GitLab Foundation, the $30 million philanthropic arm of the AI software company of the same name, needed some extra brain power to sort through all the funding requests it received in September. The company had awarded grants totaling $4 million to nonprofit organizations testing ways to better use artificial intelligence to improve economic opportunity. So, naturally, grant makers turned to their own AI systems to make sense of the 800 applications that poured in.
Ellie Bertani, president of the GitLab Foundation, said reviewing such a large number of applications would take GitLab’s three program leaders hundreds of hours without the help of technology. The whole thing took 30 minutes, as it “actively” used AI to review applications.
Bertaini’s team trained the AI tool to look at several criteria, including what kind of wage increase an applicant can generate, how strong the partnership is, and what kind of data security systems are in place.
By using AI to query large datasets, Bertani can more easily observe trends in the application, such as where resources are needed most and how the foundation can fine-tune the questions it asks of grantees.
The GitLab Foundation’s enthusiasm for AI makes it a bit of an anomaly among grant makers. Chantal Forster, an expert on the use of technology in philanthropy who has consulted with more than a dozen foundations, including the Annenberg Foundation, the Lumina Foundation, the MacArthur Foundation and the Robert Wood Johnson Foundation, on the use of AI, says that while many other foundations are using the technology as part of their grant selection process, only a few are using the tools in a comprehensive way.
Philanthropy technology experts like Forster envision how foundations and nonprofits can leverage AI to speed up grant applications. It can also help match grant applicants and grant writers with common objectives and strong approaches to problem-solving. More radical use of AI and its ability to make sense of vast data sets could transform the entire grant application process, resulting in more automated systems for connecting nonprofits and foundations.
GitLab Foundation
At GitLab Foundation, Bertani and his staff fed large amounts of data from grant applications, grant reports, and check-in calls into a large-scale language model that can respond to prompts, help identify patterns, and provide insights into things like common factors in the success and failure of the foundation’s 190 grants.
“This will help define future strategies,” she said, adding that the responsibility for actual grant decisions still lies with humans, not machines.
Ultimately, Bertani said, the use of AI will help with GitLab’s core job of quickly providing funding to nonprofits.
“We feel a huge responsibility to get the funds out as quickly as possible because the real impact is not on us, but on the grant recipients,” she said. “AI can be very helpful in speeding up the process of gathering insights, making the right bets, and recovering money.”
AI ambivalence
Foundation staff are using AI, and Forster said about 60 percent have tried free tools like ChatGPT, but most grantmakers don’t have a formal plan for how to use AI. As a result, grantmakers don’t know how AI is impacting their work or how best to use it, Forster said.
Foundation personnel are often concerned about the use of AI. Because using AI can cost you your job.
In many cases, she says, Foundation employees are concerned about the use of AI because it could cost them their jobs. As more philanthropies face pressure to provide more funding to grantees, program staff represent overhead costs that can be reduced through the use of technology tools.
Cash-starved nonprofits face a different dynamic when considering AI, she said.
“They have real financial constraints,” she says. “They want to use whatever tools are available to get the job done” and reduce costs.
Last spring, the Center for Effective Philanthropy surveyed nonprofits and foundations and found that grantmakers did not have a clear view of how grantees could benefit from AI. More than 80 percent of foundations surveyed said their staff has little or no understanding of how AI can help grantees improve efficiency, and little understanding of grantees’ technical capabilities to use AI.
Despite uncertainty about how AI can help with the grant application process, the rapid growth of automation tools is putting significant pressure on nonprofits to incorporate them into their plans. Before doing so, leaders need to plan carefully, suggests Jean Westrick, president of the Technology Grants Association.
Technology won’t address funding inequities unless people remain deeply involved, she says. Also, using it just to speed up decision-making can have unintended consequences. Only human involvement can ensure underrepresented organizations are discovered and supported, she said. What she wants is an application process that feels more like a dialogue between funders and grant recipients.
“The application process is broken and AI can’t fix it,” she said.
AI can cause overcommitment
Last fall, Habitat for Humanity Michigan held an AI training session for 43 Habitat offices across the state. Wendy Crow, the group’s director of operations who led the session, was concerned about how, or even if, groups, which range in size from two to 70 staff members, were using technology.
She wanted to get everyone on the same page, so she worked with a consultant to provide training on things like AI policies, how to use AI to craft fundraising emails, and how to automate tasks like donor verification.
Crowe encourages the use of AI in grant applications because it has helped her personally. Using Google NotebookLM, Crow can take a 20-page request for proposal from a foundation full of “legal jargon” and get an easy-to-understand overview. This tool allows you to quickly decipher what other resources accompany your application, such as data about your operations, staff hours from members of different departments, and letters of support from financial institutions and partners.
With the increasing use of AI in grant applications, Crow is unable to show a final return on investment. But she said artificial intelligence saves time and drafts the wording of proposals, giving a tighter picture of funding bodies’ missions and priorities.
With AI, it’s very easy to make yourself look bigger than you actually are. Having an AI write the story can add a lot of fluff, which can be dangerous.
—Wendy Crowe, Michigan Habitat for Humanity Director of Operations
But she said she would never use AI to create a proposal and just send it out. Without an eye to monitor everything the tool uses, artificial intelligence can become a little overkill for nonprofits.
“With AI, it’s very easy to make you look bigger than you are,” she said. “If you let AI write the story, you can end up with redundant parts, which is dangerous. It can lead to overcommitment.”
human atmosphere
Grant writers can also tell if their application was created by AI.
Alison Bajracharya, chief impact and strategy officer at the Ewing and Marion Kaufman Foundation, said that when grantmakers use AI to quickly generate large numbers of applications, it’s often “not satisfying.”
“If we see something that seems very common, or if it doesn’t really match the questions we asked, that certainly raises some red flags,” she said.
Last fall, Kaufman deployed a company-wide AI system to support research and enhance the grant review process. Bajracharya said AI implementation is happening in stages across the organization, but with the addition of AI fellows this spring, he believes grantmakers will adopt a more fully developed plan.
Vajracharya’s first concern was security. She didn’t want applicants’ sensitive information and underlying data to be siphoned off into large databanks that would enable AI. With an enterprise system, all your information is kept safe and secure within your organization.
She also didn’t want AI to replace the expert insight of her staff. While AI helps Kaufman staff derive a variety of insights about applications, program officials still perform the initial review of the 100 or so grant applications the foundation receives each grant cycle.
One way AI can help is by comparing sets of similar applications. Using AI, Bajracharya and his staff can query batches of, say, 10 applications and compare many factors side-by-side, such as impact, top spend, and other budget items.
Using this matrix will help Mr. Vajracharya intuitively check whether an applicant’s plan is feasible or too ambitious. But ultimately, she and her staff have the final say.
Bajracharya says: “We don’t want to just substitute judgment.”
