A small charity starts with AI safely by picking one repeating office job, using only information the trustees have already approved, and keeping a person responsible for what goes out. Do not begin with a public chatbot, and do not begin with beneficiary records.
- One repeating office job.
- Approved information only.
- A person still signs it off.
- A written reason to stop.
Trustees do not need a new strategy document before anyone tries a tool. They do need a boundary that a busy manager can keep. The ICO guidance on AI and data protection expects you to know what personal information is used, why, and how people remain in charge. The Charity Commission essential trustee guidance (CC3) still puts that responsibility with the board. The Charity Commission note on charities and artificial intelligence asks trustees to understand the tool they are allowing before staff start using it.
A safe first checklist
- Which job? Pick one repeating office task, such as turning approved notes into a first draft of a routine letter or a board summary. Not a public chatbot.
- Which information? Use policies, published reports or other material the trustees have already agreed can be used. Leave case files, donor lists and staff records out of the first trial.
- Where does it go? Write down whether the tool runs on your own system or on someone else’s. If you cannot answer that, do not start.
- Who checks? Name the person who reads the draft before it is sent, filed or shown to a trustee.
- When do we stop? Agree the sign that the trial is not helping, and stop it without buying a second tool.
What to leave until later
Sensitive beneficiary, donor or staff information is a later decision, not the warm-up. Read Can charities use ChatGPT with sensitive data? before anyone pastes a case note into a public chat. Where the data is processed, which sources are allowed and who reviews the output are covered in private AI for charity data.
The board questions to ask before anyone approves a tool are in the trustee section of 12 questions before an AI project.
After the first job works
When the first job is clear and the information is allowed, the work can move onto a system the charity controls, with a person still signing off what goes out. Our AI process improvement work starts from one job like this, not from a tool. If the job depends on case records, referrals or outcomes, see custom CRM and case management for charities.
Sources
- ICO: Guidance on AI and data protection
- Charity Commission: The essential trustee (CC3)
- Charity Commission: Charities and artificial intelligence
Questions
Should a small charity start with a chatbot?
No. Start with one repeating office job that already has approved information and a person who checks the result. A public chatbot can wait.
Can we use beneficiary records in the first project?
Not as the first step. Start with policies, published reports or other information trustees have already approved. Sensitive records need a separate decision about where the data goes.
Who is accountable?
The trustees remain accountable. Name one member of staff who reviews what the tool drafts before it is used.