Build a governed knowledge base
The assistant should search regulations, procedures, opening times and pages with a defined editorial owner. Relevant answers should identify their source and the date of the information.
Resolve duplicated or contradictory content before automating it. AI cannot create a coherent municipal policy from sources that disagree.
Define what the assistant may and may not do
Explaining required documents or directing a citizen to a service is different from interpreting an individual case. The system should recognise these limits, avoid collecting unnecessary data and escalate questions that require human analysis.
An alternative channel must remain available for people with digital-access or accessibility needs.
Escalate with context
A transfer to a person should not force the citizen to repeat everything. The system can prepare a summary, list the sources it consulted and identify the unresolved issue.
Staff should be able to correct answers and flag outdated content. These corrections improve knowledge-base maintenance and future testing.
Evaluate the service, not just automation
Measure first-contact resolution, appropriate transfers, corrected answers, abandonment, accessibility and satisfaction. Fewer calls are not a success if citizens can no longer resolve their requests.
Frequently asked questions
Can the assistant answer questions about individual cases?
Only with appropriate integration, authorisation, authentication and rules. Otherwise it should provide general information only.
How can information be kept current?
Identify official sources, editorial owners and review dates, and test them periodically.
Can a citizen request human support?
A clear route should exist, especially for sensitive, complex or rights-affecting situations.
Architecture: AI for complaint handling in local government →