Grantmaking in the AI age: keeping humans in the loop is harder than you think

A funder uses AI to help design a grant programme. An applicant uses AI to write an application. An assessor uses AI to score it. Somewhere out there, machines are having a conversation about a community project, and the humans are hovering but not truly present.

At SmartyGrants, we've spent more than 20 years watching grantmakers work, and we’ve built what we’ve learnt into our software. We’ve had great success creating efficiencies and improving grantmaking effectiveness. Now decision-making itself is in the frame.

Recently our team reviewed an in-house prototype of an AI assessment tool, a descendant of an experiment (“Tessa the Assessor”) we began in 2019. The bot worked exactly as it was supposed to, handling eligibility assessment well. I have no doubt that with instruction, it would do the same for merit assessment.

Yet we didn't deploy it.

The problem was that there was no honest way to keep a human in the loop. A time-pressed grants officer casting an eye over machine-generated scores and clicking “confirm” is not a human process, whatever the policy says.

If you don’t know what the bot missed, what it hallucinated, or what biases clouded its judgement, you are not really in the loop. Checking the output properly means, in effect, redoing the assessment, and if you're redoing it there’s no point automating it.

We went back to the drawing board with a new discipline built into the design process. This is a test I would put to any funder weighing up an AI tool: picture the moment the decision gets challenged. An unsuccessful applicant appeals, or a journalist asks why that town got the money and this one didn't. The Public Accounts Committee or the board asks: who decided, and on what basis?

Could the person who’s accountable answer in their own words?

“The system scored it 62” isn't an acceptable answer.

Underpinning our development of AI tools for SmartyGrants is what our chief data scientist, Dr Paola Oliva-Altamirano, calls our “ethical muscle”. Anything we deploy must pass our “dignity lens” review, developed with AI ethics researcher Dr Lorenn Ruster, which asks:

• Are people’s contributions and ideas validated, rather than reduced to a score?
• Do people stay in control of decisions?
• Does it feel fair to the people on the receiving end?
• What is its environmental cost?
• Whose work does it automate or displace, and whose labour built it?

With all of that in mind our new tool is designed to help grant assessors think, rather than thinking for them. That means showing them where the relevant evidence sits (in the application, in past reports, or on a government website, for example), reminding them of the assessment criteria that have been agreed and published, and flagging inconsistencies and gaps in the applicant’s argument. The judgement stays with the person who has to defend it.

We expect that grantseekers will keep using AI to draft their applications. That doesn’t mean they’ll always submit slop; in fact, in some cases, AI lifts the standard. When AI enables all applicants to express themselves clearly, other criteria take on greater significance: a grantseeker’s values and track record, verified outcomes data, and community relationships and partnerships.

If AI leaves all applications equally polished, the qualities that differentiate them are what funders should have been judging all along. And that’s a good thing.

This article, written by Kathy Richardson, Executive Director of SmartyGrants, was originally published on October 8, 2026 in Philanthropy Ireland Magazine.