UXR • Civic Technology • 2025
Role
Team
Timeline
09/2025-12/2025
Tools
Overview
The City of Dearborn had AI-powered online services up and running. Our job was to get more residents to use them. But the deeper we got, the more the question shifted — not "how do we drive adoption?" but "is this tool actually serving the people it's supposed to?"
The Problem
Dearborn launched its AI tool around 2024, but residents didn't trust it, couldn't get answers from it, and kept asking for a human instead.
The surface problem was a tool that wasn't working well enough. The deeper issue was that the city and its residents were using it with completely different goals in mind.
The Research
Early Solution
What stakeholders said
The pivot wasn't a choice. It was what happened when the plan met reality.
The Pivot: Three times we were wrong
Midway through the project, our professor stopped us and asked: "What do the residents actually need?"
I didn't have a good answer.
We'd been so close to the city's framing — their goals, their constraints, their language — that we'd stopped asking whose problem we were really solving. That question reset everything.
The decisions that shaped this project didn't come from ideation. They came from hitting a wall.
The Two Things We Could Change
Not everything from the feedback session was rejected, but not everything was actionable either. Working within what was available to us, we focused on 2 areas we could actually design, prototype, and test:
The chatbot UI — making it visible with animation, different color and a bilingual greeting
BEFORE
AFTER
The phone menu — rewriting the conversation within a system we couldn't modify
Before: "Hi, thank you for calling the City of Dearborn. I'm your AI Assistant, you can ask me anything, or speak to a live agent."
After: "…For trash and recycling, say 'Trash and Trees.' For permits and inspections, say 'Permits.' For water bills, say 'Water.', ' For ….', …"
We ran a usability test to validate both.
Final Deliverables
Three scenarios where the AI genuinely helps: after hours when no one's picking up, when a resident's first language is Arabic, and when a voice call is difficult for accessibility reasons.
Our solution targets those moments — not everyone, all the time.
Web Chatbot
Animated pop-up button with high contrast — visible regardless of the seasonal background image
Bilingual greeting (English + Arabic) on open — signals the tool was built for the whole community
Named "DB" — makes it feel like a person you're talking to, not a system you're navigating
"Did I answer your question?" — a short satisfaction check after each response
Phone Menu
"Trash and Trees" first — highest call volume at the top, not buried under department names
Natural language prompts — residents say "missed pickup" or "broken cart," not department terminology
Two levels max — shallow structure reduces wrong turns and cognitive load
Say "live agent" at any point to reach a person — no need to navigate to the end of the menu; Say "repeat" to hear the options again
Auto-handoff after 3 failed attempts — the system doesn't keep trying when it's clearly not working
Separate complaint route to Mayor's office voicemail — complaints need a different channel than service requests
We presented our final designs to city staff and community members at the Dearborn Public Library. Residents said they'd try it, and that it might actually help.
Reflection
This project reshaped how I think about UX in practice. I've always known it's more than UI, but working in civic tech made that concrete. Residents don't choose their city services the way they choose an app; they depend on them. That changes the stakes, and it changed how I approached the design..
I also learned that constraints aren't obstacles; they are the actual shape of the project. Knowing when to stop is a design decision too. Naming what's out of scope is more honest than pretending every problem is solvable. Most importantly, this project reminded me that the client and the user are rarely the same person. Good design serves the people using it first. Everything else follows.
