Public-sector AI should be treated as organisational change with technology inside it. The useful starting point is a clearly defined service or operational problem, supported by accountable ownership and evidence.
1. Begin with service pressure, not a shopping list
Begin by describing the current work, pressure and desired outcome in plain language. Teams should document assumptions, involve the right service, technology and governance colleagues, and agree how success and risk will be monitored.
2. Map processes, data and ownership
The approach should remain proportionate to the impact, information involved and people affected. Teams should document assumptions, involve the right service, technology and governance colleagues, and agree how success and risk will be monitored.
3. Choose a contained, measurable pilot
The approach should remain proportionate to the impact, information involved and people affected. Teams should document assumptions, involve the right service, technology and governance colleagues, and agree how success and risk will be monitored.
4. Set governance before implementation
The approach should remain proportionate to the impact, information involved and people affected. Teams should document assumptions, involve the right service, technology and governance colleagues, and agree how success and risk will be monitored.
A practical next step
A readiness conversation can turn the topic into a bounded discovery exercise: establish the baseline, identify constraints, compare opportunities and select a first pilot only when the evidence supports it.
We don't start with technology. We start with the problem.