Your AI Task Force Is About to Study a Rollout That Already Happened

I have watched three associations this year stand up an AI task force with roughly the same charter: evaluate staff-facing AI tools, recommend which ones to adopt, build an adoption plan, write the documentation. It is a serious, well-meant charter, and I think it is aimed about eighteen months behind where the organization actually is.

Start with what the sector data says about the starting line. Virtuous, a fundraising-software company, ran a benchmark with Fundraising.AI that surveyed 346 nonprofits in December 2025 and found 92 percent already using AI in some capacity. ASAE's first State of Associations report, out this March, found AI use widespread: 87.5 percent for content and 44.3 percent for data. Whatever your task force decides about adoption, adoption has happened. It happened without a plan, without documentation, and without anybody telling the task force.

The rollout was quiet on purpose

Here is the part that should reshape the charter. A 2026 PagerDuty survey of 1,250 office professionals at companies with $500 million or more in revenue, across the US, UK, Australia, and Japan, found 66 percent had used an AI tool at work believing it was not permitted under company policy. Eighty-eight percent had put work information into a public tool. Seventy-two percent said they believed they understood AI better than their own technology team, and 81 percent believed leadership operates under a different set of AI rules than the rest of the company.

Read those four numbers as one sentence. These are big-company office workers, not association staff, but I suspect the picture in your building is not far off. Most of them are already using these tools, they think they are breaking a rule by doing it, they do not think you are the expert, and they think you have an exemption they do not. That is not a population waiting for a rollout plan. That is a population that has quietly concluded the official channel is slower and less honest than the unofficial one, and has routed around it.

So when the task force sends out its first survey asking what tools staff would find useful, it gets a polite, largely fictional answer. Not because anyone is lying, but because you cannot ask people to volunteer that they have been doing something they believe is against policy, in a form with their name on it, and expect the truth back.

The 81 percent are already inside your building

The same Virtuous research found 65 percent of nonprofits describing their AI use as reactive and individual, 81 percent using it ad hoc, and only 4 percent with documented, repeatable workflows. The 4 percent is easy to read as a maturity ceiling. I would look at the 81 percent instead. That is the one-person pattern: somebody on staff figures out a use, alone, gets good at it, and never writes it down. The knowledge lives with individual staff members instead of becoming part of how the organization operates.

That means the highest-value thing in your building right now is undocumented, owned by one person, and unbacked-up. When the membership coordinator who built a working process for cleaning chapter roster imports leaves in March, that process leaves with her. Nobody will notice for a quarter, because nobody knew it existed. This is a continuity problem before it is an innovation problem, and it is the kind of thing a tool evaluation matrix will never surface.

Run the census before the evaluation

Here is what I would put in front of the task force as its first deliverable instead, and it costs nothing, needs no vendor, and works with your stack exactly as broken as it is today.

Send every staff member one anonymous form with three questions. What did you use an AI tool for in the last two weeks. Which tool. What would you have done instead if you did not have it. Then, above those three questions, put a sentence the ED or COO signs personally: nobody is in trouble for anything they write here, including tools we have not approved and information we would rather had stayed internal. Mean it, and say it in that order, because the amnesty has to arrive before the question or you have just run an audit with a friendly font.

You will know it worked by the volume. If a forty-person staff returns four or five uses, you have not learned that your people are not using AI. You have learned that your framing read as enforcement, and my guess is the two-thirds number above is closer to what is actually going on. Send it again, from a different signature, with the amnesty sentence made blunter. Silence here is a measurement failure, not a finding, and treating it as a finding is how a task force spends six months solving a problem the organization does not have.

What comes back is your real inventory: the actual workflows, the actual tools, the actual exposure. Now the tool evaluation is a comparison against something instead of a guess, the documentation has source material, and the training has a curriculum drawn from what your own people already worked out.

There is one more test worth applying, to the charter rather than the staff. Ask what the task force's first deliverable is. If the answer is a tool evaluation, a vendor shortlist, or a policy document, it is chartered for the wrong decade. If the answer is a list of named workflows with a person attached to each one, it is pointed at the thing that will still matter in a year.

I wrote earlier this summer about the gap between having an AI policy and having governance you can prove, and separately about the distance between 92 percent adoption and 7 percent impact. This is the third side of that shape, and the cheapest. Governance asks what you could defend to an auditor. Impact asks what changed. The census asks the question underneath both, which is simply what is true right now, and it is the only one of the three you can answer before Friday.

Quick takes

The shadow usage is a data-exposure question too. Eighty-eight percent putting work information into public tools is not an abstract risk for an association. My guess is that for many associations some of that work information is member data somebody else trusted you with. The census gives you the scope of that in a week, which is faster than any policy review will get you there, and you cannot scope what you have not counted.

Be careful what the enthusiasm survey measures. If your task force polls staff on which AI tools they want, I suspect the answers skew toward whoever has the time and confidence to have an opinion. That is often not the person doing the most repetitive work. Ask about the last two weeks of actual work instead of about preferences, and the answers come from a different half of the org chart.

Small staff may be ahead of you here. The Virtuous data found organizations under 50 staff reporting moderate impact at 41 percent against 34 percent for those with 50 or more. Note that this measures impact, not usage. The report's own reading is that organizational complexity, not resources, may be the bigger barrier to AI impact. Fewer layers means fewer places for a good idea to stall.

Worth a read

The 2026 Nonprofit AI Adoption Report, from Virtuous and Fundraising.AI. It is a fundraising-software vendor's benchmark of 346 nonprofits, not a random sample of the sector, but it is a useful baseline, and the breakdown between reactive, operational, and strategic use is more useful than the headline.

ASAE's State of Associations report. The first edition, drawn from a year of pulse polls. Worth reading alongside the nonprofit numbers to see how closely associations track the wider sector.

StationX's roundup of shadow AI statistics. A collection rather than original research, but it names and links the organisation behind each number and flags sample limits where they matter, which is more than most roundups do, and it is the fastest way to see how consistent the pattern is across studies.

My prediction is narrow. The first genuinely useful AI workflow your task force documents will not be one it recommended. It will be one it found, already running, built by somebody who assumed they would get in trouble for mentioning it.

---

From the Mind of Ravi Rooprai is a weekly column on association tech, data, and AI. Read the perspectives for the longer arguments behind it.

Researched with AI assistance and fact-checked against primary sources. The analysis, judgment, and writing are mine. How this column is made →