Bizbaz

Inference from alternative signals is the only thing sold, since the customers have no bureau file. No data product, workflow tool or distribution asset sits underneath the scoring model.

Lenders make the decision, but what a credit officer sees, whether contributing factors are visible and whether an assessment can be overridden are not described. Inferred traits are hard for a credit officer to check, which raises the stakes of that missing review path.

Personality and voice based assessment are not established credit techniques, and accuracy is given for neither. A validator would have nothing to test.

RCBC, a national retail bank in the Philippines, announced a pilot of the small business and underbanked credit tool in November 2023, quoting its chief executive. Revenue up two to three times, acquisition costs down 40 percent and defaults down 50 percent are claimed for clients including an unnamed Indonesian bank, with no period or method, and no move from pilot to production has been announced.

One profile serves both a lender's credit decision and a recommendation engine that sells clients' customers further financial products. Whether a profile built for one client is reused for another, or what happens to it when a relationship ends, is not said.

Health, biometric and voice inputs are special category data under the GDPR and several newer national laws in its markets, yet it publishes no privacy posture, retention schedule or consent framework.

Biometric templates and voice recordings cannot be reset like a password, so a breach is permanent for borrowers with few practical remedies. The firm names no certification, attestation or testing program.

Licensing is not addressed in any market, nor any registration or supervisory relationship. Producing a credit assessment on an individual engages credit reporting rules in several of its markets, including a subject's right to see and dispute it.

Each input is a proxy problem: health tracks disability and age, voice carries gender and regional origin, and physical appearance has no place in a financial assessment. Positioned on financial inclusion, the firm leaves fair lending testing, outcome analysis across borrower groups and adverse action explanations all unpublished.

A borrower assessed on inferred traits cannot know which contributed, verify them or correct them, because they are inferences rather than facts on file and the applicant never deals with the firm. Who bears the loss when a score is wrong is never addressed.

Advanced Human Imaging, listed in Australia and the United States, disclosed a 2022 agreement letting its face scan and body measurement SDKs be integrated into the platform. Its scoring engine, voice assessment and personality inference have no named provider, and that one supplier surfaced through its own filing rather than the firm's disclosure.

Integration of the scoring and risk models is described in the bank's pilot announcement, implying API delivery into an origination flow. Origination systems, bureaus and telecom data partners are not named, and how the product reaches ecommerce and telecom channels is undescribed.

Several of its markets restrict cross border transfers of personal data, and no hosting arrangement, region or residency commitment is published. Staff sit in Singapore, Israel and Vietnam, among other countries.

Whether scoring costs per assessment, per approved loan or by subscription is never stated. In thin margin emerging market lending, cost per assessment decides which loan sizes are economic.

Banks, fintech lenders and telecom operators are among the buyers, across markets including Indonesia, the Philippines and Vietnam, with separate profiles for individuals and small enterprises. Material for each buyer type is not evident.