Sabrina Fonseca / UX
Taking IRS Direct File from a 12-state pilot to a 25-state launch
Period
2023 - 2025
My role
Researcher and User Research Team Lead, leading a team of 5 researchers
Agency
Coforma
Methods
Observation studies, moderated and unmoderated usability testing, customer satisfaction surveys, Spanish content and usability testing, content testing, testing with people who use assistive technology, feedback interviews, journey mapping, ecosystem mapping
Roles
Researcher and User Research Team Lead, leading a team of 5 researchers
Recognition
Listed in the FedScoop 50 under Innovation of the Year, 2024
Challenge
The IRS had about 8 months to build Direct File, a free way to file federal taxes directly with the IRS, for a 12-state pilot that couldn't afford to fail. A year later, it opened in 25 states. When I joined, designers were running their own usability tests and were running out of time to finish design work.
What I did
I made the case for a dedicated research team and led it through the 2024 pilot and 2025 launch, developing trauma-informed research protocols, setting up research operations, running the research roadmap and leading observation and content testing studies.
Outcome
The research team tested all designed features well ahead of launch in 2024 and 2025. One finding from a study I ran prioritized a fix that cut one of the most common causes of rejected returns by about a third. The 41 studies and 227 sessions with taxpayers from my team helped provide an experience that taxpayers scored +80 NPS in 2025 and 94% rated it excellent or above average.
In 2023, the IRS was building Direct File, a free way to file federal taxes directly with the IRS, as a pilot for the 2024 filing season. The team had about 8 months to build what needed 18, and the pilot couldn't afford to fail. Tax data is the most protected data in the federal government, so every research method needed privacy approval before we could use it.
Designers were testing their own work
I joined in June 2023 as a designer and researcher, and my first task was designing and testing the handoff from Direct File to state filing tools. To help alleviate designers’ design and testing workload, I made a case to move from to doing research exclusively across many topics, alongside another researcher.
I set up a research roadmap anyone on Direct File could send requests to, and took up running research weekly meetings and pursuing research requests to alleviate the UX director's plate. I also pushed new templates through the privacy approvals so studies could use people's own tax documents. Soon I was asked to lead the research team, and the IRS asked Coforma for more researchers. The team became official: 5 researchers from Coforma and the IRS, and me.
We finished every planned study before the deadline, with time left for lower-priority work.
Consent and trauma-informed practices
Many people are intimidated by the IRS, and some of our studies asked them to share their tax documents during the session when IRS designers might be present. So every study asked people, in the screener, who they were comfortable having in the room, and confirmed that in writing before any session with sensitive tax documents. Sessions with real tax forms weren't recorded, and no session had more than 3 research staff. This was one example of the trauma-informed research protocols I workshopped and created with the research team.
Signing a return in the 2024 pilot. Before, people typed last year's AGI or a self-select PIN.
After, they could bring in last year's AGI from their IRS account.
A PIN that might work
Before launch, we could only test parts of Direct File, with made-up data. Once the pilot was live, we started end-to-end observation: people filing their own returns on the live product with their own tax documents.
In March 2024, I was moderating one of those sessions when the participant reached the signing step. To sign, people entered last year's adjusted gross income (AGI) or a self-select PIN. The participant typed a PIN that might have been right and expected Direct File to check it. It didn't. That check could only happen after the return was submitted, when the IRS accepted or rejected it. Most of the people we watched preferred the AGI, because they could look it up.
Following the rejection past the screen
During the pilot, 18.2% of submitted returns were rejected over those two fields. The team could see the rejections in the data, but not why, and the cost didn't stay on the signing screen. A rejected return sent people back to fix something they had no way to check, and AGI and PIN questions made up 30% of customer service volume after rejections.
A designer had watched the session with me, and right after it I reported the finding at a pilot evaluation check-in with design, data analytics and product. The team was already working on a way to pull last year's AGI from the person's IRS account, so the answer would come from the IRS's own records instead of someone's memory. Once they saw what was behind the rejections, they prioritized it. It shipped within two weeks, and looking up the AGI became the recommended way to sign. Rejections for those two fields fell to 12.1%, about a third below where they started (earlier design and content changes had brought them to 16.4%). In 2025, after work with legal counsel, Direct File stopped asking for either.
The second season
From the 2024 filing season through April 2025, the team ran about 41 studies and 227 sessions with taxpayers under my lead, and I ran or co-ran 13 of them. We tested in English and Spanish, and with people who use assistive technology or have cognitive disabilities.
I also turned the studies into a detailed journey map of filing taxes that starts before Direct File and ends well after it, from choosing how to file through state taxes, the refund and a possible audit. With it came an ecosystem map of what shapes each person's filing, from the person and the people around them to institutions and government.
These materials also helped reshape design principles followed by the designers and writers on the team.
Direct File ran for two filing seasons. The IRS didn't offer it in 2026. Its code was released on GitHub open source.
Results
The reception to Direct File was overwhelmingly positive:
437,334 accepted tax returns in 2024 and 2025
+80 Net Promoter Score in 2025
94% survey respondents ranked their experience as Excellent/Above Average in 2025
User testing prioritized a fix that cut one of the most common causes of rejected returns by about 1/3
86% respondents said their experience with Direct File increased their trust in the IRS in 2024
IRS Direct File was listed in the Fedscoop 50 2024 under Innovation of the Year
“It was amazing, easiest taxes I've ever prepared! Really impressed that this was put together by the IRS.”
— Direct File User in 2024
“La información y pasos a seguir fueron muy claros de entender y muy fácil de seguir.”
“The information and steps to follow were very clear to understand and very easy to follow”
— Direct File User in 2024
"My experience with entire system made me trust that my taxes were done correctly."
— Direct File User in 2024
“[Direct File is] a small step toward a world where people don’t need commercial tax software. But it’s a meaningful step, and the case for taking things slow is surprisingly compelling.”
— Dylan Matthews, Vox News