Somewhere in New Jersey right now, a board member is asking an executive director what the organization is doing about artificial intelligence (AI). The honest answer is usually some version of this: We’re still closing last month, the grant report is due Friday and the one person who understands the donor database is out this week.

That gap, between the pressure to do something with AI and the reality of a two-person finance team, is where most nonprofits sit today. It is also the right place to start. The problems AI can solve for a nonprofit are ordinary ones, easier to see once you stop looking at the technology and look at the week.

The Problems Underneath

Four problems come up in nearly every nonprofit we sit down with, whatever the mission:

• Too few people, too many hats: Salaries are tight and talent is scarce, so the accounting manager also runs HR and operations. Anything that eats their hours eats the mission.
• Revenue that resets every fiscal year: Donations, grants, event income and program fees are harder to predict than customer revenue. Planning beyond 12 months can be guesswork.
• Reporting that has to tell a story: The financials go to donors, grantors and the board, not just to a bank. They want dollars tied to programs, so the statement of functional expenses and the Form 990 depend on data coded consistently all year, by someone with three other jobs.
• Thin controls, regular target: Phishing, fake donor inquiries and payment fraud hit nonprofits often. The outsourced IT vendor handles break-fix tickets. Often, no one is watching for this, let alone planning for AI.

Where AI Actually Helps

AI will not repair a broken process. Pointed at one repetitive task, though, it gives hours back:

• Coding and allocations: A finance lead who re-codes transactions into functional categories every quarter can have an assistant, set up with the chart of accounts, propose the coding and review only the exceptions.
• Donor intelligence: Giving history that sits unused in the customer relationship management (CRM) system becomes a short weekly brief pulled from CRM reports and summarized by AI: who gave again, who drifted away, which relationships brought in the largest gifts.
• Reporting that explains itself: Board summaries and grant narratives can be drafted from numbers already in the system, in the organization’s voice. A person still edits every word, but the first draft no longer takes a weekend.
• A second set of eyes: Suspicious “donor” emails and unusual payment requests can be flagged before anyone clicks.

One caution that applies to all four: AI does not know which dollars are restricted unless the ledger does.

The Groundwork Most Organizations Skip

One foundation we are working with is upgrading its accounting system and grant database before layering any AI on top. That order matters. AI amplifies whatever sits underneath it. Clean data gets more useful. Messy data gets confidently wrong, and a consumer AI account on a staff member’s personal login carries none of the security or data controls donor information requires.

A best practice is to perform the following readiness check, in order:

1. Map the data. List every system holding financial, donor or payment information, who owns it and how often it is reconciled.
2. Fix the chart of accounts and cost centers. Functional reporting depends on it and so does any AI you point at the ledger.
3. Set guardrails. Use approved tools only, on organizational accounts where your data is not used to train the model. Decide what never goes into any tool.
4. Look at what you already own. Your accounting system and CRM may already include AI features. Start there before buying anything new.
5. Pick one painful task. Choose something staff do every week and dread. Automate it, count the hours saved and report back.
6. Involve the board. AI decisions are risk and governance decisions, and boards are already asking.

No two nonprofits are alike. A private school, a grief support center and a foundation with rental income need different things, so the plan starts from the mission and works backward. The sequence above holds for all of them.

Every hour AI saves is an hour that goes back to the mission.