Store with no usability testing and reviews that repeated the same complaints. I was hired to design one feature. The feature was the reason to run the first test, not the goal of it. Ten interviews came first, and all ten described the app as overwhelming. One said it felt like homework, which is the word a financial literacy app for children cannot afford.

How the evidence was graded

Two findings were Verified: only 20% chose competing against a rival as what they liked most, and participants asked to see what the rival answered. Those shipped as design changes. Requests for more game-like mechanics beyond the quiz were Directional, from a smaller number of sessions, and did not change the build. Separating the two kept the round from turning into a wish list.

What testing did not settle

Twenty-two sessions is enough to find a broken entry point and not enough to prove a retention effect. Task success read 100% while misclicks ran at 65.9%, which is the clearest evidence that the metric the company watched was not measuring the thing it believed. I left before the release cycle closed and cannot confirm what shipped.

Reflection
01

Introducing testing changed more than any screen

Once everyone had a shared evidence base, the PM stopped defending preferences with intuition and engineers stopped pushing back with their own assumptions. The arguments got smaller and the work got faster.

02

A design system is judged by adoption, not coverage

Four decisions, not forty, plus a short documentation set. Other designers used it for screens I never touched. A complete system nobody uses is worse than a smaller one that ships.