If you sell software, professional services, financial infrastructure, or research tools into the U.S. nonprofit and foundation market, your customers file an annual disclosure to the IRS that runs 20 to 200 pages, names many of their largest vendors and contractors, and is published on the public internet within months of filing. That document is IRS Form 990, and it is one of the most underutilized public customer signal sources for any B2B vendor selling into the tax-exempt sector.
This sits in the broader regulatory disclosure family alongside customer-cdp-climate-disclosure-and-tcfd-aligned-sustainability-report-product-mentions-extraction-workflow-from-public-climate-disclosure-archives and customer-fcc-form-477-broadband-deployment-report-and-broadband-data-collection-disclosure-product-mentions-extraction-workflow-from-public-telecommunications-archives, but the customer set is different: nonprofit operating organizations, private foundations, public charities, hospitals, universities, museums, and trade associations all file Form 990 in one of its variants.
What Form 990 Is and What It Discloses
Form 990 is the annual information return that every U.S. organization recognized as tax-exempt under Internal Revenue Code section 501(c) must file with the IRS. The base form is supplemented by sixteen schedules (A through R) that capture specific disclosures depending on the nature of the organization. The full filing is submitted on a fiscal-year cadence and posted to the public record by the IRS through its Tax Exempt Organization Search and by intermediaries including ProPublica's Nonprofit Explorer, GuideStar (Candid), and the Foundation Center.
There are four main variants:
- Form 990 — the full form, required for organizations with gross receipts above $200,000 or assets above $500,000.
- Form 990-EZ — a shorter version for organizations with gross receipts between $50,000 and $200,000.
- Form 990-N (e-Postcard) — a minimal electronic notice for organizations with gross receipts at or below $50,000.
- Form 990-PF — the version filed by private foundations, with significantly more detail on grants and investment activity.
For vendor mention extraction, the full Form 990 and Form 990-PF are the high-signal variants. The 990-EZ has limited contractor disclosure and the 990-N has essentially none.
The filings are public the moment they are received and processed. Most large nonprofits file within five and a half months of fiscal year-end, often with extensions that push the public availability to seven to nine months after the fiscal year closes. The result is a predictable, annually-updating signal on the operational stack of essentially every funded U.S. nonprofit.
Where Customer Product Mentions Appear in Form 990
The product mention extraction value is concentrated in a small number of items and schedules. Knowing exactly where to look is the difference between a workflow that takes 30 hours per filing and one that takes 20 minutes.
Form 990 Part VII — Compensation of Officers, Directors, Trustees, Key Employees, Highest Compensated Employees, and Independent Contractors
Part VII Section A lists the named individuals. Section B is the section that matters for vendor extraction: it names the five highest-compensated independent contractors that received more than $100,000 in the tax year, along with the type of service provided. This is the most structured vendor disclosure in any federal U.S. tax filing.
Typical mentions in Section B for a mid-size nonprofit include law firms (corporate counsel, employment counsel), accounting firms (audit, tax), management consulting firms, fundraising consultants, technology service providers, and architecture or engineering firms for capital projects. For a hospital, the list extends to anesthesiology groups, radiology groups, and physician staffing firms. For a university, it extends to construction managers, food service vendors, and licensing agents.
Schedule O — Supplemental Information
Schedule O is the narrative supplement where the organization provides explanations the structured form does not accommodate. This is the highest-density unstructured product mention surface in the entire Form 990. The schedule contains:
- Narrative answers to multi-part questions from Parts III through XII of the base form (mission, program accomplishments, governance, policies, financial statement footnotes).
- Explanations of audit, compensation, and conflict-of-interest policies — often naming the audit firm, the compensation consultant, and the legal counsel that reviewed the policy.
- Description of program activities — frequently names the technology platforms, data providers, and service partners used to deliver programs.
- Compliance with state regulatory requirements — sometimes names registered agent services and state filing agents.
Schedule O has no length cap and is written in prose. Mentions are inline, not structured. Extraction requires NER or substring match against a vendor name lookup table.
Schedule J — Compensation Information
Schedule J expands on Part VII compensation and discloses additional detail on retirement plans, severance, and non-cash compensation. The schedule routinely names the retirement plan provider (Fidelity, Vanguard, TIAA, Empower) and sometimes the executive compensation consultant.
Schedule R — Related Organizations and Unrelated Partnerships
Schedule R discloses related corporate-family entities and unrelated partnerships in which the nonprofit holds an interest. These are not "vendor" mentions but reveal organizational structure that affects sales targeting.
Form 990-PF Part XV — Supplementary Information on Grants and Contributions
For private foundations filing 990-PF, Part XV is the grants list — every grant the foundation made above the disclosure threshold, with grantee name and amount. This is not a vendor disclosure but is the highest-value document in U.S. philanthropy for understanding foundation funding patterns. For B2B vendors selling into foundations, the relevant section is the foundation's own administrative expenses on Part I and the contractor list on Part VIII.
The Extraction Workflow
The workflow we recommend has five stages. Each is mechanical enough to automate, but the first run on a new nonprofit cohort is usually worth doing manually to calibrate.
Stage 1: Build the organization universe
Start by deciding which nonprofits you care about. The IRS publishes the Exempt Organizations Business Master File (EO BMF) as a free monthly download — a structured CSV of every currently tax-exempt organization in the U.S. with EIN, name, address, ruling date, asset and income ranges, and NTEE classification codes. This file is the right starting point for building a target universe by mission area, size, or geography.
For higher-resolution targeting, the IRS also publishes the Form 990 Series Returns dataset on AWS — every electronically filed 990, 990-EZ, and 990-PF since 2010, available as XML at no cost. ProPublica's Nonprofit Explorer wraps this dataset in a searchable UI and is often the cleanest manual research path before automating.
Stage 2: Pull the filings
For automated extraction, fetch the XML files from the IRS AWS bucket. The XML is well-structured and includes all Part VII and Part VIII data as parseable fields. Schedule O appears as a long narrative field in the XML.
For non-electronic filings (older filings, smaller organizations, and some 990-PFs), the filings are scanned PDFs that the IRS releases through its Tax Exempt Organization Search. Most of the volume since 2018 is electronic; older filings are mixed.
Cache the filings locally with a deterministic naming convention ({ein}_{tax_year}_990.xml) so you can reprocess as your vendor lookup table grows.
Stage 3: Parse the structured fields
For XML filings, parse Part VII Section B (highest-compensated independent contractors) directly. The fields are: contractor name, business address, type of service, and compensation amount. This is the highest-confidence vendor mention extraction in the entire filing and requires no NER.
Schedule J retirement plan provider and Schedule R related organization fields are similarly structured.
Stage 4: Mention extraction from Schedule O narrative
Schedule O is the unstructured narrative section. Two extraction approaches and a hybrid:
Approach A — Vendor lookup table substring match. Maintain a list of vendor names with common variants and run case-insensitive match against the Schedule O text. Capture the surrounding sentence as context. This is high-precision for known vendors but misses unknown ones.
Approach B — NER on the narrative text. Run a fine-tuned NER model trained on organizational mentions and filter against a stop-word list of common non-vendor entities (government regulators, IRS, the filing organization's own name, common grantmakers if you are not selling to grantmakers). Higher recall, higher false-positive rate.
In production, run Approach A first to capture high-confidence known mentions and Approach B over what is left to surface novel vendors for human review.
Stage 5: Validation and source metadata stamping
Every extracted mention should carry: EIN, organization name, tax year, schedule and field of the source (e.g., "Part VII Section B" or "Schedule O explanation of Item 12c"), and the surrounding sentence or table row. This metadata is what lets you cite the mention with confidence in a sales motion, in a customer logo wall, or in an analyst briefing.
See testimonial-with-source-link-to-evidence for the broader practice of attributing customer mentions to verifiable public sources, and how-to-collect-product-mentions-from-public-regulatory-filings for the cross-regulator framework this Form 990 workflow plugs into.
What Form 990 Is Not Useful For
Three categories of customer signal that Form 990 does not reliably contain:
Specific software products below the materiality threshold. Form 990 does not require a nonprofit to itemize software subscriptions or SaaS spend unless they meet the highest-compensated contractor threshold of $100,000 per year for a single vendor. For a SaaS vendor with $20,000 annual contracts, your customers are invisible on this disclosure unless they aggregate spend through a reseller above the threshold.
Pricing and contract terms. Form 990 discloses that a vendor is used and the total compensation, but does not disclose unit pricing or contract structure. This is not a useful source for competitive pricing intelligence.
Mid-size vendor relationships. Below the five-vendor cap in Part VII Section B, vendors are aggregated and not named. A nonprofit may have 50 contractor relationships in the $20,000 to $80,000 range that are not individually disclosed. This is structurally invisible.
The corollary: Form 990 is most useful for vendors whose typical contract size with a nonprofit customer is above $100,000 per year — law firms, audit firms, management consulting, retirement plan administration, executive search, technology implementation, and enterprise software in the upper tier. For these vendors, Form 990 produces a defensible, annually-updating customer signal that requires zero opt-in from the customer.
The Annual Cadence and the Freshness Window
Form 990 is due five and a half months after fiscal year-end with extensions available for up to six additional months. For organizations with a December fiscal year-end, the initial filing deadline is May 15 of the following year and the final extended deadline is November 15. Most large nonprofits use the extension and file in the August to November window. Smaller organizations file closer to the May deadline.
The IRS posts electronic filings to the AWS bucket within roughly two to six weeks of receipt. Combined, the practical freshness window is:
- February to May. New filings beginning to land for prior tax year December year-end organizations that did not extend.
- June to August. Mid-cycle. Volume picks up.
- September to December. Peak filing volume. The bulk of large-organization filings land here.
- January. Cleanup window before the next year's cycle begins.
For ongoing monitoring, weekly re-fetch of the IRS Form 990 XML bucket catches new filings throughout the year. The cadence is more predictable than SEC filings because the nonprofit fiscal-year distribution is heavily concentrated at December and June year-ends.
Sector-Specific Notes for the Highest-Value Verticals
Three nonprofit verticals are disproportionately high-signal for vendor extraction.
Hospitals and healthcare systems
Tax-exempt hospitals file Form 990 plus Schedule H (Hospitals), which provides additional disclosure on community benefit, charity care, and operational data. Vendor mentions for hospitals cluster heavily in Part VII Section B (physician group contractors, anesthesiology, radiology, hospitalists), Schedule H Part V (community benefit consultants and audit firms), and Schedule O policy explanations (electronic health record platforms, revenue cycle management vendors, group purchasing organizations).
Universities and educational institutions
Universities file Form 990 plus Schedule E (Schools). Vendor mentions cluster in Part VII Section B (construction managers, food service, athletics-related contracts), Schedule J retirement plan disclosures (TIAA dominance, with Vanguard and Fidelity gaining share), and Schedule O explanations of policy implementation (research compliance consultants, Title IX consultants, athletic compliance consultants).
Private foundations
Private foundations file Form 990-PF. Vendor mentions cluster in Part VIII (compensation of officers, directors, and contractors), Part XV (grants list — relevant for grantmaking software vendors), and the supplementary information sections. Investment managers and custodians for foundation endowments are disclosed in Part X (Books) and the investment schedules.
Final Thoughts
IRS Form 990 is the most reliable public source of customer product mentions in the U.S. nonprofit and foundation sector. The structured highest-compensated contractor disclosure in Part VII Section B, combined with the unstructured but high-density Schedule O narrative, makes it disproportionately valuable for vendors whose typical nonprofit contract size sits above the $100,000 disclosure threshold.
The work to build the extraction pipeline is real but tractable: monthly EO BMF download for the master list, AWS XML fetch for the filings, structured field parsing for Part VII and Schedule J, NER and lookup-table mention extraction for Schedule O, and source-metadata stamping for downstream defensibility. Once built, it generates customer evidence at an annual cadence without requiring any customer outreach or opt-in, across a market that includes essentially every funded U.S. nonprofit.
For complementary public-disclosure workflows in related markets, pair this with customer-eu-ai-act-conformity-assessment-and-high-risk-ai-system-disclosure-product-mentions-extraction-workflow-from-public-ai-regulation-archives for the European regulatory archive and customer-dol-form-5500-employee-benefit-plan-annual-report-product-mentions-extraction-workflow-from-public-erisa-disclosure-archives for the U.S. retirement plan archive that often complements Form 990 for retirement plan administrators selling into the nonprofit sector.