ZenBudget
Concept project · Product designer and researcher · 2025–2026
A budgeting app for people paid partly in cash tips. I took it from an open bootcamp brief through research, testing, a change of direction, and a working coded prototype.
01 / Overview
Designing a budget for income that arrives in cash
Traditional budgeting assumes income arrives as a paycheck. It doesn't account for cash tips.
I researched how people manage money, designed and tested a first version with six people, and changed direction when testing exposed a problem with cash. I then designed and built a working prototype for hospitality workers.
My role: research, interaction and visual design, a design system, and the coded prototype. Status: the cash-tip flow is built but not yet usability tested - that's the next step.
02 / Starting point
A bootcamp brief that became a product
ZenBudget started in my CareerFoundry UX bootcamp as an open brief: design a personal-finance app from scratch, end to end: ideation, survey, interviews, user journeys, affinity mapping, a design system and usability testing.
Competitive analysis of Bank of America, Cash App and Venmo showed AI guidance and subscription tracking together were uncommon, so my first versions explored that. What ZenBudget is today came later, from what testing showed me.
03 / Research
What 20 people told me, and 4 explained
A survey of 20 people, interviews with 4, and a card sort with 5.
65%
didn't track a budget at all
60%
didn't track subscriptions, but wanted to
5.8/10
average comfort with AI, only if it explains itself
4 of 4
interviewees had dropped a tool that asked more than it gave them
-
Survey: 20 people, mostly multiple choice, under 10 minutes — habits, attitudes to AI, current tools, a short profile. 70% were 25–34; 60% worked full-time.
Interviews: 4 people, 30–40 minutes, semi-structured — tracking money, subscriptions, AI and trust, what they'd change.
Card sort: 5 people. Subscriptions, reminders and alerts clustered as one job; "goals" and "shared budgeting" split people.
Limits: a small convenience sample from my own network, so I treated findings as direction, not proof. The survey never asked about income type or cash.
04 / Design and design system
Turning findings into structure
The card sort showed people treat subscriptions, reminders, and alerts as one job, so they became one place. Because effort decided everything, I designed for the fewest steps per task.
The visual system was built around calm: soft blues, red kept for alerts only (red around money causes stress), and a tone closer to a coach than a bank.
05 / Usability testing
Six people, one prototype, what changed
Moderated think-aloud sessions with six people: three on mobile, three on desktop. Everyone completed the tasks; the struggles were about meaning. Numbers needed explaining, people expected a confirmation after any change, and AI should be something you switch on, not something that pops up.
So I made AI an explicit opt-in with its own tab, and made every add or cancel confirm what happened. One theme came up with five of the six people: cash.
-
In person, think-aloud, 20–30 minutes, on the high-fidelity Figma prototype.
Tasks: onboarding, the AI assistant, subscriptions, budgeting and adding a transaction.
Themes: onboarding felt long to some; people missed that screens scrolled; numbers needed context; red/green with arrows confused; AI should be opt-in; confirmations after changes; Home and Budget overlapped; cash handling.
What I didn't do: collect timed metrics, or synthesise the desktop sessions as fully as the mobile ones.
06 / The pivot
Why I changed direction
In testing, cash broke the numbers. One participant explained that with $100 in cash, the app would only ever show his bank balance, so any cash he spent made the total wrong. Once the numbers didn't add up, people stopped trusting the rest.
I'd already designed a way to log cash spending, but that was only half the picture: for many people, cash is also income. From about eight years in hospitality, I knew tips can be a large part of a worker's real earnings and rarely make it into any app.
No participant asked for tip logging. Refocusing on tipped workers was my judgment, built on a real finding, and the next round of testing is designed to check it.
-
About 4 million US workers were in tipped occupations in 2023 — about 2.5% of all employment — with median weekly pay of $538, against $1,000 for other workers (Yale Budget Lab, 2024).
Waiters and waitresses held 2,284,400 jobs and bartenders 762,600 in 2025 (US Bureau of Labor Statistics).
A deliberately small audience: I chose depth over breadth, for the group where the gap between what an app shows and what they really have is widest.
Not known yet: how much of those tips are paid in cash.
07 / Design decisions
Key decisions, and what I chose not to do
Shifts, not dates
Workers remember earnings by shift, like "Saturday dinner". The shift is pre-set from the time of day, and always visible, because testing showed people distrust things an app decides silently.
Log what you took home
Tips are often pooled before a worker leaves. I log only the cash they actually took home and keep pooling out of the app, it happens before the money reaches them, and would add steps at the end of a shift.
Say what's in your wallet
Cash on hand is calculated from what you log. When it drifts, you count your wallet and ZenBudget works out the difference, people know what they're holding, not what the app is missing.
Colour means good or bad for you
"$20 up, but it's green" confused testers. Now the arrow shows direction, colour shows meaning, red appears only when something needs attention, and colour never works alone.
Ask how people get paid
A reviewer pointed out that asking every new user about cash tips doesn't fit people who don't earn them. One question after sign-up now sets things up: paycheck-only users skip the tip features, and people paid mostly in tips start with cash. Nothing is pre-selected, so a default can't nudge the answer. It's untested, and part of my next round.
Make the main action stand out
The Log cash tips card was fading into the background, so I kept its warm amber and gave it a defined edge and a solid icon. Category tags became dots: as outlined pills they looked like buttons, and testers had mistaken status labels for controls.
What I left out: a proposed "spendable" figure. A new number on the main screen needs a clear definition and a reason to be there; that one had neither.
08 / The prototype
A working prototype, not just screens
I built ZenBudget as a coded prototype from my Figma designs, with AI as a build partner, so the flows can be used, not just viewed. The walkthrough follows the core job: logging tonight's tips, and names the design decision behind each step.
Try it yourself
One task: log $85 of tips for your shift. Look for:
the shift is already set from the time of day - tap to change it
a numpad instead of a keyboard
the comparison with your previous shift of the same kind after you save
Everything is simulated; nothing you type is sent anywhere.
09 / How I used AI
AI as a collaborator, not the designer
Analyse
AI checked my research, Figma and early screens against each other and flagged where claims went further than the evidence, including claims from earlier versions of this case study.
Challenge
Every suggestion came as a proposal I accepted, changed or rejected, and every decision is on record. When AI's cleanup made the screens consistent but generic, I rejected it and kept the original visual character.
Build
The Figma file and a coded prototype kept in step, built with AI and deployed so people can use it.
10 / Reflection and next steps
What I'd do differently, and what I'm testing next
I'd screen for income type from the start. My survey never asked about cash or tips, so when I changed direction there was no survey data behind it.
I'd measure, not only observe. I planned task times and success rates but ran the sessions as qualitative only.
Next is a second round of testing with hospitality workers: can someone log a shift correctly without help, how long does it take, and do shifts match how they remember their earnings?