The question is almost never “should we use AI”. It is “where do we start, and what will embarrass us”. Here is a practical answer for an organization with no technology staff.
Start with what you already pay for
Most Canadian not-for-profits already have Microsoft 365, often at a nonprofit rate. That means capable AI is frequently already available or available cheaply, without procurement, a new vendor or a new data-processing agreement.
Before evaluating anything new, find out what your existing licences include. This one step resolves a surprising share of “which AI tool should we buy” conversations.
The three tasks worth automating first
These are chosen because they save real time, carry low risk, and do not require anyone to change how they work.
- Meeting notes and summaries. High volume, low stakes, and immediately obvious value. Nobody misses writing minutes.
- Drafting routine correspondence. Thank-you letters, acknowledgements, standard replies. A human still reviews and sends.
- Finding information. “Which policy covers this?” across a shared drive nobody has reorganised since 2019.
Notice what is not on the list: anything donor-facing that goes out unreviewed, anything that makes a decision, and anything involving sensitive personal information.
Where AI should not be used
- Decisions about individuals. Who receives a service, who is eligible, who is prioritised. Even as a suggestion, this introduces bias risk your organization cannot audit.
- Sensitive personal data, without a policy. Client records, health information, anything about a vulnerable person. Not because the tools are necessarily unsafe, but because “we had no policy” is not a defensible answer to a funder or a privacy commissioner.
- Anything published without review. Not because AI writing is bad, but because accountability for what your organization says has to sit with a person.
The governance you actually need
Not a forty-page policy. For most small organizations, one page covering:
- Which tools are approved
- What may and may not be pasted into them
- That anything published is reviewed by a person
- Who to ask when unsure
That page, understood by everyone, is worth more than a comprehensive policy nobody reads.
A realistic sequence
- Find out what your existing licences already include
- Pick one task from the list above and try it for a month
- Write the one-page policy before the second use case, not after
- Measure whether it actually saved time, honestly
- Only then consider buying something new
The honest caveat
AI is genuinely useful for the specific problem most small not-for-profits have, which is too much administrative work and not enough people. It is not useful for the problems that get the most attention. Treat anything promising transformation with the scepticism you would apply to any other vendor claim.