Public-sector AI rollout should begin with service accountability, not generic automation. The first design question is who remains responsible when an answer, route, or escalation affects a citizen, employee, or regulated process.

Define transparency requirements before model selection. Teams need source-visible responses, accessibility checks, clear escalation paths, and logs that explain why the system routed or drafted a response.

Measure consistency, queue reduction, evidence quality, and reviewer workload together. A public service desk is successful when it improves operational flow without making accountability harder to inspect.