An AI strategy for your not-for-profit in 2027: what to budget, and in what order

Every November I used to sit with a spreadsheet and a board that wanted to know what technology would cost next year and what it would give back. For nineteen years I was the chief executive of a registered charity and, because we never had the money for a technology department, I was also the person who built everything it ran on. That gave me a view of this question that most advisers do not have: I had to find the money, justify it to funders, and then make the thing actually work with a team that was already stretched.

This year the question has an AI in it. Funders are starting to put money specifically into not-for-profit AI adoption, boards everywhere are asking their chief executives what the plan is, and most of the sector has staff already using Copilot or ChatGPT with no plan at all. I wrote in August about what the EU AI Act means for mental health services in Ireland; this piece is the practical sequel for any not-for-profit, whatever the mission: homelessness, disability, youth work, community development, the environment, education, older people, migrants, the arts.

Here is the order I would spend the money in, and roughly what each step costs.

Step one: find out where the hours go. Two weeks.

Do not start with a tool. Start with a fortnight of looking at what your people actually do. In every organisation I have worked in, the same four things eat the week: the front door (the phone, the email inbox, the enquiry form, answered by a human who could be doing something harder), the paperwork around every appointment or visit, the notes that get written and never become data, and the search for the right procedure or policy when a new staff member needs it. None of these is specific to mental health. A homelessness service has an intake process. A disability organisation writes support-plan notes. A community development group reports to three funders in three formats.

The output of this step is a short, costed list of two or three projects, ranked by return and risk, and a one-page paper your board can approve. It is also the document that gets you through a funder’s door: every grant application asks you to name a problem you have already identified, and “our coordinator spends nine hours a week on intake” is a problem. “We want to use AI” is not.

Step two: train the people before you buy the tool. Half a day.

This is the cheapest step and the one most organisations skip. If your staff already use Copilot or ChatGPT at work, and most do, you have had a legal duty since February 2025 under Article 4 of the EU AI Act to take measures so they have an appropriate level of AI literacy. The duty has no size exemption. The wording was revised in July but kept, and national supervision began in August.

More to the point, trained staff are the ones who catch the tool’s mistakes. A half-day on your own real tasks, the monthly report, the twenty-times-a-week email, the policy nobody can find, with a one-page set of house rules adopted the same week, prevents most of the data-protection problems that follow. Budget it first.

Step three: write down what the machine may see. One week.

Most everyday AI risk in a charity is a well-meaning staff member pasting a service user’s details into a public tool. A written decision about which categories of information can go into which systems, pseudonymisation before anything sensitive goes near a model, a human sign-off on anything that affects a client, and a risk register the board actually reads: that is responsible AI in practice. It is one week of work, and it is what the Charities Regulator, the HSE, Tusla or HIQA will ask for first if they ever look.

Step four: build one thing, in your own environment, with a number attached. Six to eight weeks.

Now buy, or build. Pick the project from step one that a named person on your team feels the pain of daily. Build it on the Microsoft 365 or Google Workspace you already pay for, not on a new platform. Make it answer only from your own documents and show where its answers came from. Train the people who will use it. Then run it for a month with someone watching, and write down the before and the after. That number is what you take back to the board, and to the funder, in twelve months.

The three places I would not start: anything client-facing that makes a decision on its own, replacing a system that works, and a chatbot on the website for its own sake.

What it costs, and what it saves

Done in that order, a full programme for a small or mid-sized not-for-profit comes to the price of about a third of one salary for the year, and the first project typically gives back more than that in staff hours. Done in the wrong order, with the licence bought first, the money is usually gone by March and nobody is using the tool by June.

For 2027 I am offering this sequence to the Irish not-for-profit sector as a single programme at a flat charity day rate, with each step also available on its own and help with the funding application included. The details are on the offer page. If you would rather start with a conversation, that costs nothing: thirty minutes, one question, where does your time go?

Krystian Fikert founded and led MyMind for nineteen years and now works with not-for-profits on technology, AI, governance and operations through Fikert Limited. He is an Ashoka Fellow and INSEAD Social Entrepreneur in Residence. Book a call or email krystian@fikert.com.