The gap between a world-changing innovation and its funding typically comes all the way down to 4 minutes—the typical time a human reviewer tends to spend on an preliminary grant utility. In these 4 minutes, reviewers should assess alignment, eligibility, innovation potential, and group capability, all whereas sustaining consistency throughout 1000’s of purposes.
It’s an not possible ask that results in an not possible alternative: both decelerate and evaluation fewer concepts or velocity up and threat lacking transformative ones. At MIT Clear up, we’ve spent a yr exploring a 3rd possibility: educating AI to deal with the repetitive elements of evaluation so people can make investments actual time the place judgment issues most.
WHY AI, AND WHY NOW
In 2025, Clear up acquired almost 3,000 purposes to our World Challenges. Even a cursory four-minute evaluation per utility would add as much as 25 full working days. Like many mission-driven organizations, we don’t need to commerce rigor for velocity. We would like each.
That led us to a core query many funders at the moment are asking:
“How can AI assist us consider extra alternatives, extra pretty and extra effectively, with out compromising judgment or values?”
To reply this query, we partnered with researchers from Harvard Enterprise Faculty, the College of Washington, and ESSEC Enterprise Faculty to check how AI may help early-stage grant evaluation, one of the crucial time-intensive and high-volume phases of the funding lifecycle.
WHAT WE TESTED AND WHAT WE LEARNED
The analysis group developed an AI system (based mostly on GPT-4o mini) to help utility screening and examined it throughout reviewers with various ranges of expertise. The purpose was to know the place AI provides worth and the place it doesn’t.
Three insights stood out:
1. AI performs greatest on goal standards. The system reliably assessed baseline eligibility and alignment with funding priorities, figuring out whether or not purposes met necessities or match clearly outlined geographic or programmatic focus areas.
2. AI is extra useful to much less skilled reviewers. Much less skilled reviewers made extra constant choices when supported by AI insights, whereas skilled reviewers used AI selectively as a secondary enter.
3. The largest acquire was standardization at scale. AI made judgments extra constant throughout reviewers, no matter their expertise, making a stronger basis for the second stage of evaluation and human decision-making.
HOW THIS TRANSLATES INTO REAL-WORLD IMPACT
At Clear up, the primary stage of our evaluation course of focuses on filtering out incomplete, ineligible, or weak-fit purposes, liberating human reviewers to spend extra time on essentially the most promising concepts.
We designed our AI software with people firmly within the loop, centered on the repetitive, pattern-based nature of preliminary screening that makes it uniquely fitted to AI augmentation. The software:
- Screens out purposes with no lifelike path ahead.
- Helps reviewers with a passing likelihood rating, a transparent suggestion (Cross, Fail, or Evaluation), and a clear rationalization.
When the 2025 utility cycle closed with 2,901 submissions, the system categorized them as follows: 43% Cross; 16% Fail; and 41% Evaluation. That meant our group may focus deeply on simply 41% of the purposes—slicing whole screening time down to 10 days—whereas sustaining confidence within the high quality of the outcomes.
THE BIGGER TAKEAWAY FOR PHILANTHROPY
Each hour saved throughout the early phases of analysis is an hour redirected towards the higher-value work that people excel at: partaking extra deeply with innovators and getting daring, under-resourced concepts one step nearer to funding.
Our early outcomes present robust alignment between AI-supported screening and human judgment. Extra importantly, they exhibit that it’s potential to design AI methods that respect nuance, protect accountability, and scale decision-making responsibly.
The philanthropic sector processes hundreds of thousands of purposes yearly, with acceptance charges typically under 5%. If we’re going to reject 95% of concepts, we owe candidates—particularly these traditionally excluded from funding—a real evaluation. Dividing duty, with people making choices and AI eliminating rote evaluation, makes it that rather more potential at scale. It’s a sensible step towards the thoroughness our missions demand.
Hala Hanna is the chief director and Pooja Wagh is the director of operations and influence at MIT Clear up.
