People Do Not Think in Departments

The Hidden Barrier

Administrative categories are useful inside an institution. They establish ownership, legislation, records, and reporting. But the categories that help an organisation operate are not automatically useful choices for a citizen. A person may know that a licence was damaged, an address changed, or a vehicle was sold without knowing which authority, certificate, permit, or transaction name sits behind the event.

Research on choice overload needs careful handling here. More options are not always worse. A 2010 meta-analysis found an average effect close to zero across studies, with wide variation. Later work found that overload becomes more likely when the choice set is complex, the task is difficult, preferences are uncertain, or the person is trying to minimise effort. A public-service user facing unfamiliar terms, state rules, and real consequences can meet several of those conditions at once.

Decision paralysis is therefore not a diagnosis applied to every visitor. It is a design risk. If the first page asks people to choose between labels they cannot confidently distinguish, some will defer the choice or switch routes. In a public service, that may mean an incomplete application, an unnecessary office visit, or a support call rather than a missed purchase.

A Smaller First Decision

The prototype replaces the catalogue as the main entrance with one question: what do you need help with? A person can enter “I lost my licence” instead of comparing duplicate, replacement, renewal, correction, and state-specific service names. Search aliases, Hindi aliases, state context, and ranked matches translate the phrase into the duplicate driving licence service.

Intent-first navigation does not delete options. It changes when they appear. The system first narrows the catalogue to a plausible service, explains why it matched, and lets the user correct the interpretation. Relevant alternatives remain available without demanding that every visitor inspect the entire administrative tree.

The distinction matters. An opaque recommendation can create a new kind of uncertainty, especially if the system quietly guesses wrong. The match should be legible: show the recognised intent, the proposed service, the state rule being applied, and a route back. The prototype uses deterministic search while the interaction model is being tested; its MCP tools begin only after the user chooses the supported duplicate-licence journey.

Before Sign-In

Finding the right service removes one decision. It does not answer whether the person is ready to apply. Before sign-in, the guide shows required documents, cost, likely timing, online stages, state differences, and whether an office visit or appointment may be required.

That sequence lets a person prepare instead of discovering requirements one error at a time. It also makes the service easier to discuss with a family member, support worker, or local official because the explanation is available before a transaction begins. W3C guidance recommends clear instructions before or beside an activity and complete steps for complex work. The same principle belongs at service level, not only beside individual form fields.

A mentally demanding site is not fixed by hiding every detail. People need enough information to make the next decision, in the order they need it. The design removes irrelevant choices from the entrance and brings the relevant commitments forward.

Beyond the Happy Path

A successful demonstration moves from sign-in to upload, payment, and confirmation without interruption. Real services inherit blurred photographs, expired documents, mismatched records, unavailable appointments, interrupted networks, failed payments, and users returning after several days. Each failure can reopen the same uncertainty the intent-based entrance was meant to reduce.

In the prototype, a document correction keeps version history and blocks resubmission until a replacement is provided. A failed payment becomes a visible state with a retry path. State rules can require pre-approval or change whether an appointment is needed. The tracker preserves progress across these changes rather than scattering the application across receipts and disconnected pages.

Visible recovery cannot remove every wait, but it can remove the decision about what to do next. The person can see what failed, what remains valid, who must act, and whether an action is safe to repeat. Support staff get the same traceable history. That is a quieter experience than any colour palette can create on its own.

The Integration Layer

The build separates the interface from service rules and provider contracts. Identity, documents, applications, payments, appointments, and notifications use local or HTTP adapters. The new stdio MCP server calls a bearer-protected application route through those boundaries. It does not receive direct database access.

The server exposes eleven tools for one supported transaction: replacing a lost or damaged driving licence in Uttarakhand. The sequence starts the journey, verifies synthetic identity and contact details, prepares a prefill, waits for confirmation, records document metadata, requests pre-approval, confirms a synthetic payment, suggests appointment times, books one selected slot, and reports status.

The constraints are part of the product. Real Aadhaar data must never enter the AI host. Only a masked ending is retained. Files travel through a secure direct-upload interface rather than model context. Saving the declaration, recording payment, and booking an appointment each require an explicit confirmation or exact selection. The status tool is marked read-only.

Intent-first navigation reduces the number of choices presented to the citizen. MCP gives an authorised assistant a controlled way to continue the matched journey. Neither removes administrative complexity, and the current tools are a synthetic demonstration rather than production integration. Together, they show how less choice at the interface can be supported by tighter capabilities behind it.

The same principle informs Workflow Automation Hub: automation can carry structured work forward while consequential choices remain visible to a human. The interface may become simpler, but responsibility does not disappear behind it.