Pick one resident-facing service, such as housing repairs or the customer contact hub, and assess everyone who will review an agent's work there. The test, set in local government, covers held actions, refusals, vulnerable residents and the record. The report decides the first agent, keeps it away from decisions about people's entitlements, and shows what to train first.
Why does a council need its own approach?
A council is dozens of services under one roof: waste, housing, council tax, parking, planning, social care, libraries. Each has different rules, different residents and different risks. An agent that suits missed bin reports would be wrong for council tax hardship cases. So the assessment must be run service by service.
Councils also carry public duties of transparency and fairness, and residents who contact them include some of the most vulnerable people in the area. A reviewer who approves an agent's reply must be able to spot when a routine request hides a safeguarding concern or a hardship case.
The head of customer services usually owns the assessment, with the service manager for the chosen service. For the concept, see what a readiness assessment is.
Who should take it?
Up to 25 people from one service. Good first candidates:
- Housing repairs: call handlers, repairs coordinators and the team leader who decides emergency classifications.
- The contact hub: advisers who handle general enquiries across several services and route them on.
- Waste and street services: officers who handle reports and complaints.
Include the person who handles members' enquiries for that service. Councillors raise cases on behalf of residents, and those cases need careful handling whatever the agent drafts. Advice on order is in who in a local council needs AI agent training first.
The steps
- Person (head of customer services): chooses the service and, with the service manager, lists what the agent will never decide: anything about entitlement, enforcement or safeguarding.
- Person (head of customer services): asks for the test set in public service, with questions about resident requests, repairs and complaints.
- Person (each member): sits the test alone, in working time.
- Person (head of customer services and service manager): read the report by concept.
- Person (service manager): picks the first agent and its depth. In repairs, that might be an agent that drafts job tickets from residents' reports at depth 2 for a coordinator to raise.
- Person (team leader): trains the weakest concept in short sessions around peak contact times.
- Agent, depth 1: reads past closed requests in shadow, proposing a category and priority for comparison with what the team decided.
- Person (cohort): reassesses before the agent drafts on live requests.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Choose service | Nothing yet | None | Scope and never-decide list |
| Sit the test | Nothing yet | None | Individual answers |
| Read the report | Nothing yet | None | Gaps by concept |
| First agent | Named as candidate | None | Agent and depth |
| Train the gap | Runs in practice labs | Practice | Learning |
| Shadow | Proposes category and priority | 1 | Comparing with real decisions |
| Reassess | Nothing new | 1 then 2 | Go-live |
What does the test usually find in councils?
Hidden vulnerability. A resident reports damp. Another mentions in passing that their child has asthma and the heating is off. Staff generally recognise the second as urgent once prompted, but in the test many answer as if the agent's category settles it. The point to learn: an agent's bounds can hold messages that mention health, children or risk, but a person must then act on the hold.
Held versus refused. Council staff are used to saying "that is not a council service". They need to tell the difference between an agent holding a request for a person and refusing it as outside its bounds, and to know a refusal still needs a person to pick it up and redirect the resident.
The record. Complaints, ombudsman referrals and requests for information all depend on reconstructing what happened. Staff should know every verdict and every approval is filed, with their name.
What does the report change?
It picks the first service, sometimes a different one. If the repairs team scores poorly on vulnerability and the waste team scores well, the first agent may belong in waste, where the stakes are lower.
It shapes the bounds. The never-decide list and the vulnerability triggers become part of the agent's bounds, so the agent holds rather than categorises those messages.
It sets training before go-live, and it gives the council evidence for its members and auditors of how staff were prepared.
What should the council keep as evidence?
The never-decide list, agreed with the service manager.
The report by concept.
The decision on the first agent and its bounds.
The reassessment before live work.
What should the head of customer services measure?
- Scores by concept for each service assessed.
- During shadow, requests where the agent's priority differed from the team's, especially emergencies.
- After go-live, the time from a held vulnerability message to a person acting on it.
Where does it go wrong in councils?
One assessment for the whole council. The average across services hides the service that is not ready.
Leaving out members' enquiries. Cases raised by councillors are high-profile. The officers who handle them must understand what the agent drafted.
Starting where the pressure is greatest. The service with the longest queue is often the one with the most vulnerable residents. Start where the team is ready, then move.
The council-wide plan follows in how to roll out AI agent training across a local council, and elected members and senior officers need their own preparation, covered in what AI agent training a local council board needs. A regulated private-sector comparison is the lender version. The topic guide is training teams for AI agents.
Questions people ask
Should social care staff take it at the same time?
Not at first. Social care decisions are high-stakes and need their own scoping. Start with a lower-risk service and bring social care in once the council has experience.
Do agency and temporary staff need assessing?
Yes, if they will approve agent drafts. Councils rely on temporary staff in busy services, and the record names whoever approves.
How often should a council reassess?
Each time a new service adopts an agent, and before raising depth in any service.