Learn to decode foundation financial documents and use AI to reveal funder priorities, patterns, and capacity.
A foundation's Form 990 is its financial truth document. While a funder's website proclaims interest in "innovative youth initiatives," their 990 reveals exactly which youth programs they funded, how much they gave, and whether they're increasing or decreasing investment in that area. This gap between stated interests and actual giving patterns is where grant strategy is made or broken.
Form 990-N, 990-EZ, and 990 filings are public documents required by the IRS. They contain the universe of grants a foundation made that year, grant amounts, and grantee names. For grant writers, this is intelligence gold. Instead of guessing what a funder wants, you can see exactly what they've supported, spot patterns, and determine if your organization fits their actual priorities.
AI accelerates this research dramatically. What previously required hours of manual document review—downloading, scanning, highlighting, comparing—AI systems can do in minutes, extracting patterns and identifying thematic priorities from thousands of grants.
Foundation 990s contain Schedule I, which lists every grant made during the fiscal year, typically including: grantee name, grant amount, and often a brief description of the grant purpose. This one schedule is the foundation's complete grant history for that year. By reviewing three to five years of Schedule I, you develop a comprehensive picture of funder priorities.
The grants listed aren't filtered or curated—they represent the actual portfolio. Some foundations make 20-30 grants annually; others make hundreds. A foundation claiming broad interest in "education" but making most grants to STEM programs with occasional international education grants reveals their true priority through distribution.
The 990's Part I contains total assets and revenue. A foundation with $50 million in assets typically makes different-sized grants than one with $5 million. The 990 also shows required distributions (5% of assets for most private foundations), revealing realistic annual giving capacity. If a foundation's assets declined significantly, they may have reduced funding capacity even if their guidelines haven't updated.
By analyzing multiple years of 990s, you can determine a foundation's typical grant size range, whether they make larger strategic grants, and whether ask sizes have shifted. A foundation averaging $50,000 grants is different from one averaging $250,000—not just in capacity but often in program maturity requirements.
Many foundations' 990s list grantee locations or grant purposes that indicate geographic focus. Some claim national scope but fund predominantly in a few regions. Others explicitly geographic but have begun expanding. These patterns, visible in three-year grant histories, reveal whether your location is viable or historically outside their range.
Beyond stated interests, your analysis will reveal actual thematic clusters. A family foundation might claim broad interest in "community development" but actually fund consistently in housing, workforce development, and youth services. Other potential impact areas (health, environment, education) get minimal or no funding despite being plausible for their stated mission.
Form 990s provide unfiltered data about what funders actually support. Stated interests on websites are aspirational; 990s reveal reality. Analyzing grant patterns across multiple years provides the most reliable funder intelligence available.
Modern AI systems can ingest multiple 990 PDF documents and extract structured data: all grants listed, amounts, descriptions, and grantee information. Instead of manually reading and highlighting, you receive a formatted summary showing all grants from a specific period. For a foundation that made 150 grants over three years, AI can provide a clean list in seconds.
AI excels at identifying patterns humans might miss with limited time. It can analyze grant descriptions and categorize grants thematically, showing that a foundation claiming interest in five areas actually allocates 60% of funding to two areas. It can identify geographic patterns by analyzing grantee locations. It can calculate average grant sizes, identify outliers, and track trends across years.
AI can compare multiple foundations simultaneously, showing which have similar portfolios, which fund at different scales, and which have diverging thematic focuses. If you're researching five prospects, AI can create a comparison matrix showing funding size ranges, thematic concentration, and geographic coverage for all five.
Advanced AI systems can compare your organization's description against a foundation's actual grant history and calculate alignment. Not just "do you both work in education," but "what percentage of foundation grants went to organizations similar to yours? How much did they give to organizations at your maturity level? Is your geographic location represented in their funding?"
Start with ProPublica's Nonprofit Explorer (projects.propublica.org/nonprofits), GuideStar (now Candid), or the IRS tax-exempt organization database. Download three to five years of 990s for each major prospect. Most foundations file annual 990s; larger foundations file extensively detailed 990-PF forms with detailed grant information.
Using AI document processing tools, upload the PDFs and request extraction of all grant information. Quality AI systems will provide clean, structured output: a table or spreadsheet showing each grant, amount, grantee, and any available description. You now have the foundation's actual portfolio in analyzable format.
Ask the AI to categorize grants thematically and show the distribution. Request it identify primary focus areas (where 70%+ of funding goes), secondary areas (10-30%), and marginal areas (under 10%). Many foundations claim broader missions than they actually fund.
Select three foundation 990s in your sector. Using AI tools (ChatGPT, Claude, or specialized grant research platforms), extract all grants from the past three years. Categorize them thematically, calculate average grant size, identify geographic patterns, and compare foundation priorities. Document your findings in a one-page summary showing thematic distribution, average grant size, and assessment of fit for your organization.
Calculate the foundation's annual grant distribution (from Part I of the 990), identify median and average grant sizes, and determine your ask size relative to their typical grants. A foundation making average $75,000 grants isn't well-suited for a $400,000 request. Understanding capacity prevents pursuing misaligned prospects.
Compare 990s across three years. Is annual giving increasing or declining? Are grant sizes growing or shrinking? Are thematic priorities shifting? A foundation that funded health initiatives heavily three years ago but has shifted to education may not be receptive to traditional health proposals. Trends matter more than single-year snapshots.
AI analysis of 990s is powerful but not infallible. If an AI system categorizes a grant description ambiguously, verify its interpretation. Don't blindly trust automated categorization without spot-checking a sample of grants. The best practice is AI-assisted analysis verified by human judgment.
Grant descriptions on 990s range from minimal ("Grant to ABC Organization") to detailed narratives. Detailed descriptions reveal what the foundation found compelling about that grant. Look for repeated language, specific outcome expectations, or partnership requirements that appear across multiple grants. This language indicates what matters to the funder.
Analyze other grantees in your subsector. Do they share characteristics with your organization? Are there three organizations very similar to yours that received funding? No organizations similar to yours in five years of grants? That's significant intelligence. Look at grantee sophistication, size, and type to understand whether the foundation funds organizations at your maturity level.
Notice what the foundation doesn't fund. A foundation claiming education interest but with zero grants to K-12 schools in a three-year period has effectively excluded that sector. Their actual priorities have narrowed compared to their stated mission. These gaps are as important as areas they actively support.
Some grants don't fit a foundation's typical pattern—larger than usual, outside their geographic area, thematically divergent. These outliers often signal emerging interests, board member influences, or special initiatives. Outliers worth noting, but a single outlier grant doesn't necessarily indicate new priority.
After analyzing a foundation's 990s, document your findings in a one-page intelligence summary:
The ultimate value of 990 analysis is not the data itself but how it informs your strategy. A foundation with perfect thematic match but declining assets and shrinking grants may not be worth significant effort. A foundation with lukewarm website interest but demonstrably funding organizations like yours (shown in recent 990 grants) is a stronger prospect than the algorithms suggest.
As you progress through this course, you'll use this intelligence foundation to prioritize prospects, customize proposals, and target asks appropriately. The grant writers who win are those who combine algorithmic matching with the human intelligence that 990 analysis provides.
Form 990s reveal actual funder priorities, capacity, and patterns. AI accelerates analysis of multiple documents and identifies patterns. Use AI for extraction and pattern discovery, then apply human judgment to verify alignment and assess strategic fit. The combination is more powerful than either approach alone.
In the next lesson, you'll expand beyond individual funder research to competitive intelligence—understanding your entire funding landscape and the strategies of other organizations competing for the same resources.
Learn Competitive Analysis