Why isn't our nonprofit's AI actually producing results?

Most nonprofit AI projects never start with a decision. They start with a tool.

A new report on AI operations at nonprofits makes that specific. Among nonprofits that already have at least one AI initiative running in production, just 12% say their most recent AI initiative began with a clearly defined problem. The rest started elsewhere: 37% adopted a vendor's recommended use case, another 37% chose a platform first, and 13% never settled on a scope at all (Coastal / Oxford Economics, "The AI Operations Report 2026: Nonprofit," accessed September 23, 2026).

That's not a story about slow adoption. Every organization in the sample already has AI live somewhere. It's a story about what happens when "live" arrives before "why."

A note on the source before we go further: the report's page shows only a 2026 copyright, no month or day for when it was published or fielded. That's worth saying plainly rather than dressing it up with a date I don't actually have. Cite it as I found it, on September 23, 2026.

How do most nonprofits actually start an AI project?

Usually with a demo, not a decision. Someone on staff sees a vendor pitch, or a board member forwards an article about a competitor's chatbot, and within a few weeks there's a contract. Nobody wrote down what problem the tool was supposed to solve, because nobody asked that question out loud before signing.

I've seen this pattern up close. It rarely looks reckless in the room. It looks efficient. A platform gets chosen because procurement is slow and this particular vendor already passed security review.

A use case gets adopted because the vendor's own customer stories made it sound proven elsewhere. Each step is reasonable on its own. The sum of those steps is an organization running AI it never actually defined a purpose for.

Why does starting with a tool instead of a problem matter?

Because it removes the one thing that would tell you, later, whether it worked. Start with a problem, and "did this work" already has an answer waiting: fewer missed grant deadlines, faster responses to donor inquiries, less staff time spent on the same repeated task. Start with a tool, and you only find out what success would have looked like after the fact, if you ever bother to ask.

That gap shows up directly in the same report. Despite every organization in the sample already running AI in production, only 19% can point to a measurable result. Not 19% saw a small improvement and are calling it a win. 19% could actually name one.

Who ends up owning AI when nobody assigns it?

IT does, by default. 61% of the nonprofits surveyed place AI under IT leadership. Only 16% put it under a program or operations lead, and 11% under a dedicated AI team.

That default makes a certain sense on paper. IT already owns the login, the vendor contract, the security review. What it doesn't own is the mission the tool was supposed to serve.

A program director knows what a missed grant deadline costs an organization. An IT lead knows what an expired API key costs. Those are not the same accountability, and right now one department is quietly holding both.

This is a distinction I keep coming back to with human teams, and AI inherits the exact same risk. More activity is not the same as closer to the goal. A person, or a piece of software, can be busy all day without moving the organization's actual priorities forward. Nobody notices, because nobody outside IT was ever asked to answer for it.

What should a director do before greenlighting the next AI pilot?

Write one sentence before you sign anything: what result, measured how, by whom, by when. Not a strategy document. One sentence.

Then name a person outside IT who owns that sentence. Usually the program lead whose work the tool is meant to touch, not the person managing the vendor relationship. If nobody outside IT can explain, in one sentence, what a given AI tool is supposed to change, that's the actual finding, before a single feature ever gets evaluated.

This is close to the discipline behind the work I do now. DeTask turns your mission into goals and keeps your leaders, people, and AI aligned around the results that matter. The AI-ownership question above is really the same question in a smaller frame: does anyone know how this piece of work, human or automated, contributes to what the organization is actually trying to accomplish?

The next AI pitch that lands on your desk will describe what the software does. Before you approve it, ask who will be able to tell you, six months from now, whether it worked, and whether that person has anything to do with IT.

FAQ

What percentage of nonprofits start AI projects with a clearly defined problem?

Just 12%, according to Coastal and Oxford Economics' 2026 nonprofit AI report. The rest split three ways: 37% adopted a vendor's recommended use case, another 37% chose a platform first, and 13% never settled on a scope at all.

Who typically owns AI at a nonprofit?

IT leadership, in 61% of cases, according to the same report. Program or operations leads own it in 16% of cases, and a dedicated AI team in 11%.

Does starting with a defined problem guarantee measurable results?

No single report proves that causally. But the same survey found only 19% of nonprofits already running AI in production can point to a measurable result, alongside the finding that most did not start with a defined problem. It's a pattern worth taking seriously in your own organization, even without formal proof of cause and effect.

What's the first question a director should ask before approving an AI pilot?

What result, measured how, and who outside IT is accountable for it. If nobody can answer that in one sentence, that's the finding.