What's actually happening
I have sat in enough of these conversations to recognize the pattern. A director gets a demo from a vendor at a conference, or a peer organization mentions a tool that "changed everything," and the instinct is to move fast. Fast feels like leadership. So the organization buys the platform, or signs up for the pilot, and only then does anyone ask what it's for.
This is the same mistake I see nonprofits make with strategic plans. A forty-page document gets written, approved by the board, and filed.
Nobody translates it into what changes on a Tuesday.
The plan is real. The connection to daily work is not.
AI tools are landing on top of that same disconnect, except now the artifact isn't a binder, it's a subscription.
I'm starting to hear a version of this at the board level too. Funders are asking nonprofits how they use AI, and a board wants an answer that sounds current. That pressure creates the same shortcut: adopt something, anything, so there's an answer to give. A platform bought to satisfy a funder's question is still a platform bought before anyone named the problem it needs to solve inside the organization.
The survey backs this up in a way I did not expect.
Only 28% of these nonprofits name budget as their real constraint on AI. Sixty-seven percent point to a lack of strategic direction.
Money was never the bottleneck. Direction was.
Why willing staff still don't trust the output
Here is the part that should worry a leader more than the adoption numbers do. 79% of leaders in the survey say their staff are eager to use AI. 53% say those same staff don't trust the output enough to act on it.
At first that reads like a contradiction. It isn't.
If nobody told an employee what problem a tool is supposed to solve, that employee has no way to judge whether its answer is right.
Eagerness and distrust can live in the same person at the same time. Willingness to try something is not the same as knowing what "correct" looks like.
Picture a program coordinator using a new tool to draft outreach emails. She's enthusiastic on day one.
By week three, she's rewriting most of what it produces, because nobody told her whether the tool was optimizing for warmth, brevity, or donor conversion.
She can't tell if the draft is good. Neither can her supervisor.
Nobody set the standard before the tool started producing answers.
A nonprofit COO quoted in the same report put it plainly: "we should have spent more time understanding real workflows before building anything." That is not a technology complaint. It is a leadership one.
The pattern shows up long before AI enters the picture. Organizations adopt a tool, a platform, or a plan without first deciding, out loud, what needs to be true on the other side. AI just makes the gap visible faster, because the tool produces an answer within seconds and someone has to decide, immediately, whether to trust it.
A few weeks ago I wrote about naming an owner for the AI tool your organization already has running. This is the step before that one. That piece assumed the tool existed and asked who's accountable for it. This one is about what happens before you ever pick the tool.
The one move that costs nothing
I don't think the fix here requires a framework, a consultant, or a line item. It requires a sentence.
Before your next AI decision, or the next time you sit down to review a tool that's already running, write one sentence that names the problem it is supposed to solve. Not "improve efficiency." Something specific enough that a stranger could read it and know what success looks like: fewer hours spent drafting grant reports, faster response time to intake calls, fewer donor records that go stale after a single gift.
Then name a person, by name, who will say in 30 days whether it worked. Not a committee. One person, one date, one honest answer.
That's it. No budget required. No procurement process. Just the discipline of deciding what the tool owes you before you find out what it gave you.
Does your team know how today's work contributes to your mission?
Most of the leaders in that survey never got that far with their AI decision. They knew the platform. They didn't know the problem.
I'd rather see a nonprofit ask the question first and buy less.