Deciding your nonprofit's AI position

Ask everyone at your organization where it should be with AI tools in two years and you may get four answers in roughly equal proportions. Some people want to hold off. Some want to try a few careful things. Some want to use them wherever they obviously help. A few want to move fast. Nobody has a majority, which means deciding your nonprofit's AI position has not happened yet, whatever anyone has said in a meeting.

A split like that is not a failure and it is not indecision on anyone's part. It is the ordinary state of an organization that has never been asked the question directly. What makes it a problem is that the vacuum gets filled anyway: by whoever is most confident, by whatever a vendor switched on by default, or by the staff member who found something useful and kept going because nobody said not to.

What a position actually is

Not a value statement. A position is a sentence that changes what someone does on a Tuesday, plus the rules that follow from it.

Four positions are worth naming, because they are genuinely different and organizations tend to blur them.

Hold off for now. We are not using these tools for our work at present. This is a real position with real reasoning behind it, and it is treated as one in deciding not to use AI.

Try them carefully in a few low-risk areas. Named tasks, named limits, a person checking the output, and a review date. The most common landing spot for a small organization.

Use them where they clearly help. A general permission with a short exclusion list, rather than a short permission list.

Build them into how we work. A deliberate change to processes, which implies budget, training, and somebody whose job includes it.

The test for whether you have chosen one is simple. Can a program manager read it and know whether she can put her draft newsletter into a tool tomorrow. If not, what you have is a sentiment.

Deciding your nonprofit's AI position, step by step

Get the split in front of the board in writing. Not one person's summary of the mood. The actual distribution, board and staff shown separately, because the two groups often differ and a blended number hides it. If they do differ, that gap needs its own conversation first, which is the subject of when the board and staff disagree.

Do the education hour first if the board needs it. A board that is voting on something half of it does not understand will produce a position that does not survive contact with the first real question.

Ask what people are actually worried about, one at a time, before anyone proposes anything. Go around the table. The point is to get the concerns named rather than argued with. Boards that skip this get a vote that closes the topic without settling it, and then the same discussion reopens in six months as an operational complaint.

Put the four positions on the wall and ask people to place themselves. Then ask the people at each end what would have to be true for them to move one step. That question does more than any amount of debate, because it converts a preference into a condition, and conditions can be checked.

Try for consensus, take a vote if you do not get it, and record the split. A 6 to 3 vote is a legitimate decision and a useful record. Pretending it was unanimous is how a position quietly stops binding anyone.

Give it a review date before you adopt it. Six or twelve months. A position with a review date is much easier to agree to, because nobody is voting on forever.

What to write down

Five things, in the minutes and in a document somebody can find later.

The position, in one sentence. Plain enough to repeat out loud.

What it means in practice. Two short lists: what people may do without asking, and what is off the table. The second list matters more.

What information never goes into one of these tools. This is a separate decision from the direction, it applies whatever the position is, and it should already be made. If it is not, make it first, on the reasoning in what never goes into an AI tool.

Who decides the details. The board sets the boundary. Somebody named on staff decides which tools and how, within it.

When you look at it again, and what would change it sooner. Invented example: a community health nonprofit adopted a careful-in-a-few-areas position with three named triggers for an earlier review, one of which was a funder asking about their use of AI. That happened eleven months later, and the board had already agreed what it meant, which turned an awkward email into a scheduled agenda item.

Then the document has to reach people, which is a separate job from writing it, and the failure mode is well documented: a written policy that most of the staff cannot describe. The fixes are in the AI policy nobody knows about.

Two ways this goes wrong

The position that says nothing. We will use AI responsibly and ethically, in line with our values. Everyone can vote for it, it forbids nothing, and it leaves every real question exactly where it was. If your draft would be agreed to by all four camps without complaint, it is not a position.

The position adopted to end the discussion. Sometimes a board settles on something in the last five minutes of a long meeting because everyone wants to go home. That decision has no support behind it and will not be defended by anybody the first time it is inconvenient. If you are out of time, schedule it properly rather than resolve it cheaply.

Boards that handle this well tend to be boards that already have a habit of deciding things rather than receiving reports, and the general version of that argument is in what to do with assessment results once you have them. For broader research on how nonprofit boards make and record decisions of this kind, BoardSource is the standard reference in the sector.

If you want the split measured rather than estimated, the AI readiness assessment asks board members and staff the same question about direction and shows the two distributions side by side, along with where each group thinks these tools should and should not be used. It is free, it takes about seven minutes, and nothing in it can be traced back to an individual.

The organizations that get into trouble here are not the ones that chose the wrong position. They are the ones that never chose, and then found out what their position was by reading it in an email a staff member sent to a funder.