AI in Smart Cities and the Public Sector: What African Cities Can Learn from Hangzhou and Shenzhen

For an African city delegation studying AI in China, the starting point should be a public service rather than a control-room display. Select a problem residents experience, such as delayed permit responses or unresolved maintenance requests. Then examine whether a proposed system improves that service without weakening accountability, access or the ability to challenge a mistake.

Follow one service request from start to finish

Ask a host to trace a representative request through intake, classification, assignment, resolution and resident feedback. Identify which department owns each stage and where staff use judgement. Record the information exchanged between departments and the permissions that support it.

A dashboard may display many data feeds while the underlying service remains fragmented. Ask for completion times, unresolved cases and the reasons requests are reopened. Check whether an apparent improvement reflects automation, staffing changes or a simpler procedure.

An illustrative permit-information assistant

Imagine a city testing an assistant that explains the documents required for a permit. Its initial scope could be limited to retrieving approved guidance and directing unresolved questions to staff. It should not be assumed to have authority to approve or reject an application.

Build a test set covering ordinary questions, outdated terminology, incomplete information and questions in the languages residents actually use. Include people who need an assisted or offline service. Evaluate whether answers cite the current official guidance, acknowledge uncertainty and provide a usable escalation route.

Measure incorrect guidance, staff correction time and repeat contacts alongside response speed. Fast answers are not a service improvement if residents arrive with the wrong documents. This is a proposed assessment exercise, not a claim about a particular city's deployment.

Questions that need an accountable owner

Area	Question for the host or supplier
Service delivery	Who remains responsible when an automated response is wrong?
Data	Which fields are necessary, who may access them and when are they removed?
Accessibility	What service remains available to people who cannot use the digital channel?
Review	How can a resident reach a person and challenge an adverse result?
Procurement	Can the city export records and change suppliers without losing service history?
Operations	Who funds staffing, maintenance, security updates and training after the pilot?

Privacy, administrative authority and procurement rules depend on the destination jurisdiction. A process observed in China cannot establish what a city elsewhere is permitted to deploy. Relevant local officials and qualified advisers should resolve those questions before a live trial.

Compare arrangements, not just city names

Use the China destination guide to explore possible visit locations, then select hosts that can discuss the relevant service and operating model. Request participation from the service owner as well as the technology provider. A municipal presentation and a supplier demonstration answer different questions; neither alone establishes the other party's performance.

During synthesis, separate observations from supplied claims. If a host cannot share detailed records, retain that limitation in the decision note instead of converting an impressive presentation into verified evidence.

Bring back a scoped public-service trial

The delegation's output should name the service, its baseline, the intended users, data permissions, a responsible department and the conditions for continuing or stopping. Start with a bounded workflow where errors can be detected and corrected. Build consultation and feedback into the evaluation rather than waiting until procurement is complete.

A sector-focused executive program can help structure these discussions. Specific municipal access, company visits and records depend on prior confirmation. The city government study-mission guide provides additional questions for the delegation brief.

References and scope

• OECD AI Principles: a reference for transparency, robustness and accountability.
• NIST AI Risk Management Framework: a voluntary method for organising AI risk assessment.

These sources support the evaluation approach. They do not certify any city platform or establish compliance with local public-sector requirements.