A 1% difference nobody can see
This case study is password protected.
- My role
- Product Designer. Owned the visual system, every card layout, the edge-case states, the illustration set, the surfaces around the deck and the teaser film.
- Problem
- August set the bar for delight with a pixel-art dark deck. September had to be a new visual world, not a reskin — and the PRD forbids a language model from touching any numeral.
- What changed
- A single hairline construction language that works three ways: as the decoration under every card, as the chart vocabulary, and as thirteen illustrations.
- Outcome
- A complete default deck, drawn states for the four branching cards, five entry states, the home banner and sheet, and a teaser film.
Need access? Email ishitabhorework@gmail.com
september unwrapped · 1%club · monthly money review · oct 2026
Sixty seconds of your money
Thirteen cards that tell you what you are worth, what you spent, what your money did and what needs your attention — in a product where a language model is never allowed to write a number.
MY ROLE
Product Designer. Owned the visual system, every card layout, the edge-case states, the illustration set, the surfaces around the deck and the teaser film.
TEAM
1 designer (me) · PM · data · tech · legal review
TIMELINE
Explorations 24 Sep · system locked 26 Sep · deck complete 1 Oct 2026
STATUS
Default deck complete. Four PRD data gaps and legal sign-off still open.
Cover · the thirteen-card deck
1508 × 900
THE 30-SECOND VERSION
PROBLEM
August set the bar for delight with a pixel-art dark deck. September had to be a new visual world, not a reskin — and the PRD forbids a language model from touching any numeral.
MY APPROACH
Settle the data layouts on plain black first, find the theme second. Eight directions tried and killed in a week before anything was approved.
WHAT CHANGED
A single hairline construction language that works three ways: as the decoration under every card, as the chart vocabulary, and as thirteen illustrations.
OUTCOME
A complete default deck, drawn states for the four branching cards, five entry states, the home banner and sheet, and a teaser film.
MY LEARNING
A theme is not a skin. It only earns its place if it survives being used as a chart.
01
The second month is harder than the first
13
CARDS IN THE DECK
60s
TARGET READ TIME
4
CHAPTERS
1,998
NON-MEMBERS MEASURED FOR CARD 12
August shipped the first Monthly Money Review: pixel art, a glass mark, a dark premium feel. People liked it, which made September harder rather than easier. The first one of anything defines what delight means. The second has to match that without repeating it.
The brief I set myself was narrow on purpose: keep the room August built — dark, crafted, deliberate — and change everything inside it.
THE BAR
August’s deck. Dark, premium, pixel-led. Anything lighter or flatter read as a downgrade.
THE TRAP
Reusing August’s pixels. It would have been fast and it would have read as a repeat, not a sequel.
THE RULE
A theme’s meaning comes from the brand, the user’s own money and a fresh style each month — never from a calendar or a festival.
August’s deck beside September’s, for the delta
1508 × 800
02
Eight directions in a week
Everything below was built, looked at and killed between 24 and 26 September. I am showing all of it because the approved direction only makes sense as the last thing standing — and because three of the rejections produced rules the deck still runs on.
24 SEP
Clean editorial, full-bleed colour, monochrome, light product screens
REJECTED
Delight lives in the dark, crafted world August set up. Light and editorial read as a different product.
24 SEP
Calendar and festival themes
REJECTED
A theme has to come from the brand and the user’s own money. Hanging it on the month’s festivals makes it decoration with no argument.
24 SEP
Stepped gradient bands, Wrapped-like
REJECTED AS A COPY
Take a reference’s principles, never its geometry. This one was recognisable as somebody else’s deck.
24 SEP
Brand glass-block forms and pixel variants
REJECTED
Pixels were August’s theme. Reusing them is a repeat even when the forms are new.
25 SEP
Pixel-light art from the moodboard, used as the charts themselves
REJECTED
Moodboard art is decoration. The moment it carries data it has to be precise, and it was not.
25 SEP
Themed card sets — glass on clouds, halftone, checkpoint boxes
REJECTED
Fake glass reads as a button. Panels and legends slow the read on a card meant to take four seconds.
25 SEP
Cards first on plain black, with construction-grid deco
APPROVED
Settle the data layout before the theme. Once the cards worked unstyled, the theme had something to sit on.
26 SEP
Grid and dot backgrounds, waterfall charts
REJECTED
Grids behind data read as clutter, and waterfalls confuse on a story card. Decoration moved to the bottom edge and stayed there.
The rejected directions, as frames
1508 × 1100
03
The system that survived
What came out the other side is small enough to hold in your head, which is the point when thirteen cards have to feel like one object.
TYPE
Season Mix for hero figures and titles, Manrope for everything else. A maximum of three sizes on any card.
COLOUR
One chapter colour per part — blue, orange, purple, yellow — as a soft top-down gradient on near-black.
DECORATION
A hairline construction layer at the bottom edge of every card. Twenty animatable steps, dark and light.
GREEN
Reserved for AI and nothing else: the mark, the word Unwrapped, every AI CFO entry point.
The cover and the wrap close the loop: September, Unwrapped opens on the AI mark rising as a sun over a horizon, and That’s a wrap sets on the same horizon thirteen cards later.
The system — type scale, chapter colours, construction deco
1508 × 800
Cover and wrap on the shared horizon
1508 × 700
04
One number, one sentence, one visual
The PRD set six rules before any design existed. They are the most useful constraints I have been handed, because each one removes a decision rather than adding a requirement.
ONE OF EACH
One number, one sentence, one visual per card. Everything else is cut.
SAY WHY
If a number moved, the card has to say what moved it.
SHOW THE WHOLE
A proportion has to show what it is proportional to.
EARNED SIZE
Big type only for numbers that earned it. Nothing is large because the layout had room.
FLAT IS A RESULT
A quiet month is an outcome, not an empty state. Every card has a steady version.
NO AI NUMERALS
A language model never writes a numeral. It writes the sentence around numbers that were computed.
That last one shaped the whole deck. Every card carries a two or three sentence AI summary sized as the hero — so the thing a user reads first is written by a model, while every figure inside it comes from the pipeline. XIRR, alpha, benchmark, drawdown and allocation are banned words.
One card dissected against the six rules
1508 × 900
05
The card that needed eight states
Card 7 compares what the index did to what the user did. It is the only card where both numbers can go either way independently, which means it has eight honest outcomes and no fallback.
THE GEOMETRY
Bars sit on a zero baseline and go below it for falls. Names pin to the zero line, values sit at the bar end.
THE SCALE
Floored at 2% so a quiet month still reads as a shape rather than two flat lines.
WHAT BROKE FIRST
Labels floating under the bars looked like a rendering bug. Pinning them to zero fixed it.
The eight states cover the index falling while the user falls less, the index rising while the user falls, a month where neither moved, and the case where there is no index data at all. Each one is drawn, because a card that only works when the market is interesting is not finished.
Card 7 · eight states
1508 × 1000
Card 11 · the rank chip with no rank, three variants
1508 × 600
06
Thirteen findings, thirteen drawings
Card 12 is the only card that asks the user to do something. The PRD defines sixteen themes, measured across 1,998 non-members in September; three were deferred, which leaves thirteen live themes that between them give 97.3% of users at least three findings.
Each theme needed its own drawing. Not an icon — a small diagram that carries the finding, in a language consistent enough that thirteen of them look like one set.
69.5%
Top 3 sectors above 50%
One arc covers more than half, split in three
60.9%
No debt and no gold
One slot filled, two empty
50.9%
3-year win rate below 50%
Six periods, two wins, halfway marked
42.8%
Fund rated Poor or Below Average
A five-band dial, needle in the lowest band
39.5%
Regular plan held
Same start, the regular line stops short; dashes are what commission took
36.8%
Mid and small cap above 50%
One big circle against a cluster of small ones
31.8%
Large cap under 40%
The big circle is only partly there
29.4%
Up-capture below 100
The market rises further; dashes are the rise missed
24.3%
Drawdown deeper than 30%
Falls far below its old high, drop measured
24.2%
Expense ratio above 1%
A ring with a bite taken out
23.7%
High-overlap fund pair
Two rings, the shared part lit
23.1%
More than 10 funds
Ten funds on a ring, two drifting outside it
16.2%
Top 5 stocks above 25%
A dashed quarter slice, and a lit slice in five that runs past it
Share of 1,998 non-members each theme fires for, September 2026. Themes 4, 5 and 16 are deferred.
Three exclusion rules keep the card honest: two themes that measure the same thing never render together, and up-capture always carries down-capture beside it, so the card cannot flatter by omission.
The thirteen drawings as one set
1508 × 1000
07
Finding the drawing language
Three rounds, two of them wrong, and the wrong ones were wrong for different reasons.
ROUND 01
Icons — Phosphor-style duotone and a pixel set
REJECTED
The first two passes were called ugly outright. The third used researched metaphors and was better, but an icon names a thing; it cannot carry a finding.
ROUND 02
Pixel illustrations on a 16 × 16 grid
REJECTED
Too big and too intricate at card size, and pixels still belonged to August.
ROUND 03
Construction lines, drawn in the bottom deco’s own values
APPROVED
The decoration already on every card turned out to be a drawing language. Nothing new had to be introduced.
The resulting rules are tight enough that a new theme can be drawn without asking anyone: 1px white lines at 16 to 42% opacity, small nodes, dashes for anything missing, and exactly one orange accent marking the thing that is wrong.
Three rounds — icons, pixels, construction lines
1508 × 900
Themes 12 and 14, five ways each
1508 × 700
08
Finding the layout
Getting three findings onto one card without it reading as a list took five rejected layouts and then a frame she had already drawn.
THREE EQUAL ROWS
Tested across five theme combinations. Readable, but nothing told the user which finding mattered most.
FIVE PLACEMENTS
Leading icon, airy, hero, ghost, index. All rejected — none of them were clean enough at card size.
ONE LEAD, TWO LATER
Start here, then Also worth a look. The hierarchy does the work the equal rows could not.
Three versions of that layout differ only in where the small drawings sit: beside each row, absent, or on the left. V3 went into the final deck.
The five rejected placements
1508 × 800
V1, V2, V3 of the one-lead layout
1508 × 800
09
States are semantic, not a template
Five screens stand between a user and the deck: not connected, stale data, generating, already up to date, and not available yet. My first pass gave them one shared look, and it was rejected — correctly.
These are five different meanings. Making them look alike makes the screen say nothing.
NOT CONNECTED
The full cover, undimmed. Nothing is wrong yet — there is just no data. Connect accounts.
STALE
The same screen with a warning row. The deck exists, it is just out of date.
GENERATING
A progress bar and no action. The only honest thing to do is say come back.
ALREADY CURRENT
You are ahead of the curve. A compliment, not an error.
NOT YET
The sun in red, half above the horizon so it is still legible as the same world.
THE RULE
Every semantic state keeps its own look. Generic states lose their meaning.
Five entry and error states
1508 × 900
10
Beyond the deck
The deck is the deliverable people see, but a story that lives in the app needs a way in, a way to react to it, and a reason to open it at all.
HOME BANNER
The wrap screen’s horizon at 312 × 180 — the AI mark as the sun, three rings, eight chapter-coloured pixels on an eight-second orbit, and a pulsing live pixel.
BOTTOM SHEET
First built with the same orbit, then deliberately made static. The orbit was being overused, so the sheet was redrawn in the card style instead.
FEEDBACK
Not every card has an AI line, so thumbs up and down needed a placement that scales across the deck without colliding with the bottom decoration.
The teaser film went through the same correction. A phone slideshow of screens was rejected; the brief became motion graphics built from lifted design elements. Two concepts were made — a pixel that collects the four chapter colours, and numbers that split along a green seam to reveal why they moved. The first was picked and refined twice.
GRIDS
Appear only where the story needs them, never as wallpaper.
CAMERA
Never rotates. An unmotivated tilt reads as a template.
EXPLANATION
Never drawn as diagram lines. If it needs a diagram, the idea is not clear enough yet.
Banner, bottom sheet, teaser stills
1508 × 900
11
The rules this established
Decisions that outlived the month they were made in. Anyone extending the MMR inherits these.
COLOUR
Green means AI and nothing else
Keeps the AI voice recognisable across thirteen cards and the surfaces around them.
COLOUR
Orange marks the issue in a drawing; everything else stays grey
One accent per illustration is what makes it readable at a glance.
DECORATION
Dashed means empty or missing, never a button
Dashes already carry that meaning everywhere else in the product.
DECORATION
No grids or dots behind data; decoration stays at the card’s bottom edge
Anything behind content reads as clutter at card size.
CHARTS
Falls go below a zero baseline, names pin to the line, values sit at the bar end
Labels floating under bars looked like a bug on Card 7.
CHARTS
No waterfalls on a story card
Tested on Card 2 and called confusing. A story card gets four seconds.
CHARTS
A comparison must read as market, plus extra — never two unrelated bars
Card 13’s first attempts showed both returns and conveyed no relationship.
STATES
A missing value lives inside its component, never as a sentence
A rank chip with no rank keeps the card’s structure stable; Not on our board does not.
COPY
PRD and designer copy is immutable — problems are flagged, not reworded
Copy is reviewed and owned upstream, and silent rewrites break that review.
MOTION
Motion carries meaning; no unmotivated camera moves
Straight from teaser feedback.
PROCESS
Revisions go beside the original, never over it
Versions get compared side by side, and the older one is often the one that ships.
12
What is still open
The deck is drawn. Some of it cannot ship until questions outside design are answered, and I would rather list them than imply the work is finished.
FOUR DATA GAPS
Start-of-month holdings, cash balance, plan type and expense ratio per holding are not available yet. Five of the eleven actionable Card 12 themes depend on them.
THE SUMMARY LINE
Nobody has decided whether the AI sentence is generated at render or a template filled with computed values. It changes what the card can promise.
LEGAL
Fund ranks on Card 11, findings sitting beside a paid call to action on Card 12, and the basket comparison on Card 13 all need sign-off.
THE 2.7%
About one user in forty gets fewer than three findings, and the PRD has no title written for that case.
AS OF WHEN
No financial figure on the deck carries an as-of date yet. On a monthly product that is a trust problem, not a formatting one.
CARD 1
Its summary line predates the insight rules and repeats what Cards 2 and 8 already say.
The open question I care most about is the second sentence. Every theme has two sentences written for it and the card shows only the first. Where the second one goes — on the card, behind a tap, or nowhere — decides whether this deck explains a finding or just announces one.
NEXT CASE STUDY
Reach without depth
Turning mutual fund metrics into something people can understand and act on, without crossing into financial advice.