Trust-As-Collateral Credit Scoring
Turn invisible community trust signals into a measurable credit identity using AI on phone, video, and social data.
Trust-As-Collateral Credit Scoring reframes underwriting around the signals communities already use to extend credit informally: reputation, presence, consistency, and visible business activity. Instead of demanding paperwork the applicant cannot produce, it captures the digital exhaust they already generate — short-code SMS confirmations, a one-minute video of the storefront, and public social profiles — and converts that raw evidence into a structured financial identity.
The framework runs on three proprietary scores that complement each other. A telecom score parses bill-pay, order confirmation, and wallet SMS to infer income, spend, and disposable balance — open banking without a bank account. A storefront-video score replaces the costly in-person risk-officer visit by using computer vision and language analysis on a self-recorded clip to read inventory, location, business type, and willingness to pay. A social-media score measures engagement, profile depth, and posting cadence as proxies for follow-through and reputation.
Fused, these signals predict not only approve/decline but also amount, tenor, installment shape, and seasonal adjustment — enabling loan products tailored to each business rather than one-size-fits-all. Built deliberately on a balanced dataset (half women) so the model learns from the population it will serve, the framework converts what looks like absence of data into abundance of context.
- Trust is an invisible currency built over time, and the same reputational signals neighbors use to extend informal credit can be made legible to formal lenders.
- Models can only predict what they have already seen, so serving an underserved segment requires deliberately building a dataset that includes them rather than reusing legacy bureau data.
- The data needed to underwrite informal entrepreneurs already exists on their phones — it is simply in formats banks were never trained to read, not formats that are missing.
- A balanced training set (such as half women borrowers) is a precondition for fairness, because skewed data hard-codes exclusion no matter how sophisticated the downstream model.
- Good underwriting goes beyond yes-or-no to specify how much, when, and under what terms, so credit fits the borrower's cash-flow reality rather than a single off-the-shelf product.
Developed by Mercedes Bidart from a 2019 MIT Master's thesis pilot in Colombian informal settlements, after observing that Latin American shopkeepers already extended credit on reputation and that 99% of regional businesses — excluded by banks — leave rich digital traces on the phones they own.