The first AI project sets the tone for everything that follows. A well-chosen one builds confidence and skills; a poorly chosen one makes the next approval harder.
Look for repeatable work
Tasks that follow a pattern, use available information and have a clear definition of done are good candidates. Examples include sorting requests, drafting standard replies or summarising long documents.
Avoid starting with something that depends on rare judgement or sensitive decisions.
Keep people in charge
Design the first release so a person reviews the output before it is used, and capture where they make corrections. Those corrections are valuable feedback.
This approach also helps staff build trust in the tool.
Check the information is ready
AI is only as useful as the information it can reach. Make sure the data or documents are accessible, reasonably accurate and permitted for this use.
Fixing this early avoids delays later.
Decide how you will know
Agree on what success looks like before building, such as time saved or fewer errors, so that the result can be judged on evidence.
Share the outcome openly, including what did not work, to guide the next step.
- Pick bounded, repeatable work.
- Start with human review.
- Confirm the information is ready.
- Define success before you build.
