Research Methodology
How BDI analyzes corporate equity commitments using AI-assisted research and evidence-based classification
Our Approach
The Black Dollar Initiative employs a systematic, evidence-based methodology to analyze corporate equity commitments affecting Black economic prosperity. Our process combines AI-assisted document analysis with rigorous human oversight to extract, classify, and assess corporate actions from publicly disclosed reports.
We prioritize transparency, specificity, and accountability—documenting what companies actually do, who benefits, and what outcomes are reported. Our methodology distinguishes between commitments and results, tracks resources across domains, and measures impact using evidence-based criteria.
Evidence Collection & Ingestion
We systematically extract verbatim evidence from corporate ESG reports, sustainability disclosures, diversity reports, and annual filings. Our AI-assisted extraction process identifies and preserves relevant text while maintaining strict page-level citations.
What We Extract
- Explicit references to Black people, Black communities, People of Color, and broad Diversity Signals
- Workforce programs, policies, and demographic disclosures
- Supplier diversity commitments and spend data
- Community investments, philanthropy, and place-based initiatives
- Governance structures, oversight mechanisms, and accountability systems
- Grievance and remedy processes
- Metrics, targets, and progress reporting
Context Lock Principle: We extract complete program descriptions including eligibility criteria, dollar amounts, timelines, enforcement mechanisms, and reported outcomes—not isolated quotes or fragments.
AI-Assisted Classification
Using large language models (LLM) with custom classification protocols, we transform raw evidence into structured data. Our classification system answers four core questions for every documented action, ensuring consistency and rigor across thousands of corporate commitments.
The Four Core Questions
- What did the company do? — Programs, policies, investments, disclosures, or remedies
- Who is impacted? — Black People, Black Communities, People of Color, or broader diversity initiatives
- What is the level of impact? — Systemic, Transformative, Moderate, or Limited scope
- What is the direction of impact? — Positive, Negative, Neutral, or Inconclusive outcomes
Human Review & Quality Control
Every AI-generated classification undergoes human review to verify accuracy, resolve ambiguities, and ensure alignment with our evidence standards. BDI researchers validate demographic routing, assess contextual nuance, and flag items requiring deeper investigation.
Classification Framework
Focus Areas
Black People: Explicit references to Black/African American individuals, Black-owned businesses, and Black-specific institutions (HBCUs, cultural organizations)
Black Communities: Place-based actions in majority-Black areas with explicit Black identification
People of Color: Multi-racial commitments including BIPOC, Latinx, AAPI, or aggregated minority groups
Diversity Signals: Race-neutral equity language including “underrepresented,” “underserved,” or general DEI initiatives
Scale Assessment
Systemic: Company-wide or system-wide policies with durable structures
Transformative: Major change in one business unit, market, or geography
Moderate: Bounded programs with measurable but limited scope
Limited: One-time events, pilots, or symbolic gestures
Impact Assessment
Positive: Demonstrated benefit or harm reduction with evidence
Negative: Demonstrated harm, rollbacks, or outcomes below equity thresholds
Neutral: Action exists with no meaningful change in outcomes
Inconclusive: Outcomes not yet knowable (new initiatives, commitments without results)
Primary Domains
Culture & Policy: Internal workforce, hiring, pay equity, leadership, ERGs, and HR practices
Supplier: Procurement, supplier diversity, vendor requirements, and sourcing commitments
Community: External investments, philanthropy, scholarships, housing, health, and infrastructure
Cultural & Historical Context
BDI applies cultural and historical context, informed by ongoing research, to interpret disclosures that are often misclassified or overlooked.
Unlike traditional corporate analysis that treats these references as “marketing,” BDI recognizes cultural engagement as verifiable data when tied to concrete actions, resources, or outcomes.
Transparency Standards
We track six disclosure-level transparency signals that indicate the quality and comprehensiveness of corporate reporting, independent of the actions themselves.
EEO-1 Reporting
Disclosure of detailed workforce demographics by job category
Disaggregated Data
Race-specific breakouts vs. aggregated “diversity” metrics
Multi-Year Data
Historical trends enabling year-over-year analysis
Third-Party Verification
Independent assurance of reported data
Supplier Disclosure
Quantified diverse supplier spend or contracts
Community Disclosure
Quantified community investment or geographic targeting
What We Don’t Do
No Inference from Geography Alone
We do not classify place-based investments as benefiting Black communities unless the company explicitly identifies Black residents, Black impact, or the location appears in BDI’s vetted list of majority-Black areas. Generic terms like “urban,” “inner-city,” or “underserved” without racial identification are classified as diversity signals, not Black-specific commitments.
No Credit for Stated Intent Without Action
Commitments and promises are documented as “inconclusive” until outcomes are reported. Companies receive full recognition for demonstrated results—and full accountability when results fall short of commitments.
No Conflation of Scale and Quality
A “systemic” action means company-wide reach—not inherently positive impact. We separately assess scope (how far it reaches) and direction (whether outcomes are beneficial, harmful, or neutral). Broad policies can have negative outcomes; limited programs can deliver measurable benefits.
Evidence-First Principle: Every classification must be supported by verbatim text from the source document with page-level citation. We do not extrapolate, assume, or fill gaps in corporate disclosure. If a company does not report it, we do not claim it.
Limitations & Boundaries
Data Availability & Reporting Quality
Our analysis is constrained by what companies choose to disclose. Lack of evidence in our database does not prove absence of action—it reflects absence of disclosure. Companies with comprehensive, detailed reporting enable more thorough analysis than those with vague or limited disclosures.
AI-Assisted, Human-Verified
We use AI to accelerate extraction and classification, but human researchers review outputs, resolve edge cases, and validate demographic routing. AI assists; humans decide. Classification accuracy depends on both model performance and researcher judgment.
Temporal Boundaries
We analyze current-year reporting. Historical actions documented in older reports are not systematically re-ingested unless referenced in current disclosures. Trend analysis requires multi-year data, which not all companies provide.
Continuous Improvement
BDI’s methodology evolves as corporate disclosure practices change and as our understanding of equity measurement deepens. We refine classification protocols, update cultural context, and incorporate community feedback to ensure our analysis remains rigorous, relevant, and accountable to the people most impacted by corporate equity commitments.
Major methodology changes are documented and disclosed. Classification updates may cause variance in historical comparisons as we improve accuracy and consistency.