If you are unsure where to start with AI, ask people what they have to do again every week. They may prepare reports, gather customer information or copy data between systems. Recurring work gives you a concrete way to see what takes too long and what needs to improve.
I would choose a task that someone on the team understands well. You need a person who will notice mistakes and can judge whether the new approach helps.
A report may be a data problem
Suppose a weekly management report takes two hours of finding figures and twenty minutes of writing commentary. Those are illustrative numbers, but they show why it is worth following the whole task.
If AI writes the commentary in a minute, most of the work remains. Connecting data from existing systems might help more. So might agreeing which spreadsheet has the correct figures.
When choosing a first project, I would count the time spent preparing information and checking the result. Generation speed is only a small part of the job.
What will the business do with the time?
It is easy to multiply an hour saved each week by the number of employees and their hourly cost. That does not mean the same amount disappears from the company’s expenses.
Be specific about how the freed time would be used. A salesperson could prepare another proposal. A team could clear requests that have been piling up. Sometimes the benefit is simply less work in the evening. That can be worth doing even if payroll costs stay the same.
Try the task on a few ordinary cases and one that tends to cause problems. Have the person who does the work check the results. Once they can tell you how much time is left after corrections and review, you have a basis for deciding whether to continue.