Your team already adopted AI. Your organization hasn't.

TLDR: Somebody on your team is already running part of their job through AI, your organization has no idea what that adds up to, and the best data anyone has collected says that gap is not closing on its own.

Somewhere in your organization right now, someone is pasting a grant narrative into ChatGPT. You didn't approve it, there's no policy that covers it, and honestly, it's probably the only reason their Tuesday is survivable.

The humanitarian sector has a name for this, shadow AI, and it has the best data anyone has collected on it. The Humanitarian Leadership Academy and Data Friendly Space surveyed 2,539 aid workers across 144 countries in 2025, then went back in January 2026 and asked 1,729 more. Two waves, eight months apart, and the pair of them tells a sharper story than either one alone.

In the January 2026 wave, 75 percent of respondents use AI tools daily or weekly, up from 70. The share who believe AI has improved their operational efficiency jumped almost 18 points in eight months, to 64.6 percent. So the people got more convinced, and more capable, and more embedded.

And their organizations? 9 percent report AI as widely integrated, up from 8. Formal AI policy coverage went from 22 percent to 23.3. In eight months, the individuals moved 18 points and the institutions moved one.

Before you quote any of this at a board meeting, here is what this survey is not. It did not sample nonprofits at random, and its authors say so, in a labeled limitations section: distribution ran through the publishers' own networks and training platform, which "likely resulted in respondents with higher digital literacy and more positive technology attitudes than the broader humanitarian workforce." Respondents opted in. Nearly half work in Sub-Saharan Africa. So no, 93 percent is not your staff's adoption rate, and I won't pretend it is. The honest reading is kind of worse: this is the most AI-friendly sample anyone has assembled in this sector, and even here, the organizations built almost nothing, twice, eight months apart. Enthusiasm bias inflates the individual number. It makes the organizational number a ceiling. (One more hygiene note, since numbers drift: you'll see 21 percent quoted for policy coverage, which is a senior-leader subgroup figure that a press release generalized. The whole-sample number is 22, now 23.3.)

Why should a nonprofit ED in the US care about a humanitarian survey? Because the mechanism transfers even where the percentages don't. Your people are the classic three people in one trench coat, doing three jobs each for below-market pay, and AI is the first tool in a decade that actually relieves some of that load without a budget line. Of course they adopted it on their own. The question the survey puts in front of you is the one it answers for its own sector: what did your organization build around that adoption? A policy, so nobody is improvising rules on donor and client data? A shared way of working, so the gains compound instead of living in one person's browser tab? For most organizations the answer is nothing, and the risk side is moving: 45 percent of respondents say no policy exists at all, and 29.8 percent are now running AI agents, tools that take actions rather than draft text, a category that didn't even register in the 2025 wave.

Full disclosure, because you should read the next paragraph knowing it: Earned Impact sells AI implementation services to nonprofits, so I have an interest in you deciding this gap matters. Which is exactly why I'll also tell you what a vendor deck would leave out. In the 2025 wave, fewer than half of respondents thought AI had improved their efficiency, only 3 percent consider themselves expert users, and none of this data describes US nonprofits. I think the argument survives those three numbers. You should check.

Closing the gap is not a moonshot, it's an implementation project, the kind that costs less than one employee in your market and gives you back capacity across the team you already have. Nobody loses a job in this story, that's the point: your people stop doing three jobs badly and start doing one job with leverage.

So here's the offer, agenda disclosed up front. We're doing free AI consults for nonprofits: an hour on where your organization actually stands, what your staff are already doing in the shadows, and what the first two or three implementation moves would be. No deck, no pitch, and if what you need is a policy template and a training afternoon, I'll tell you that and we're done. We're looking for the organizations where the gap is big enough that closing it changes their year.

PS: If you read one primary source this month, make it the January 2026 pulse briefing note. It's eleven pages, it's honest about its own limits, and it will change how you hear every AI adoption statistic somebody puts on a slide.