Where should we start with AI? Find one useful opportunity

It is easy to start an AI discussion with a demonstration. A tool answers a question, writes a summary or completes a task in seconds. The more useful question comes afterwards: where would that help our organisation, and what would we need to make it work?

A good starting point is one piece of work that people understand well. It should matter enough to improve, but be small enough to examine properly. That gives you a practical way to learn before choosing a wider programme.

Look for a task people can describe

Ask your team where they repeatedly search, copy, chase, check or explain. A customer-service team may answer the same questions using information spread across several documents. A charity may spend time finding approved guidance or moving referral details between systems.

These are promising places to explore. Ask the person doing the work to show you a typical case and one that needs extra judgement. Together, those examples help reveal where AI, automation or an existing software feature could make the work easier.

Write the opportunity in ordinary language: who needs help, what they are trying to do and what would improve. “Help colleagues find current guidance” is a more useful starting point than “we need an AI assistant”.

Check the information and the route through the work

Before comparing tools, identify what the task depends on. Where does the information live? Who keeps it current? Who should be able to see it? What happens after someone receives an answer?

A knowledge tool might find a passage quickly, but the service still needs a way to deal with conflicting documents or a question the material cannot answer. A workflow might save copying, while still needing an agreed route for an incomplete record.

Security belongs in these early decisions. The NCSC's secure AI guidance covers design, development, deployment and ongoing operation, making it a useful reference throughout a service's life. It is established guidance, rather than a new announcement. NCSC guidance

Include the people who will use it

The team needs more than access to a tool. People should understand the intended use, what to check and where to get help. Someone also needs responsibility for maintaining the information and responding to feedback.

The government's AI-upskilling research is a useful source for planning workforce support. It brings together practical principles and evidence for employers developing AI capability; it does not turn one training session into a guarantee of adoption. Skills for AI research

For your first project, make learning specific. Can a colleague complete a representative task, recognise a poor answer and take the right next step? Those questions are more revealing than whether everyone has tried the software.

Decide what you want the first test to tell you

A pilot should answer a decision. You might want to know whether approved source material supports useful answers, whether an integration handles the common cases, or whether staff find the new process easier to use.

Agree a baseline and a small set of measures. These could include task completion, answer quality, time spent checking, successful handovers or unresolved exceptions. Record the effort needed to keep the service working as well as the apparent benefit.

Then set a review point. The next decision may be to extend the service, improve its foundations, try a different approach or use an existing software feature instead. Each is a useful outcome if the test gives you better evidence.

A useful first step

Choose one task, involve the people who know it and make the dependencies visible. You will have a stronger basis for selecting technology and a clearer question for any supplier or delivery partner.

Our AI opportunity review helps turn that starting question into practical recommendations. You can also explore data and systems readiness and people, skills and adoption before deciding where to begin.