A familiar problem, a new way of working.
Retainers was, on paper, a familiar type of problem for me: complex, flexible, and dependent on careful information architecture, the same shape as other projects in this portfolio. What made it different was how I worked. This was the first project where AI tools were a core part of my process, not just polish at the end.
Asking for the ideal, not auditing the broken version.
The existing retainer experience wasn't good, so instead of auditing it directly, I asked people to set it aside and describe their ideal experience, a deliberate way to avoid anchoring research on a broken baseline. Alongside that, I dug into why and when people actually use retainers, and the edge cases involved, since retainers flex a lot and the design needed to reflect that instead of forcing rigid structure.
I also saw room to push the information architecture further: building it so one retainer could span multiple client projects instead of being scoped to just one.
Prototypes people could actually click through.
This was my first real use of AI tools in the design process: brainstorming directions, comparing ideas against competitor research, pulling the most useful thinking out of each source instead of leaning on one tool for everything. I took it further than most projects by using AI to generate working HTML prototypes, which let me test real interactions with people instead of static mockups, and get feedback on flow and feel, not just layout.
Leaning into simplicity on purpose.
Retainers get complicated fast underneath, so the design leaned hard into simplicity: a form, an index, and a dashboard for each retainer. Every piece was built from existing design system components, no new one-off patterns.
The AI-generated HTML prototypes got tested with real users, and the feedback was positive. Currently awaiting development.
Honestly, the design itself wasn't hard, and it wouldn't have taken me much longer without AI. What made this work was already having the judgment to tell which of AI's suggestions were good and which weren't, without that, I'd have accepted its suggestions too readily and ended up with something disjointed.
A few habits worth stealing.
- Keep prototypes low-fidelity on purpose. A prototype that looks too polished convinces people it's already good, a real problem when the existing product genuinely needs work. A rough prototype keeps the conversation honest.
- Use emojis or letters instead of real numbers in placeholder data. For number-driven users like accountants, real numbers invite them to check the math instead of judging the design. Swap in symbols and attention stays on the flow.
- Use low-fidelity mode on myself, too. When brainstorming, I have AI work through my own raw brain dump instead of polishing it immediately, asking clarifying questions and pushing back instead of just agreeing something "looks fine." That friction is usually where the better idea shows up.