Income Tax
Making a high-stakes filing journey explain itself

Tax filing is unfamiliar, consequential, and full of opaque language.

The brief started as a question I put to myself: ITR filing is one of the few high-stakes services most people in India touch once a year, often alone. Anyone filing alone needs to understand where they are, what matters next, and how to recover when something is unclear - first-time taxpayers and senior citizens most visibly, but a repeat filer reconciling capital gains against AIS is doing the same work with higher stakes and less sympathy. The design problem is confidence, not merely form completion.

• Audience Self-filing assessees across the range - people approaching a first return, older adults filing without help, and repeat filers handling capital gains, multiple income heads, and reconciliation against AIS
• Contribution Research, UX flows, prototypes, and user-testing materials
• Status Built and live as two interactive demos, currently collecting structured feedback - not a government service, and no real return is filed through it
• Constraint Protect participant privacy and avoid presenting portfolio content as official guidance

Before touching a screen, find out where confidence actually breaks down.

Rather than assume the interface was the problem, this drew on five strands of evidence, which are not all the same class and are kept separate here for that reason.

• Attitudinal interviews, professional Conducted during the official Income Tax website redesign (Infosys × Income Tax Department), across the filing range rather than one segment of it: salaried filers, freelancers, people with business income, people with capital gains, and first-time filers among them. Run as engagement research, and reported here only in the generalised form client work allows.
• Informal conversations, for this concept Five, run separately and on my own time: three first-time filers and two senior citizens filing without an accountant. No consent forms, no session recordings, no moderator script - candid conversations, and bounded accordingly.
• Reddit and X outreach A direct post to public tax communities; dozens of replies within hours, from first-time filers, seasoned traders, NRIs, and at least one practising CA.
• Secondary research on comparable systems Two national filing systems read in detail for mechanics India’s portal does not have - Estonia’s e-MTA and Sweden’s Skatteverket, set out further down this page. Published UX work on form comprehension and financial-task confidence shaped how these findings were framed, but none of it is cited here, so that part is background rather than evidence.
• Competitive scan What people named as already working elsewhere - secondhand from replies rather than a hands-on audit.

The distinction that matters: the interview strand is professional research from a real engagement, and everything else is self-directed work around this concept. Where this page draws a conclusion, it is the overlap between strands that carries it - not any one of them alone.

Secondary research - asking the community directly

Instead of only reading old threads, this started with a new outreach post across Reddit and X asking people who file their own returns what breaks down first: what's hardest to understand, how they figured out which form to use, and what they Googled mid-filing. Dozens of replies came in within hours, from first-time filers, seasoned traders, NRIs, and at least one practising CA.

• Form selection By far the most repeated question was some version of "which ITR form even applies to me?" - with no confident way to answer it inside the product.
• Redundant entry Salary, TDS, and broker data already sit in AIS and Form 26AS, but capital gains, dividends, and challan numbers still have to be typed in by hand.
• Terminology Section and schedule references (80C, Schedule FA, TR, FSI) appear throughout the form with no explanation attached to them.
• Recovery Validation is inconsistent - some mistakes are never flagged before submission, others surface as a notice months later.
• Silent defaulter risk The sharpest version of that: paying tax through a challan but forgetting to enter its details in the ITR draws no warning. The payment is real, but the return doesn't show it - and a "tax defaulter" notice can follow months later for tax that was already paid.
• Requests for a chatbot Several replies asked for one outright. Read next to the terminology complaints, it looks less like a request for AI and more like a request for help that doesn't require leaving the page.

Competitive analysis - the bar is already set elsewhere

Several replies didn't just name a problem - they named a competitor already solving it. That's a useful, checkable claim, so here's what each pattern looks like based on what people described:

This isn't a teardown of those products - it's secondhand, from replies, not a hands-on audit. The point is narrower: nothing here is a novel idea. The government portal is the one place people are required to use that doesn't yet do it.

International benchmarks - what a live-reconciliation flow could still add

A side conversation asked a simpler question: which country has the easiest filing process, and is it worth reading how they actually do it? Two systems came back with concrete, checkable detail - Estonia's e-MTA and Sweden's Skatteverket - and Skatteverket in particular had mechanics that went further than anything in the mockups above.

• Live preliminary calculation Skatteverket's dashboard recalculates tax owed or refunded in real time as any section is edited - the number is always visible, not just revealed at the end.
• A dedicated "Check" step Reviewing pre-filled data is its own named screen with its own state, not something implicit in scrolling past a field.
• Auto-generated schedule annexes Forms like K4 (securities sales) and K5/K6 (property) are built from data the tax authority already has - the filer can accept as-is, edit line-by-line, or ignore and build manually.

None of this is a new principle - the "reuse what the system already knows" theme above already named the problem. What Sweden adds is a specific mechanism for it, mapped onto the exact schedule structure India's own portal uses. It's a bounded borrow, not a claim that AIS is clean enough to auto-file anything: broker and RTA reporting to AIS has known lag and mismatch issues today, nowhere near the near-complete employer and bank reporting Skatteverket's pre-fill depends on. The live-reconciliation concept takes the first two mechanics - a running total that updates as you review, and a draft schedule you can accept, edit, or flag - while treating "is AIS reliable enough to draft from" as the open question it still is, not an assumption this note gets to make.

Informal conversations - covering who the replies missed

The Reddit and X replies skewed toward confident, self-directed filers - traders, freelancers, people fluent in tax vocabulary. Five informal conversations followed to cover who those replies mostly missed: three first-time filers and two senior citizens, filing without an accountant's help. These were candid conversations, not formal research sessions - quotes below are paraphrased and composited to protect participant identity. They sit apart from the engagement interviews above, which reached a wider span of assessees but cannot be reported here in this kind of detail.

Thematic analysis

Grouping the outreach replies and interview notes together surfaced four clusters that mattered more than any single feature request:

Turn a dense process into a calm guided path.

Each decision below responds directly to one of the four themes above, rather than to a generic best practice.

• Guided eligibility, not self-selection Ask plain questions about income and let the sequence of answers determine the ITR form - instead of asking someone to pick from a legal list on day one.
• Reuse what the system already knows Pre-fill and reconcile against AIS and Form 26AS wherever the data already exists, and only ask again when something doesn't match.
• Plain language at the point of confusion Explain a term where it appears, rather than sending people to search engines and losing their place.
• Calm, correctable recovery Name what's mismatched and what to do next - including flagging a genuinely-paid but unentered challan before submission, not after a defaulter notice arrives months later.

On the chatbot requests specifically: read on their own they look like a feature ask. Read next to everything else, they look like a symptom of the same problem the plain-language decision above already targets - help that lives outside the flow instead of inside it. The design bet here was to fix that first, rather than add a chat window on top of a process that still doesn't explain itself. Whether there's a smaller, genuinely open-ended need underneath that - edge cases too specific for an inline explanation - is a real question for the next round, not something to assume either way.

Sketch the sequence before the surface.

Early exploration stayed low-fidelity on purpose, to test whether the sequence of decisions made sense before any visual design existed.

Test the calmness of the journey, not just task completion.

Wireframes went back in front of four of the five original conversation partners in a second informal round. There was no formal script - watching for hesitation meant noticing where someone paused, re-read a line, or asked "wait, is this normal?" out loud, rather than timing task completion. The original outreach posts also stayed open through this period, so some of what changed here came from replies arriving there, not only from those conversations.

Not everything worked the first time. The earliest version of the Resolve screen stated the recommended form as fact, no way to question it. In testing, that read as more confident than it should - for something with real consequences, participants wanted a visible way to say "that's not right" even before they needed it. That's the one change that came from the design being wrong, not the research: the "that's not right" path only exists because the first version didn't have it.

Show the reasoning without exposing the service.

These mirror the six steps of the live filing demo, using its default "Salaried + capital gains" profile - anonymised, illustrative, and not tax guidance.

This is a portfolio case study, not an official tax service or source of filing advice. Reddit and X replies referenced here are paraphrased from public comments on the author's own outreach posts and are not attributed to individual usernames.

Confidence breaks down differently for a senior citizen than a first-timer.

The senior-citizen participants weren't struggling with the same thing the first-time filers were - jargon and a red error screen read as a personal failure to them in a way it didn't for a 24-year-old who could just Google the term. That difference shaped what actually shipped in the concept, and what's still just a flagged requirement.

• Plain language at the point of confusion Explaining a term where it appears, instead of sending someone to search and lose their place - built into the concept, not an afterthought (see Design choices above).
• Calm, specific error states P4's reaction to a red warning - "I saw a red warning and just stopped. For two days I thought I'd broken something." - is why mismatches are named and given a next step instead of shown as generic red errors.
• Keyboard access A skip-to-content link is present on every page of this concept for keyboard and screen-reader users, ahead of the main navigation.
• Not yet solved Larger text, slower pacing, and high-contrast modes for senior users were named as requirements early on but aren't built into this concept - they're a gap, not a decision, and belong in the next round alongside the language gap below.

What this concept cannot claim.

• Nothing here is a production result This is an independent concept, not a shipped change to the live service. No real return is filed through it, and nothing above should be read as a measured effect on real filers. The demos are now collecting structured feedback, but no finding on this page is drawn from that response set yet - when it is, it will be labelled as such.
• The strongest strand is the one I can say least about The engagement interviews were professional research across a real span of assessees, and the segments they reached are named above. What does not follow them here is a participant count or any verbatim material, so the strand carrying the most weight is the least checkable on this page - the detailed quotes earlier come from the informal conversations instead.
• The self-directed strands were informal, on purpose No consent forms, session recordings, or moderator script on the five conversations or the walkthroughs. That was the right trade for moving quickly outside an engagement, and it bounds what those strands carry: directional signals, not instrumented results.
• The walkthroughs were not an independent check Four of the five original conversation partners returned for the second round, so the people who described the problem are largely the people who assessed the fix. That keeps context and costs distance - they were primed to recognise their own complaints being answered.
• Five conversations, and a skewed wider channel Three first-time filers and two senior citizens, alongside an open outreach post on Reddit and X. The post reached people fluent in tax vocabulary far more readily than people approaching a return for the first time, which is why the five exist at all - they cover who the replies missed, but five is a floor, not a sample.
• Named accessibility requirements are not built Larger text, slower pacing, and high-contrast modes were identified early and are absent from the concept. They are a gap rather than a decision, so the accessibility section describes intent, not something tested with the users it is for.
• The screens are reconstructions Every interface shown is an illustrative concept screen made for this case study, mirroring the live filing demo rather than reproducing the service.
• One dependency stays unresolved Whether AIS data is reliable enough to draft a return from is treated here as an open question, not an answered one. The draft-and-confirm model assumes it; that assumption has not been tested.

Trust is part of the interface.

Impact

The first round of walkthroughs pointed one way: guided eligibility, in-place explanation, and calmer recovery states changed how people responded in session, not just what they completed. That is a directional read from a handful of sessions, not a validated result - what it earns is a next round, not a conclusion.

What is running now

The two items this write-up originally listed as next steps - widen past five participants, and put a testable version in front of people rather than treating the write-up as the final word - are no longer plans. Both concepts are built and public, and each carries the same structured feedback form.

• Two directions, not one Live reconciliation extends the Ask → Resolve → Recover sequence with a running tax calculation and a draft Schedule CG built from AIS data - aimed at repeat filers handling capital gains. The WhatsApp concept moves the same guided flow into a chat thread, for the first-time and senior-citizen audience whose barrier is an unfamiliar interface rather than the tax logic. Both stop short of filing: they collect and reconcile, then hand off.
• An instrument, not a comment box Each demo asks which of four personas the respondent is, how far they trust the draft on a five-point scale, which step was hardest, whether the live recalculation helped, and whether they would use it - plus free text and optional consent to follow up. Responses are separated by source, so the two concepts can be compared rather than pooled.
• What it is deliberately measuring Where confidence dips, not only where someone abandons the form. The hardest-step question maps onto the same five orientation moments this research started from, so incoming responses can be read against the original findings instead of alongside them.

What I'd still do next

• Language Three of five participants filed in English despite it not being their first language, and "explain the term" doesn't fully address that. The WhatsApp direction exists partly because that audience already types in its own language there - but the concept is still English-only, so the gap is named, not closed.
• Move from responses to a sample An open feedback form is self-selecting, and self-selection is not a sample. Reading it will mean treating volume as a signal of interest and the free text as the actual finding, until there is enough to segment by persona with a straight face.