The 7% Problem: Why Nonprofit AI Stalls at "Faster Drafts"

A benchmark study of 346 nonprofits from Virtuous and Fundraising.AI, released in early 2026, found that 92 percent of them use AI in some form. Only 7 percent said it had meaningfully expanded what their team could do. Everyone else landed somewhere in the middle: faster drafts, quicker emails, the same results they were already getting.

The report calls this the efficiency plateau. I think it is the most important number in the sector right now, and almost nobody is talking about why it happens.

Adoption is not capability

Here is the short version. AI did not fail these organizations. It did exactly what they asked. They pointed it at the work they already do and it made that work faster. But speed is not capability. A grant writer who drafts twice as fast is still doing the same grant writing, in the same process, with the same bottlenecks downstream. You sped up one step in a system you never redesigned, and the system sets the ceiling.

The same study found that 81 percent of nonprofits use AI individually and ad hoc. One person opens a chatbot, gets a result, and pastes it into a process that has no idea the AI was involved. There is no shared workflow and no documented way of doing it that survives that person leaving. Nobody measured whether it helped. Nearly half the organizations have no AI governance policy at all. So the picture is not a sector that adopted a powerful tool. It is a sector where a lot of individuals quietly adopted a tool, on their own, with no structure around it.

That is why "use more AI" is bad advice. The 7 percent that broke through did not use more. They changed how the work gets done so the AI had something real to plug into. That is organizational work, not technical work, and it is slower and less fun than trying a new model. It is also the only thing that moves the number.

What the gap actually looks like

"Redesign your workflows" is the kind of phrase that means nothing until you make it concrete, so take intake at a service organization.

The efficiency version: a caseworker uses AI to write up their notes faster after each session. Real, useful, and it changes nothing structurally. The capability version: you look at the whole intake process, decide what a human has to do and what a system can do safely, build a shared way of capturing and routing information that every caseworker uses the same way, put a check on it so you know when it is wrong, and write down how it works so it does not live in one person's head. The first is a shortcut. The second took design, agreement, and governance to build, and only the second is a new capability.

Most organizations stop at the first one because the second one is hard and there is no obvious owner for it. The technology is the easy part now. The hard part is the same thing it has always been in this sector: getting a stretched team to agree on a process and then follow it.

The quiet liability in high-stakes work

There is a second reason the plateau matters, and it is the one I care about most. In high-stakes services, ad hoc and ungoverned AI is not just a missed opportunity. It is a risk surface. When a tool with no oversight is sitting next to sensitive information about people in crisis, the absence of structure is the danger. Nobody decided what data goes into it. Nobody decided what happens when it is wrong. Nobody is watching. In a fundraising shop that produces a worse email. In a shelter or a crisis program it can produce a worse outcome for someone whose week is already going badly. The plateau is usually framed as wasted potential. In this work it is also a quiet liability.

Funders have started to notice the structure question. They are asking how organizations use data, how they measure outcomes, and how they plan to sustain any of it. "We let staff use ChatGPT" is not an answer to that, and the organizations that can describe an actual system, with governance and evaluation attached, are going to look very different in a few years from the ones that cannot.

The real test

So if you are a leader staring at the 7 percent number and wondering what side of it you are on, the test is not how much AI your team uses. The test is whether you could describe, in writing, one process your organization changed because of it, including who owns it, how you know when it fails, and what happens then. If you can, you are doing the real work. If every honest answer is "people use it on their own to go faster," you are on the plateau, and no new tool is going to get you off it.

The work was never the tool. It was the redesign around the tool. That part is unglamorous and organizational, and it is where the whole thing is won or lost.