Many organisations approach artificial intelligence by asking which tool to buy. That question usually comes too early. The organisations that adopt well tend to work through a different sequence.

**Start with tasks, not tools**

List the tasks your teams repeat weekly: summarising documents, drafting routine correspondence, categorising requests, preparing first-draft reports. Adoption succeeds where a task is repetitive, rule-bound, and low-risk if a draft needs editing.

**Decide what must never be automated**

Just as important as where AI helps is where it should not be used. Final decisions affecting people, confidential records, and anything requiring legal accountability need a human owner. Writing this down early prevents difficult conversations later.

**Address data handling before capability**

Staff need to know what may and may not be entered into an external system. A short, clear internal rule is worth more than a long policy nobody reads.

**Train judgement, not just prompting**

The skill that matters is not writing clever instructions. It is recognising when an output is wrong, incomplete, or confidently misleading — and knowing what to do next. That is a reviewing skill, and it needs practice with real examples from your own work.

**Expect the first attempts to be unimpressive**

Early results are usually mediocre. Teams that treat the first month as learning rather than proof tend to keep going. Teams expecting immediate transformation usually stop.

**Give people a reason to adopt**

If the benefit accrues only to the organisation, adoption stalls. If it removes work people dislike, it spreads on its own.