MIT Solve Uses AI to Screen 3,000 Grant Applications Fairly

Imagine having just four minutes to decide if a world-changing idea deserves funding. That's the impossible reality most grant reviewers face when sorting through thousands of applications.

MIT Solve decided there had to be a better way. In 2025, they received nearly 3,000 applications for their Global Challenges program. Reviewing each one properly would have taken 25 full working days.

So they partnered with researchers from Harvard Business School, the University of Washington, and ESSEC Business School to test whether AI could handle the repetitive parts of screening. The goal wasn't to replace human judgment but to free up reviewers to focus where it matters most.

They built a system using GPT-4o mini and tested it across reviewers with different experience levels. The AI checked objective criteria like eligibility requirements and alignment with funding priorities. It flagged applications that clearly didn't fit and highlighted ones that deserved deeper review.

The results were striking. The AI categorized applications into three groups: 43% passed initial screening, 16% were filtered out, and 41% needed human review. That meant reviewers could focus deeply on just 1,190 applications instead of all 2,901.

The research revealed three key insights. First, AI excelled at checking baseline requirements and geographic focus areas. Second, less experienced reviewers made more consistent decisions with AI support, while veterans used it as a helpful second opinion. Third, the system standardized judgments across all reviewers, creating fairness at scale.

The Ripple Effect

Every hour saved on initial screening becomes an hour spent engaging deeply with innovators who need support. For MIT Solve, cutting screening time to ten days meant more time helping bold, under-resourced ideas get closer to funding.

The impact extends beyond one organization. The philanthropic sector processes millions of applications annually, with acceptance rates often below 5%. If 95% of ideas get rejected, applicants deserve a genuine review, especially those historically excluded from funding opportunities.

Hala Hanna and Pooja Wagh from MIT Solve emphasize that humans stay firmly in control. The AI handles pattern recognition and repetitive tasks while people make the final calls. It's about dividing responsibilities wisely, not replacing human judgment with algorithms.

The system provides transparent explanations for every recommendation, ensuring accountability and trust. Reviewers see a probability score, a clear recommendation, and the reasoning behind it. They can accept, reject, or override any AI suggestion.

This approach offers a practical solution to a widespread problem: how to maintain thoroughness and fairness when resources are limited. By letting AI handle the routine work, funders can finally give promising ideas the attention they deserve.

The future of grant review isn't choosing between speed and rigor—it's finding smart ways to achieve both.