The Real Cost of Not Using AI to get Grants

There's a narrative gaining traction in the fundraising sector that goes something like this: “Grant-seeking must remain exclusively human. Be cautious. Don't upload your data. Don't trust the machines with anything important.” Caution is reasonable.

But when you follow this to its logical conclusion, the people it hurts most are the small, under-resourced community organisations who can least afford it. When used well, AI has huge potential to bring positive change, and to realise that, we need an accurate understanding.

The cost of doing it by hand

Let's talk about what "human-only" grant writing actually looks like for a volunteer-run organisation. A typical grant application takes 10-20 hours to research and write. Many community organisations apply for 10-30 grants a year. That's somewhere between 100 and 600 hours annually, often done by volunteers, often on top of the mahi the organisation actually exists to do. A significant proportion of those applications won’t succeed—they are for funding the organisation was never going to get. Wrong eligibility criteria. Wrong funding priorities. Hours spent crafting a beautiful application for a grant that was never a match. That's wasted effort that could have gone into the community.

The opportunities you never find

The flip side of wasted applications is missed ones. There are around 3,000 contestable grant opportunities available in Aotearoa. No human, no matter how experienced, can track all of them, know when rounds open, and assess alignment for every organisation they work with. When we talk to community organisations who use Fundsorter, we regularly hear, "We had no idea that fund existed." Right now, organisations are missing valid, well-aligned opportunities because the landscape is too large and too fragmented to navigate manually.

The filter-and-list problem

With a traditional grants database, the process is painful. You do your best to understand and set up some of the limited filters, hit search, and get a long list of potential funders. Then the real work starts: open each one, read pages of criteria line by line, and try to figure out whether your organisation is actually a good fit. Multiply that by dozens of results, and you've burned hours before you've written a single word. To navigate this successfully, you need pre-existing expertise about funding, which is just not the reality of most volunteer-run community groups.

This is what Fundsorter fixes. When you run a scan, the AI reads the documentation for you. It checks eligibility. It assesses fit. It tells you why a fund is or isn't a good match for your organisation. Instead of a long list and a weekend of reading, you get a curated list of genuinely aligned opportunities. Human judgment still matters, but it's applied where it counts, not spent on basic triage.

The hallucination problem is a design problem

One common concern is that AI fabricates information. Large language models can and do make things up, confidently and convincingly. If you paste a question into ChatGPT and take the answer at face value, you will eventually get burned. But the answer to hallucination isn't to avoid AI, it's to use well-design systems, that put humans in the loop.

Fundsorter links every recommendation back to the source and explains why it was surfaced. Every draft application can be checked against the original funding criteria. Unlike generic ChatGPT, everything Fundsorter generates is in relation to rich data you’ve provided about your specific organisation, and rich data about the unique funding landscape of Aotearoa that real, local humans have thoroughly checked, drastically reducing inaccuracies. Responsible AI means making it simple for people to verify, question, and override. This is a design choice, which comes from a deep understanding of how this technology functions. The question organisations should be asking isn't "Does AI hallucinate?" It's "Does this tool make it easy for me to check?".

Your data is yours

We've seen advice urging organisations not to upload their information to external platforms, warning that their data will be used to train AI models. This is a reasonable concern, and organisations should absolutely ask the question. At Fundsorter, the answer is straightforward: we do not train AI on user data. Your organisational information is used to match you with funding opportunities and to help draft your applications, not to improve outputs for anyone else. Further, unlike using free ChatGPT, Fundsorter uses AI through an enterprise API agreement that prohibits the LLM company from using your data for training either. On top of this, to be extra safe, Fundsorter anonymises your data before sending it to the LLM for processing, meaning you can access the intelligence without having your organisational details exposed. Your data stays yours. Not every platform makes this commitment. The existence of bad data practices in some tools is not a reason to reject the entire category.

Who benefits from caution?

It's worth asking who the "go slow" message serves. Organisations with professional fundraisers on staff and the budget to hire grant-writing consultants can afford to stick to the old ways of operating. They already have the resources to navigate the system. It also serves the legacy platforms whose business models were built before AI changed what's possible. If your product is a database with keyword filters and your service is manual grant writing, then AI-powered tools that do both of those things faster, cheaper, and more accurately are an existential problem.

The natural response is to urge caution, to emphasise the risks, to argue that the old way is the safe way. But safe for whom? Organisations that run on the smell of an oily rag, by volunteers, can’t afford to pass up huge efficiency gains. Would you advise such an organisation to post letters instead of using email? Certainly not, because the saving of time and resources is too valuable to pass up. We’re entering a similar world when it coming to AI technology—and we need purpose-built tools designed for the community sector. For organisations with limited resources, who otherwise won’t be able to find and apply for the grants they need to operate, AI isn't a shortcut, it's basic access.

Better tools, not more hours

The community sector in Aotearoa is stretched. Organisations are being asked to do more with less, and the people doing critical work are often the least resourced. Telling them to reject tools that could save them hundreds of hours is, at best, unhelpful. At worst, it protects an industry built on the assumption that grant writing has to be done the old way. It doesn't have to be so hard anymore.