A broad ambition to use AI is a starting point. A first project needs something more specific: a piece of work you understand, a clear owner and a way to tell whether it has improved.
1. What happens today?
Follow a real task from beginning to end. Write down the inputs, the people involved, the tools they use and each point where the work waits. Look at ordinary cases and the exceptions that make the process difficult.
For example, preparing a customer response might mean reading an enquiry, finding the relevant policy, checking account context, drafting a reply and asking a colleague to approve it. Each step is a more useful starting point than “automate customer service”.
2. What would a better result look like?
Choose a measure before building. That might be the time needed to prepare a draft, the number of requests waiting for information or how often a reviewer has to correct a particular error. Record a baseline from the current process.
3. What information and judgement does the task need?
List the sources the task depends on and the people allowed to access them. Check whether they are current, complete and usable. Then decide which actions a system may take and which need a person’s review.
4. Can you test it in a contained way?
A focused pilot should let a team compare the proposed workflow with the current one. Start with a manageable set of tasks and include difficult cases. Keep a way to return to the existing process while you learn.
- Name the person who owns the workflow.
- Define one meaningful improvement and its baseline.
- Agree on access, review and escalation.
- Test with the people who will use the solution.
- Review the evidence before expanding the scope.
A successful first project gives the business more than a working tool. It creates a shared understanding of how to choose, test and adopt the next useful change.