Funding Follows Networks, Not Keywords
I spent months building a foundation prospecting tool using keyword matching: embeddings, semantic similarity, and other modern NLP methods. It worked reasonably well. Then I analyzed co-funding networks and found a signal 12.8 times stronger.
A clear pattern emerged across 1,000,000+ grants, 15,000 foundations, and 300,000 nonprofits: foundation giving follows network structures that are invisible to traditional prospecting tools.
What traditional prospecting misses
Most foundation prospecting tools work the same way. They take a nonprofit's mission, programs, or focus areas and try to match them against a foundation's stated interests. The better tools use semantic similarity or embedding-based search to go beyond exact keyword matches. These methods are useful, but they operate on what foundations say they fund. The network signal operates on what foundations actually do.
Foundations that co-fund the same organizations tend to keep co-funding them. When Foundation A and Foundation B both fund three of your peer organizations, Foundation B is a strong prospect for you even if their stated mission has zero keyword overlap with yours.
This is not a theoretical insight. It is a measurable, testable signal, and in our data it outperforms keyword-based matching by a wide margin.
A real example
Lincoln Financial Foundation is based in Pennsylvania. Boys & Girls Clubs of Northeast Indiana is in Fort Wayne, Indiana. Different states. No obvious connection. Zero keyword overlap between the foundation's stated interests and the organization's programs.
Our network model ranked Lincoln Financial #1 out of 69 candidate funders for Boys & Girls Clubs of Northeast Indiana. They have since made a $50,000 first-time grant.
Geography is a powerful signal in philanthropy. But network signal is stronger. Lincoln Financial is not in Indiana, yet the network revealed a relationship that geography, keywords, and program alignment would have missed entirely.
Why networks predict funding better than keywords
Foundation giving is not a matching problem. It is a network problem.
A foundation's 990-PF filing tells you exactly who they funded and for how much. When you map those relationships across thousands of foundations and hundreds of thousands of recipients, the network structure becomes predictive. Foundations that share grantees tend to share future grantees. That co-funding signal is what keyword search cannot capture.
The same pattern appears across the full dataset
The Lincoln Financial example is not an outlier. Across the SciRise dataset, network-based predictions consistently surface foundation prospects that keyword and geography-based methods miss.
The pattern is strongest in cases where:
- A nonprofit has several existing foundation funders (providing network context)
- Those funders participate in co-funding clusters with other foundations
- The candidate foundation has funded peer organizations but not the target nonprofit
In these cases, the network signal identifies prospects that would never appear in a keyword search, because the connection runs through shared funding behavior rather than shared language.
What this means for nonprofits
If your prospecting strategy relies primarily on keyword matching, mission alignment scores, or stated foundation interests, you are working with incomplete information. Those tools will find the obvious matches. But the highest-value prospects, the foundations most likely to fund you, often have no keyword overlap with your organization at all.
The funding network contains signal that traditional prospecting tools cannot access. Foundations do not make grants in isolation. They operate within networks of relationships, co-funding patterns, and institutional knowledge that shape where dollars flow.
Today, we are introducing SciRise Network Matching. It surfaces the foundations your peer organizations are funded by, identifies co-funding clusters relevant to your organization, and ranks prospects by network proximity rather than keyword similarity.
If you want to see which foundations you are missing, visit scirise.ai.
Data source: SciRise Foundation Intelligence dataset. Analysis based on 1,000,000+ grants from 15,000+ foundations to 300,000+ nonprofit recipients, derived from IRS 990 and 990-PF filings (2023-2024).