
Solution and impact
- An AI Coach with a real person one tap away, so quick questions get quick answers and big money decisions get a specialist.
- Navigation that speaks the customers' language, interactive color-coded charts, and key information first in a clear visual hierarchy.
- One design system for a consistent, accessible experience on every platform, with simpler layouts, logically grouped content and high-contrast design.
- Mobile screens with standard navigation, streamlined interfaces and shorter multi-step flows.
About project and settings
- A regulated product. Projections carry SEC / FINRA disclaimers, so the AI could not give advice that sounded like a recommendation without a person in reach.
- 6 audiences, from beginner investors who feel overwhelmed to older savers who want an advisor to decide for them, plus a Spanish-language experience.
- Phased scope. P0 was a full rebuild of navigation and IA, with partial redesigns of the dashboard, the Coach, accounts, retirement and finance.
- WCAG 2.2 AA on desktop, tablet and mobile.
Let people try the AI before it takes over support
The original brief called for a full AI coach that would replace the help center and pass people to a human when needed. Our first usability test, 6 people in hour-long sessions, changed that plan. People told us plainly that they wanted to know when they were talking to an AI and when to a person. They were happy to ask an AI about calculations, projections and definitions, but wanted a person involved in any big money decision from the start. They also asked "Why should I trust this data?"
So the Coach arrives in 2 steps:
- Coach Lite, the MVP. A carousel of short, actionable cards with a chat feel, a typing indicator and one message at a time, placed next to Chat with an advisor. People get used to where the Coach lives without having to type a money question to an AI.
- The next phase: a chat that hands over. The Digital Coach answers basic questions and points to the Empower team on anything bigger; one tap invites a person into the same conversation. Answers carry view sources, a rating and a way to change the level of detail.
This way, AI Coach, human advisor and product support are labeled as 3 separate channels, because people asked for that line explicitly.
This strategy was proven by a follow-up usability test, where the Coach wasn't in the prototype, and participants asked for personalized, chat-like advice on their own. That told us the first step had worked.
How to handle the uncertainty of AI
The Coach answers what it can and suggests the Empower team for anything bigger. One tap invites a person into the same conversation.

Retirement is 30 years away,
so make the distance visible
Gamification was in the first brief, and we removed it in April to ship a clean MVP. In May, usability test III put goal widgets in front of 12 participants: first in context on the dashboard, then each goal type side by side.
- Goals gave a clear snapshot, and color made status stand out in a gray UI. People wanted the goals that need attention first.
- Dollars beat percentages 85% of the time, and almost everyone wanted a goal marker for long-term goals like the emergency fund and the retirement budget.
- A pace can turn against people. Some asked to celebrate milestones instead: "Markers don't acknowledge when life happens."
- People wanted help setting and prioritizing their goals.
So I came back to gamification with a narrower question: how do you show a near, reachable step on a 20–30-year horizon?
I explored 5 metaphors, a dinosaur, a city, a tree, an alpaca and a hot-air balloon, all on one data model: estimated income, income goal, savings, employer contributions, Social Security, the gap. The metaphor is a skin; the real numbers stay visible in every version. Explored, not shipped, parked for Phase 2.
One system in 3 libraries, maintained by other people
A shared Foundation and 2 platform libraries, Web Desktop and Mobile App, so a token change lands once and reaches both.
- 335 variables, 105 primitive colors and 163 semantic, with Desktop, Tablet, Mobile and Wide Desktop modes; 34 text styles, including a number scale for financial figures; 256 icons.
- 2,300+ variants across 242 component sets. Data visualization is one of the largest areas, because that's where testing found the most pain.
- The Coach is a component in both libraries, so AI surfaces follow the same rules as everything else.
- Governance: a changelog, documentation in zeroheight for Empower's design and engineering teams, and a status on every page: Ready, In progress, On hold or Archive.
Data on a phone. Holdings is a wide investment table. On mobile it comes in 2 patterns: a collapsible table, where each holding opens into its details, and a scrollable table, where the holding column stays put and the numbers scroll sideways. Filters, date ranges and accounts open in bottom sheets.
Tools
Figma
zeroheight
Standards
WCAG 2.2 AA
Not a straight line: every finding changed the plan, and it still shipped
3 usability rounds, 2 brief rewrites, 4 navigation versions and a summer of post-handoff work. Each finding was written down and changed what came next, so the winding path stays readable as a set of documented iterations.
| When | What happened | What it changed |
|---|---|---|
| Q4 2024 | Audit, stakeholder interviews, competitor research, card sorting | The product as a chore; the Coach reframed as a feed |
| Jan 2025 | Usability test I, 6 participants | Coach off the first screen; support ≠ AI; specific copy |
| Mar 2025 | Usability test II: comprehension, findability | People asked for the Coach unprompted → Coach Lite |
| Apr 2025 | Brief rewritten twice | Gamification removed from the MVP |
| May 2025 | Usability test III: goal widgets, 12 participants | Dollars over percentages, goal markers kept, milestones over pace → goal metaphors explored |
| May–Jun 2025 | Accounts, mobile, handoff | Ready-for-dev package |
| Jul–Sep 2025 | Post-handoff collaboration: design-system work, guidance for the teams, accessibility, refinements | The system kept up with what engineering built |
What I'd do earlier
- Recruit beyond confident users. Our first test leaned towards people comfortable with digital tools; the audiences Empower worried about most were the least represented.
- Agree on an engagement metric before testing engagement ideas. Goal metaphors came back on a qualitative signal; a baseline would have let us compare them on more than taste.
- Introduce AI as a presence first. Coach Lite was more than a cut-down chat: it is how people start to trust the Coach.





















