ProfileAtlas

Case study · Client work · Local government

Role
Product manager, sole client contact
Team
5, Carnegie Mellon
When
Jan to Aug 2026
Result
Delivered; adoption not yet decided

Collier Township civic platform

A project tracker that shows residents where township projects stand while their input can still change them.

  • 224stakeholder conversations
  • 2.2 → 4.4of 5, residents' understanding once a proposal became a stage timeline
  • 11 of 12testers aged 55 to 84 would use it for a real decision
A-14PrototypeCollier Blueprint: a project page with its timeline, map and pollLead with where each project stands, and what input can still change.

AskedJan 2026

A township with public data asked for AI.

"Insightful Governance": use AI to help township staff work with civic data.

Kickoff: mapping how the township works, with the Manager, Secretary and Finance Director

LearnedFeb to Apr

The information is public. Residents find it too late.

The information is public; residents cannot find it in time. 2 or 3 residents attend the meetings where decisions take shape, about 200 attend the hearing when nothing can change, and 35 of 39 survey respondents said they do not hear about decisions.

"Didn't think they had any information on there."
Interview recruitment poster, posted around the township

TurnedApr

The problem was timing, not analysis.

The Broken Cycle showed the input window closing before the township starts telling anyone. The goal moved from faster analysis to earlier reach, and AI moved backstage to the staff side.

The modelResidents heard about projects only weeks before the vote, after their input could change anything. Blueprint moves the telling to the start of the window.

  • The brief, as asked
  • The turn
  • Found and made
  • Set aside
  • Evidence, one dot per count
The broken cycle of civic trust

MadeMay to Jul

Eight prototypes. Evidence picked the two that shipped.

Eight prototypes, four of them mine. Evidence cut my rewards prototype (residents said rewards "cheapen" engagement) and set aside my map; my two coded one-stop-shops became the product. The timeline was rebuilt vertically, anchored at the current stage, after only 4 of 12 testers found next steps.

Comprehension 2.2 to 4.4 of 5 when the same proposal became a stage timeline

LandedJul 2026

Start manual. Add AI when the township is ready.

No AI on the resident side; a staff side that works fully before AI; a four-tier roadmap ($30K to $370K) so the township starts manual and adds AI over time. Presented to the Board of Commissioners and township staff. Delivered; adoption not yet decided.

Mobile: a project page with its details and map
Mobile: a gated poll that tells residents how their input is used

Details

My role
Product manager and sole client contact across both semesters. Ran the weekly client meetings, led department-head research, built four of the summer prototypes including both coded one-stop-shops that became the product, shaped the final report and case study, and presented to the Board of Commissioners and township staff.
Team
Christy Yu (design and development), Bosi Li (design and development), Laila Dancy (UX research), Esther Suh (policy research, through April)
Proof
224 stakeholder conversations: qualitative interviews, a 39-respondent survey and prototype test sessions across 8 concepts. Comprehension of a real proposal rose from 2.2 to 4.4 of 5 when it was shown as a stage timeline.
Set aside
AI on the resident side (cut: residents wanted "a real person" and the township has no IT staff to maintain it). My own rewards prototype (cut: rewards "cheapen" engagement). Ticketing and events (cut: the township's existing software already does them).
Where it landed
Delivered, not adopted. Working prototype, design system, print newsletter companion, four-tier roadmap ($30K to $370K).
More numbers
  • 4 build tiers, each useful on its own
  • 9 of 9 department heads said little or no added work
More exhibits17 · show all
AskedJan 2026
A-16DrawingKickoff sketch: the ideal loop and the problem loop between residents, government and departmentsResidents only hear back when they ask; the loop is broken.
MadeMay to Jul
A-17PhotographsHow-might-we matrix: 12 questions, about 90 concepts, voted to 7 directionsOpen wide before testing anything.
A-03PrototypeResident prototype: the proposal and its potential benefitsPut the trade-offs on the page from day one, costs and benefits side by side.
A-04PrototypeResident prototype: what you can do before the voteName every real way to be heard while it still counts.
A-05PrototypeResident prototype: the board's perspectiveBe honest about what residents can and cannot change.
A-06PrototypeResident prototype: share your feedbackAsk for conditions, not just yes or no.
A-07PrototypeEngagement quiz: how residents want to hear from the townshipLet residents choose their channel instead of guessing.
A-08PrototypeResident prototype: the decision and its conditionsReport the result, including what was not included.
A-09PrototypeResident prototype: what we heard and what changed because of itClose the loop: show what input changed, and what it did not.
A-10PrototypeResident prototype: what you can do after the voteThe vote is not the end; show what can still be shaped.
A-11PrototypeResident prototype: closing reflectionAsk whether the process felt fair, not only the outcome.
A-12PrototypeCollier Engages: my civic bingo prototype, points for giving feedbackTested whether points drive participation. Only 3 of 13 expected to know more; dropped.
A-13PrototypeCollier Community Map: my map prototypeSet aside: list and calendar views won, and the township already had ticketing.
A-18PrototypeCollier Engages: my rewards prototypeCut. Rewards "cheapen" engagement; responsiveness is the motive.
LandedJul 2026
A-22DocumentDepartment heads on the staff side: "The pie chart is dangerous"Show raw counts and comments, not only a summary chart.
A-23DocumentValue before AI: every staff feature works manually, AI only addsStart manual; add AI tier by tier.
A-24DocumentManual vs AI: responding to residentsAI drafts and flags; staff approve every reply.

ThreadsPublic systemsTrust in AIAccess

Related