Project Overview
TL;DR: As part of the projekt202 team engaged on the TaxAct account, this was a ground-up redesign of TaxAct's core filing experience, kicked off by a brand refresh, but shaped almost entirely by what direct research with filers turned up. Instead of treating the brand refresh as a coat of paint, a full research and synthesis process fed directly into a new enterprise design system, and both were used to redesign the Federal Tax Return flow, including a live, human tax assistant that proactively stepped in when filers hesitated, something no competitor in the study offered.
Role: Senior UX Designer, projekt202, engaged on the TaxAct account. Focus: research and synthesis, an enterprise design system, and the Federal Tax Return redesign.
The Challenge & Context
TaxAct had recently redesigned its brand, but the interface running underneath it hadn't moved. The Federal Tax Return flow, the one experience every filer goes through, had been built up over years with no shared design system, so headings, buttons, and form patterns looked and behaved differently depending on which part of the product you were in.
TaxAct needed its core product to catch up to its new brand, and needed a real design system underneath it so the redesign wouldn't start drifting the way the old interface already had. But repainting what existed wasn't the goal, so before changing a single screen, the team went and found out what was actually wrong.
The Process & Decision-Making
Watching people file, in their own homes
Rather than run a survey, the team went to filers directly: 11 participants across Dallas and Seattle, spanning their 20s to their 60s and a wide income range, split evenly by gender. Each session was in-context: people were watched filing for real, on their own machines, with their own documents.
To understand how TaxAct actually compared, observation didn't stop at TaxAct. The same filers were watched using TurboTax and H&R Block too, so every pain point logged had a competitive baseline attached to it.
Turning 1,200+ notes into eleven themes
Every observation, quote, and behavior became its own data point, over 1,200 in total, run through affinity diagramming: raw notes grouped into "I" statements, then "we" statements, then up into named themes the whole team could work from.
It's a slow climb from a wall of sticky notes to something a product team can act on, but skipping it is how research turns into a report nobody reads twice.
One theme changed the direction of the whole project
Most of the eleven themes pointed at things the team expected: filers wanted convenience, clarity, plain language. Theme eight was different. Filers kept saying, unprompted, that when they got stuck, what they actually wanted was a person, not another Help article.
"It'd be nice to chat with someone live. It says look for it on the Help section but I couldn't find it." That was P3.
"It looks like all these things are assistance, but I really need a person." That was P5.
None of the three products studied offered that. This was before AI assistants were a realistic option, so the theme became the seed of a very specific bet later in the project: a live tax assistant, staffed by an actual human, that could step in the moment a filer hesitated.
Three filers, three very different relationships with confidence
Personas were built straight from the eleven participants, rated across five spectra: financial understanding, confidence, initiative, how often they sought help, and tech skill. Three distinct patterns emerged, from a filer who avoids taxes out of anxiety to one who wants the software to predict her needs before she asks.
Mapping where trust actually broke down
A simple filer's full path (Prepare & Login, Basic Info, Federal, Review & File) was traced, pinning every pain point to the exact step it happened on. Login and password-reset friction. W2 imports that failed silently. Manual entry filers resented. A marketing upsell that landed right when someone was anxious about what they owed.
Plotting emotion alongside each step made the pattern hard to miss: confidence rose and fell in a predictable rhythm, and every dip lined up with a moment the product offered no help at all.
Turning findings into a standard the team could design against
The last step of the research phase was making sure it didn't stay a report. The eleven themes were distilled into ten design principles, language the whole team, design and engineering both, could hold every later decision against.
"Help Users Not Feel Helpless," drawn directly from theme eight, became the principle that justified building a live, human tax assistant instead of another layer of Help copy.
The Solution & Final Design
Auditing the product, then building the system
Ten principles gave the team a standard. Designing against it meant first knowing exactly what already existed.
A full interface inventory logged every heading, button, form element, navigation pattern, modal, card, and icon already live in the product, years of accumulated patterns with no single owner. From there, a component prioritization workshop with product and engineering stakeholders decided what belonged in a v1 design system versus what could wait.
The system was planned as roughly 26 weeks of sprints, starting with foundations (color, type, icons, buttons) and building up through forms, navigation, loading states, and page templates. It was designed in Sketch, version-controlled in Abstract, handed off through Zeplin, and built out in React.js, with a governance model so components kept moving from request to release long after the initial sprints ended.
Even the color palette was tested, not eyeballed: every primary color checked against black-and-white tints for accessibility before it was allowed into the library.
Rebuilding the flow, and putting a real person in it
With the design system in place, the Federal Tax Return flow was rebuilt around it, targeting the exact friction points research had surfaced: login and password-reset failures, W2 imports that failed silently, manual entry, and marketing upsells that landed at the worst possible moment. The redesigned flow was validated with a subset of real filers before it shipped.
The biggest departure from the rest of the category was the tax assistant: a live person, not a bot, built into the flow, that proactively stepped in when a filer paused or hesitated, rather than waiting for someone to go looking for help. This was before AI assistants were a realistic option; the differentiator was a real person, not automation.
Impact, Outcomes & Learnings
Rebuilding the flow on direct research instead of a visual refresh meant TaxAct addressed the specific moments where filers actually lost trust, like login friction, failed W2 imports, and mistimed upsells, rather than guessing at what "modern" should look like. The design system replaced screen-by-screen improvisation with a single, versioned library spanning roughly 26 weeks of planned sprints, foundations through page templates, that design and engineering could both build from. And the proactive tax-assistant model gave TaxAct a form of live human help that none of the three products in the research, including TaxAct's own prior version, actually offered.
Reflection: This project sharpens the case against treating a brand-refresh brief as license to just repaint the product: the research kept surfacing a structural gap, people wanting a real person when they got stuck, that no amount of visual polish would have fixed. Looking at competitors early was a check against that instinct too. Seeing that TurboTax and H&R Block had converged on the exact same generic, solo filing pattern made it clear that solving the problem meant diverging from the category, not matching it with a nicer coat of paint.