CDFI / Underbanked Pack

Why this Data Set exists

Building a banking product for underbanked customers is a category-design problem, not a feature-set problem. The product can't assume Social Security Number, FICO score, employer-direct-deposit, or English-language interface. It has to work for someone whose payment history is on a prepaid card and whose income is irregular cash. Most product teams design for the modal customer first and 'extend' the design to underbanked segments later — which produces products that work for everyone except the segment they were ostensibly designed for.

The data problem compounds the design problem. Production banking data captures the customers who already passed the existing-product filters. The customers you're trying to design FOR are statistically absent from your training set. You need synthetic data calibrated against external sources (FDIC unbanked surveys, CFPB consumer studies, World Bank remittance corridors) that captures the patterns the production data omits.

This Data Set is that calibration. 80 households across the canonical underbanked profiles, with structured banking-relationship history (including the prepaid-card-only or cash-only past), remittance corridor patterns, cash-economy income proxies, and the inclusion-trajectory data — what does the path from prepaid to checking actually look like for someone who's been outside the banking system for a decade? The Data Set provides the realistic answer.

Use Cases

Who uses this Data Set

What's inside

The 80 households are drawn from six archetypes: F-04 First-Generation Wealth Builders, S-02 Bankruptcy Recovery, U-01 Unbanked / Recently Banked, U-02 Low-Income Working Families, U-03 Recent Immigrants, and N-04 Cannabis-Industry Workers. The mix is deliberate — about 30% currently unbanked or recently banked, 35% in the prepaid-card-dependent category, 20% post-ChexSystems looking for re-entry to mainstream banking, and 15% cash-economy participants whose income flows mostly outside formal banking.

Every household has structured banking-relationship history: account types currently held (checking, savings, prepaid, money-order purchases), account-closure history with reason codes, prior ChexSystems flags where applicable, and the 24-month trajectory showing whether banking-relationship age is increasing or decreasing. Remittance flows are structured by corridor (origin–destination country pair), frequency, channel (bank wire, MTO like Western Union or Remitly, increasingly crypto via stablecoin), and typical amount. Cash-income proxies use the structured methodology with confidence intervals: documented income from W-2 / 1099, plus an estimated cash-income component, plus an explicit uncertainty band.

The Data Set ships as JSON and CSV. The WealthSynth Methodology PDF documents the banking-access taxonomy (FDIC's six-category framework: fully banked, underbanked, unbanked, recently banked, recently unbanked, never banked), the remittance-corridor data sources (World Bank, FDIC, CFPB), and the cash-income estimation methodology with the IRS tax-gap calibration.

Preview a sample household

A redacted summary of one household from this Data Set — names, employers, exact balances, and metro area are stripped. Ages are bucketed, income and net worth are reported as bands. The full record (and all 80 like it) ships in the ZIP.

Technical Highlights

Sample Schema Fields

{
  "banking.account_type": <value>,
  "banking.prepaid_card_history": <value>,
  "banking.check_cashing_use": <value>,
  "income.remittance_outflows": <value>,
  "credit.thin_file_flag": <value>
}

Sample queries

households.filter(h =>
  h.banking.chexsystems_flagged &&
  h.banking.chexsystems_age_years > 5 &&
  h.banking.relationship_trajectory_24mo === 'improving'
)

households.filter(h =>
  h.income.remittance_outflows.monthly > h.income.documented_annual / 12 * 0.10
).reduce((acc, h) => {
  const corridor = h.income.remittance_outflows.primary_corridor;
  acc[corridor] = (acc[corridor] || 0) + 1;
  return acc;
}, {})

households.filter(h =>
  h.banking.account_type === 'prepaid' &&
  h.banking.prepaid_card_history.monthly_load_avg >= 1000 &&
  h.banking.prepaid_card_history.direct_deposit_present
)

households.map(h => {
  const total = h.income.documented_annual +
                h.income.cash_income_estimate.value;
  return {
    id: h.id,
    cash_share: h.income.cash_income_estimate.value / total,
    ci_band: [
      h.income.cash_income_estimate.ci_low / total,
      h.income.cash_income_estimate.ci_high / total
    ]
  };
}).filter(x => x.cash_share > 0.20)

Methodology

The 80 households are generated against FDIC-aligned banking-access categorisations, with archetype-specific distributions calibrated from the FDIC's 2023 Survey of Unbanked and Underbanked Households, the CFPB's consumer studies on cash-economy participation, and the World Bank's bilateral remittance flow data. Cash-income estimates use the same structured methodology as B23 — a sampling-based approach with confidence intervals calibrated against the IRS's tax-gap analysis. Banking-relationship trajectories are generated as 24-month sequences with realistic transition probabilities (a household moves from prepaid to checking with non-zero but small monthly probability conditional on the right preconditions). The corpus passes the WealthSynth consistency validator and LLM-as-judge gate, with additional review by an inclusive-finance specialist to verify the structures align with current FDIC, OCC, and CFPB guidance on underbanked-customer segmentation. Annual refresh tracks the FDIC's annual unbanked survey results.

Included Archetypes (6)

Frequently asked questions