Government,
Simplified
A GovTech platform built for intake workflows had a usability problem. Forms were repetitive, inaccessible, and exhausting. I introduced an AI agent that changed the experience entirely.
Impact
68%
Faster intake completion
Average time to complete a full intake workflow dropped significantly after AI guidance was introduced
82%
Fewer data entry errors
Guided input reduced re-entry and correction cycles that were common with unassisted form completion
91%
Intake completion rate
Up from 61% before the AI agent, driven by reduced friction and real-time assistance
3x
Faster worker onboarding
New government workers reached proficiency significantly faster with a guided system vs. manual training
The Problem
Government intake workflows are inherently complex. The information required is detailed, the stakes are high, and the people using the system range from experienced case workers to first-day employees to individuals with varying abilities navigating the process themselves.
The platform had grown in functionality but not in usability. Forms were long, repetitive, and unforgiving. The same information was requested multiple times across different screens. There was no guidance, no context, and no accommodation for users who needed assistance.
Workers were spending more time fighting the system than actually doing their jobs. Incomplete submissions, re-entry cycles, and training overhead were all symptoms of the same root cause: the product was built around data requirements, not around the humans filling them in.
The Approach
The interface needed to change. Not the underlying data model, not the compliance requirements — just the way users interacted with it.
I designed a solution around an AI agent that sat in front of the intake workflow and guided users through it conversationally. Instead of presenting a wall of form fields, the agent asked one question at a time, confirmed answers, carried context forward, and surfaced help exactly when it was needed.
Accessibility was treated as a core requirement, not an afterthought. The agent interaction model was built to support users with disabilities from the start, including screen reader compatibility, keyboard-only navigation, and plain language throughout.
Redundant input was identified and eliminated. Information the system already had, or could infer, was never asked again. That single change removed the most cited frustration in user feedback.
What Was Built
01
Conversational Intake Agent
Replaced static, multi-page forms with an AI agent that walked users through the intake process step by step. The agent asked questions in plain language, confirmed inputs, and moved users forward without requiring them to interpret complex form fields on their own.
02
Repetition Reduction
Identified the most common points of redundant input across workflows and engineered the agent to carry context forward. Information entered once was referenced, not re-requested, eliminating the repetition that was the biggest source of user frustration.
03
Accessibility-First Design
Designed the agent interaction model to meet the accessibility requirements of a government product serving users with disabilities. Keyboard navigation, screen reader compatibility, and simplified language were built into the experience from the ground up.
04
Guided Completion Logic
The agent understood which fields were required, which were conditional, and where users historically dropped off. It proactively surfaced guidance at the right moments, reducing incomplete submissions and the back-and-forth that followed them.
The Outcome
Intake completion rates climbed and error rates dropped. Workers moved through the process faster and with less frustration. The accessibility gaps that had excluded some users entirely were closed.
What changed was not the data the platform collected. What changed was the experience of providing it. That distinction matters in government software, where you cannot ask users to just switch tools. You have to meet them where they are and make the process work for all of them.
Client details are confidential. This case study has been shared with permission.