Start with the case notes. Where small nonprofits are getting their first real hours back from AI

Ask a small nonprofit where AI should go first and the answer is often something ambitious: a chatbot for the people you serve, a new program, a predictive model. Two stories from the past two weeks point somewhere plainer. The early wins are in the back office, in the notes and the research that sit between staff and the work they came to do.

A case-note tool, shaped by the people who write the notes

On September 10, First Place for Youth, a national organization serving young people aging out of foster care or experiencing homelessness, described Note Ninja, a tool it built through an AI accelerator program with staff involved in the design. Its vice president of programs wrote the account for The Chronicle of Philanthropy.

The idea is simple. After meeting with a young person, a staff member talks or types what happened. The tool returns a summary formatted in the organization’s case-management framework, with key focus areas and next steps.

The pilot was small: eight staff members, three weeks. The results are self-reported by the organization, and they are strong. According to the piece, “100 percent of users reported a decrease in administrative burden,” and 80 percent said it gave them more time to work with young people. In week two, use became voluntary, and all eight kept using it.

Eight people for three weeks is not proof that the tool works everywhere. It is proof of something more useful to a small shop: the pain was shared by the whole team, the tool did one job, and staff chose to keep it. The organization’s own lessons say the same thing. Pick a problem the whole team feels, keep the tool simple enough that people will actually use it, and ask funders to support that kind of capacity work.

A donor dossier in minutes, at museum scale

On September 23, the Chronicle reported on museums using AI. The Detroit Institute of Arts is building an in-house system so fundraisers no longer have to search four or more systems (fundraising records, artwork donations, ticketing and email engagement) to understand one donor’s full relationship with the museum.

Its chief digital officer, Jennifer Snyder, described the goal as a “succinct one- to three-page dossier” produced in 10 to 15 minutes instead of hours or days. The three AI projects together will cost about $500,000, with the database connector alone estimated at $300,000.

Few small nonprofits have $500,000 for anything. But the shape of the project carries over: gather what you already know about a donor into one page, have a person read it before the meeting, and spend the saved hours on the relationship itself. For a five-person organization, “four systems” might be a spreadsheet, an email inbox, a donation platform and someone’s memory. The same page is within reach.

The same report notes that the Museum of Modern Art lets staff use AI tools under an internal risk framework, and the National Gallery of Art runs staff training on safe use. In one, a risk framework sits beside the tools. In the other, staff training does.

A five-question test for your first project

Before you pick a first AI project, score each candidate. Give it one point per yes.

1. Does the whole team feel it? A task that burdens everyone gets used; a task one person dislikes gets forgotten.
2. Does it repeat every week? Weekly work pays back quickly and gives you enough examples to judge the output.
3. Can a person check the result in under five minutes? Case-note summaries, thank-you drafts and donor one-pagers are easy to review. A finished grant budget is not.
4. Can it run without sending sensitive records to a consumer AI tool? If the task needs client or donor data, it needs a business account with data controls and a written rule about access. If you cannot arrange that yet, choose a different task.
5. Would you know if it failed? Write down how long the task takes today, so you can compare.

Start with anything that scores five. A four is fine if the missing point is the second question. Anything lower waits.

This is general education, not legal or privacy advice. Rules for client and donor data vary by place and by funder, so check yours with a qualified adviser.

What to do this week

• List the five tasks that eat the most staff hours, and score each with the test above.
• For the top one, time how long it takes now, across two or three people.
• Write one paragraph on what data the task touches and where that data is allowed to go.
• Choose two or three staff to try it for three weeks, and agree at the start how you will decide whether to keep it.

The organizations in these stories did not start with the mission. They started with the paperwork around it, and gave the hours back.