An end-to-end redesign of one of HMRC's most-used digital services — replacing opaque, form-heavy flows with a guided, conversational experience that works for everyone.
"I just don't know what they're asking me. The words don't make sense to me — I feel like I'm going to get it wrong and get in trouble."
Every year, millions of UK citizens are required to complete a self-assessment tax return. For many, it's one of the most stressful digital interactions they'll have — an experience built around HMRC's internal taxonomy rather than the mental models of the people actually using it.
The service suffered from dense, jargon-heavy language, a non-linear structure that forced users to hold too much in their heads at once, and no meaningful guidance when people got stuck. The result was a high drop-off rate, a flood of avoidable support contacts, and significant accessibility failures for users with lower digital literacy or cognitive load constraints.
Discovery research identified three core failure modes:
Instead of presenting the full return to every user, we introduced a short qualifying flow that established their situation — employment type, income sources, changes from last year. The service then surfaced only the sections relevant to them, reducing perceived complexity by over 60% for the majority of users.
Every question was rewritten in collaboration with content designers and plain English specialists. Legal and tax definitions were retained as expandable contextual help, not as the primary instruction. We tested language variations in 6 of 8 research rounds, iterating until 90%+ of participants could answer each question without assistance.
We restructured the experience as a single-question-per-page pattern (aligned with GOV.UK Design System principles), with clear progress signposting, contextual hints, and the ability to save and return at any point. Error states were redesigned to be specific, actionable, and non-judgmental.
Every component was built and tested against WCAG 2.1 AA standards, with additional research sessions conducted with screen reader users, users with dyslexia, and users on low-bandwidth connections. Accessibility was a design constraint from day one — not a post-delivery audit.
More users completing their return end-to-end without abandoning — a direct consequence of reduced cognitive load and clearer language.
Halving avoidable calls and web chat requests by giving users the context and confidence to proceed independently.
First version of the service to pass an external accessibility audit with zero critical or serious failures across all tested scenarios.