What will make your first steps a success
The departments that have moved furthest with AI share something in common: they did not start with technology.
They started with a problem, a person accountable for solving it, and data good enough to work with. Getting these foundations right is the difference between a pilot that scales and one that stalls.
Step 1
Pick the right first problem
Choose a use case that is high-volume, repetitive, and clearly bounded. Drafting routine correspondence, summarising long documents, answering frequently asked questions from staff or citizens. These are low-risk and deliver visible, measurable time savings quickly. Avoid starting with anything that touches complex decisions about individuals.
Step 2
Get a senior sponsor with real authority
Not someone who is ‘supportive in principle.’ You need a Director General, Chief Digital Officer, or equivalent who can unblock procurement, override institutional inertia, and make decisions when things stall. Without this, projects drift. With it, they move.
Step 3
Check your data before anything else
Poor, patchy, or siloed data is the single biggest reason AI projects fail to scale. Before committing to a use case, ask: is the data accessible, consistent, and clean enough for AI to work with? If the answer is no, fixing the data is the project, and that is valuable work in its own right.
Step 4
Involve your frontline staff from day one
The best tools fail if the people using them were not involved in designing them. Case workers, teachers, administrators, and frontline staff understand the real workflow and the real edge cases. Involve them early. They will spot problems that no requirements document will capture.
Step 5
Sort governance before you go live
Establish who is accountable for AI decisions, how outputs are reviewed, and what happens when something goes wrong. This does not need to be complex. A clear ownership structure, a simple escalation path, and documented human oversight for any consequential decision is enough to start. Build governance in; do not bolt it on afterwards.
Step 6
Measure something that matters
Define your success metric before you build anything. Hours saved per week. Cases processed per day. Citizen satisfaction scores. Reduction in repeat contacts. Without a baseline and a target, you cannot demonstrate value and you cannot secure the funding to scale.
Step 7
Communicate early and honestly
Uncertainty about AI drives resistance. Staff who do not understand what is being built, and why, will assume the worst. Regular updates, honest about both progress and setbacks, build trust and reduce friction. Frame AI as a tool that supports people, not one that replaces them.
Finally...
A useful first question to ask your team
Before any strategy document, workshop, or vendor conversation, ask: “What takes our people the most time and adds the least value?” The answer is almost always your best starting point. It is a problem people care about solving. It has a measurable baseline. And when you fix it, the impact is immediately visible to the people who matter most: the staff who do the work and the citizens they serve.