Start with the mission
Your AI strategy is a written answer to one question. Where in your programs is staff time the bottleneck? And which of those spots can a current model relieve without costing you trust, safety or accuracy?
Skip that question and you get a ChatGPT seat for everyone, a vendor demo every Friday, and no change in programs delivered.
The one-page plan
A strategy names five things on one page:
- The workflow you are targeting
- The data allowed to touch the model
- Who reviews the output before it goes out
- The metric that proves it worked
- The date you look at it again
The test: if a board member asked what your org does with AI, could someone on staff answer in three sentences without naming a product?
Five steps
Inventory
List every recurring task by program. Note weekly volume, hours, and how much judgment it needs.
Score
Rate each task high or low on payback (volume times hours saved) and risk (client-facing, regulated, needs checking).
Pilot
Pick one to three high-payback, low-risk tasks. Run a four to six week pilot with before and after numbers.
Govern
Before the pilot ships, write a one-page AI use policy: who uses what, with what data, reviewed by whom, and what gets escalated.
Scale
Only after real hours are saved, repeat on the next two tasks. Review every quarter; the tools will change.
Measuring payback honestly
AI rarely saves a nonprofit money. Staff work the same hours. The win is moving hours from drudge work back to programs. Before the pilot, write down:
- Hours a week the task takes across the team today
- How much of that AI realistically removes. A demo is not a baseline.
- What the freed hours will do: more people served, faster grant reports, more donor contact
If you can't point to that outcome at 90 days, the pilot did not work, however good the demo looked.
An example of freed hours going back to the mission: Wyoming REALTORS cut a member sync from 50+ hours to 15 minutes, and outreach can now target active and inactive members separately. That one was plain automation, no AI.
Guardrails
- Never put identifiable client data into a consumer AI tool. Use a business plan with a data-processing agreement, or a self-hosted or EU-hosted option.
- Tell people when AI touches the services they receive: intake summaries, triage, written replies.
- Keep a person on every decision about funding, eligibility or care. AI drafts. People decide.
- Check outputs for bias every quarter, especially in triage and hiring.
- Have staff sign the AI use policy, and review it every six months.
Before you pick a tool, check what it keeps and trains on. Where Does Your Data Go? (opens in a new tab) is a free 16-page guide with the setting to change in eight common AI tools.
What goes in an AI use policy
A nonprofit AI use policy fits on one page. Staff should be able to read it in five minutes and know what they can do on Monday. Cover these:
- Approved tools, by name and plan. Anything else needs sign-off.
- Data rules: what never goes in (client records, donor financial data, health, immigration status) and what can
- Review: who checks AI output before it reaches a client, donor or funder
- Disclosure: when you tell people AI was used, and the sentence you use
- Decisions AI never makes: eligibility, funding, hiring, care
- Who to tell when something goes wrong, and how fast
- Owner and review date
Red Mage's own AI policy is public if you want a worked example. Adapt it; don't copy a vendor's.
Nonprofit programs from AI vendors
Checked 17 September 2026. Terms change, so confirm on the program page before you budget.
| Program | What eligible nonprofits get |
|---|---|
| OpenAI for Nonprofits (opens in a new tab) | ChatGPT Business seats at about 20% off ($8 a user a month, billed yearly), and larger Enterprise discounts through sales |
| Claude for Nonprofits (opens in a new tab) | Up to 75% off Claude Team and Enterprise plans |
| Google for Nonprofits (opens in a new tab) | Google Workspace for Nonprofits at no cost, with discounts on Business Standard, Business Plus and Enterprise |
| Microsoft for Nonprofits (opens in a new tab) | Business Basic free for up to 300 users, Business Premium at $5.50 a user a month, and Microsoft 365 Copilot at $25.50 a user a month (all billed yearly) |
| TechSoup (opens in a new tab) | Eligibility checks and software discounts, often for an admin fee |
Pick the platform that matches the rest of your stack. A deep discount on a tool nobody opens is wasted.
Where to start this quarter
Do the inventory and scoring this week. Draft the policy. Pick one high-payback, low-risk task and pilot it before the quarter ends. Intake triage, donor thank-you drafts and grant report extraction are the usual picks.
For the board conversation, the 50 Questions Before Your Nonprofit Uses AI (opens in a new tab) checklist turns each step into questions you can answer on paper before anyone buys a seat.
Common questions
- Which model is best for us?
- The one that fits your stack. If you live in Google Workspace, use Gemini. In Microsoft 365, use Copilot. Starting fresh, I reach for Claude first and ChatGPT second. Pick one, give it 90 days, and switch only if a specific workflow needs it.
- Do we need an AI committee?
- Probably not. A short AI use policy, signed by staff and reviewed twice a year by leadership, covers most small and mid-size nonprofits. Committees earn their keep at around fifty staff, or with regulated data in health, immigration or child welfare.
- Should we self-host an open-source model?
- Only with a real reason: sensitive client data, EU residency rules, or a board concern about vendor lock-in. For most small teams, a discounted business plan is cheaper, faster and safer than running your own servers.
By Andrés Sepúlveda Morales. Top Independent on Contra, rated 4.99 (opens in a new tab) by clients and collaborators, and a speaker on AI in English and Spanish.