AI in Nonprofits: 4 decisions before you start
For a long stretch I was the search function. Volunteers across a dozen countries would message me with questions our wiki already answered. I would find the page, paste the link, and do it again the following week for somebody else.
So we put a grounded assistant on top of the wiki. It answered from our own pages and nothing else. The repeat questions dropped, and the interesting part was not the tool. It was everything we had to decide before we switched it on.
Every AI question in a volunteer organisation turns into a people question within about ten minutes.
The gap is not the tool
Most nonprofits are already using AI, whether or not anyone has decided to. Volunteers are busy, they want their work done faster, and they are curious about the tools, so a task gets handed to a chatbot and nobody hears about it. The question is not whether your organisation uses AI. It is whether leadership decides what it gets used for, or whether the people doing the work decide that for you by default. If it has been the second one so far, guard rails are much harder to add later. By then there are habits to undo, and undoing a habit takes longer than writing a rule. For a European NGO the stakes are higher again, because some of your data should never leave Europe, and a document cannot be un-pasted once it has gone into a tool.
The four decisions worth making first
This is roughly the order I would take them in.
1️⃣ Name three tasks, and name them narrowly. Not "we use AI for comms". Something like first-draft translation of a chapter newsletter, tidying meeting notes, and answering repeat questions from our own wiki. Three named tasks give people permission without giving them a blank cheque. A blank cheque is what makes leadership nervous, and rightly so.
2️⃣ Decide what counts as ground truth. An assistant is only as honest as the pages behind it. Ours was useful because the wiki was current. The moment those pages go stale, the tool starts confidently telling new volunteers how things used to work. So somebody owns those pages, by name, with a date in the calendar. That is the whole maintenance cost and it is not optional.
3️⃣ Always leave a human exit. When the assistant cannot answer, it hands the person to a named human, not to a dead end. A stuck volunteer who hits a wall does not try again so the chat bot should direct them to a person when in doubt.
4️⃣ Write one page of rules, in plain language. What never goes into a chatbot: member data, anything from a disciplinary matter, anything a person told you in confidence. What always gets a human read before it goes out: anything with the organisation's name on it. One page, so a new volunteer in week two knows where the line sits without asking.
Notice that none of the four is technical. That is the point. The literacy gap people describe as a skills problem is usually a governance problem wearing a hoodie.
One thing I would not do. I would not run a general AI training for volunteers before those four decisions exist. Training people in a tool they have no permission to use produces enthusiasm and then nothing, and you have spent the goodwill you needed for the real rollout.
Where this helped me
I built the HR team of Volt Europa from a one man show to 30 people across multiple specialised teams. You can read more about it here.
Your turn
Has anyone in your organisation written down what volunteers may and may not put into an AI tool? If not, who would you expect to write it? Leave a comment.
Ermal Asllani
People Ops for Decentralised Organisations
Behind the Mission goes out every second week, on the people side of running a decentralised organisation.