Sample Team Cheatsheet | M.A.I. Consulting

Fundraising & Partnerships

AI cheatsheet

Cleared against AI Use Policy v1.2
Approved tools: three · Review: every 6 months

1DelegateWhat AI does — and doesn't

Hand these over

First draft of a donor report, from your own notes and the last one
Reformatting an existing proposal into a new funder's template
Turning a 40-page call for proposals into an eligibility checklist
Rewriting a concept note in plain language for a community audience
Ten subject-line variants for a campaign email

Never these

Budget arithmetic accepted without recomputing it yourself
Anything containing a beneficiary's name or case detail
The decision on whether to submit, and the sign-off that follows

2DescribeHow to ask

Pattern A — role, audience, constraint, format

You are drafting for [funder].
Audience: their programme officer.
Limit: 800 words, their four headings.
Use only the attached figures.

Pattern B — give the source, not a summary

Here is last year's report and this
year's data. Match that structure
and tone. Flag anything the data
does not support.

Pattern C — make it interview you

Before drafting, ask me the five
questions you need answered to
write this well.

3DiscernFour checks before you use it

Figures. Every number traced to a source document you can point at. If you cannot find it, it is not a number
Facts. Every organisation, deadline and eligibility rule checked against the original call, not against the summary
Language. Read once for deficit-based wording about the people you serve. Rewrite it
Defensibility. Would you say this sentence out loud in a donor meeting? If not, it does not ship

The usual failure

A plausible figure that appears in no source. It is the most common error and the most expensive one

4DiligenceLog, disclose, own

Never leaves the building

Personal or beneficiary data, in any tool
Unpublished financials or draft budgets
Anything at all in a tool not on the approved list

Say so when

The funder's terms require disclosure of AI use
A colleague will build on the output as if it were checked

Ownership

The person who sends it owns it. AI is not a co-author and carries no liability
Log which tool produced which deliverable. That log is your evidence file

Stuck? Escalation: your AI focal point, then the DPO for anything involving personal data. Sample. Yours is built from your team's tasks and your own approved tool list.

How yours gets made

Four steps, about two days of our time per department.

Task inventory

A short interview with the team lead: what this department actually spends its week on, and which of it is repetitive.

Tested on your files

Every prompt pattern on the page is tried against your real documents before it goes on the page. The ones that fail are cut.

Cleared against policy

The red lines and data rules come from your own AI use policy, so nothing on the page contradicts what your organisation has approved.

A 90-minute lab

The team works through the page on live tasks. They leave having used it once, which is what makes them use it again.

Why one page and not a course. Adding AI tools raises productivity up to about three, then reduces it. Staff carrying heavy AI-oversight loads report measurably more information overload, not less. A catalogue of 900 courses is the same disease in a new bottle. The product here is subtraction: the shortest set of instructions that changes what a team does on Monday.

This is what a team actually gets.