Start by looking for a recurring task that causes frustration, takes time or leads to delays. AI may help, but some tasks need only a simple rule or a process change.
Look at the work
Ask your team where they copy information, chase replies, prepare reports or pass work to another person. Record how often it happens and what goes wrong.
Choose a manageable task
Compare likely benefit with setup cost and risk. A draft reply for review or a reminder for a colleague may be easier to test than a fully automated decision.
Check the result
Agree what success looks like, who reviews the output and what happens if the system fails. Measure time and quality before deciding whether to expand.
Compare candidates
For each task, note its frequency, the time it takes and the effect of an error. Ask the people doing it which parts require judgement and which follow the same steps each time. A frequent task with clear inputs and a straightforward review point can be a practical first test.
Try representative examples, including incomplete or unusual ones. Compare the new process with the current one for time, accuracy and the effort needed to check it. Keep the original process available until the team is confident in the change.
A simple way to compare tasks
Imagine a team copies 40 enquiries each week, taking three minutes each. That is two hours of work before checking or corrections. A weekly report takes one hour, but has consistent inputs and a named reviewer.
The enquiries take more time overall, but the report may be the safer first trial if enquiries contain sensitive information or frequently need judgement. Compare a small sample of each, including setup and review time, before choosing. These numbers illustrate the comparison, not a promised saving.
When you have a candidate, use the AI readiness checklist to plan a small trial.