Public Sector · 2022
Slow AI, on purpose
A municipal services team wanted to help residents navigate a maze of benefits programs. The temptation was a chatbot. The right answer was something quieter.
- Client
- Mid-sized European city
- Role
- Product & Policy Advisor
- Duration
- 5 months
- Year
- 2022
01 — The challenge
What we walked
into.
Residents distrusted automated systems. Caseworkers were overloaded. Any tool had to make both groups' lives meaningfully better — or it would be quietly abandoned.
02 — Approach
How we got
there.
01
Shadowed caseworkers for two weeks before writing a single requirement.
02
Designed a hybrid flow: AI surfaces likely eligibility, a human always confirms.
03
Wrote plain-language explanations co-edited with residents from three neighbourhoods.
04
Built an audit log that residents could actually read.
03 — Outcome
What shipped.
+34%
Benefits successfully claimed
−22 min
Average caseworker time per applicant
92%
Resident trust score (post-pilot survey)
"The best AI in public services often looks like better paperwork — and that's the point."
Jenny Kan — reflection