The short answer
A good first AI policy covers five things: which tools are approved, what client data can go where, how AI output is reviewed, what training staff need, and what to do when something goes wrong. Keep it short enough that people actually read it.
Key takeaways
- Short and read beats long and ignored.
- Name the approved tools explicitly.
- AI output is always a draft until a qualified person reviews it.
- Review the policy every six months. AI moves fast.
Why now
If you don't have a policy, your team is already making their own rules. Some will avoid AI entirely and fall behind. Others will paste things they shouldn't. A policy gives everyone permission to use AI well, inside clear lines.
The five sections
1. Approved tools
List them. Anything not on the list needs sign-off.
2. Client and personal data
What can go into which tool. Green, amber, red works well.
3. Review and quality
AI output is a draft. Numbers are checked to source. A qualified person signs off anything going to a client.
4. Training
What staff need to complete before they get access.
5. Incidents
If something goes wrong, who to tell and how quickly. Make it safe to own up.
Common mistakes
- Copying a corporate policy that's 30 pages long.
- Banning AI outright (people use it anyway, just without rules).
- Writing it once and never updating it.
Questions accountants ask
Who should own the AI policy?
A partner or director with an AI lead doing the day-to-day. It needs someone senior enough to make decisions.

Chartered Accountant · Founder, Off Balance
Chartered Accountant with a background at Deloitte and Xero, plus years in sales, leadership and startups. Hayden helps accounting firms across New Zealand and Australia work out where AI fits, then actually put it to work.
More about Hayden →