Boost Capital

Identity parsing with liveness and facial recognition, business document validation and agent based fraud checks are model work. Distribution is the wedge, through messaging onboarding without an app download, and a chat funnel with manual checks behind it would still run without the models.

Loan officers make the credit decision from the eKYB dashboard, so verification is automated while the decision sits with people. Cited confidence levels are not tied to a threshold that routes a case, and an applicant's experience when validation fails is not described.

Document reading above 98 percent and fraud detection above 95 percent on Philippine documents are claimed, with a homepage figure above 99 percent. None carries a method, test set or false rejection rate, and the facial recognition figure is a confidence level rather than a measured error.

Institutions named include GCash, Wing Bank and BDO Network Bank. The one outcome figure, a client bank growing its applicant pool by 30 percent after a dashboard insight, names no bank and gives no period.

Applications are cross referenced against external databases and patterns, which are typically learned across an applicant base spanning competing lenders in the same markets. Whether fraud signals from one institution inform another's decisions, and whether applicant data persists after a decision, are not stated.

The privacy policy is written for Boost's own loan services, keeping data up to five years, and does not say how it applies when onboarding runs under a bank's brand. Identity documents and selfies also travel through consumer messaging platforms the bank does not control, and what those platforms retain is not addressed.

Bank grade, ISO certified security is claimed without naming the standard, certifying body or scope. Static and dynamic application security testing is a concrete practice, but an unnamed certification invites an assumption a buyer cannot check.

Five markets differ on whether remote onboarding is allowed, what evidence satisfies due diligence and whether verification may run over third party messaging. No regulator, statute or rule is cited for any of them.

Onboarding through messaging apps avoids excluding people with older handsets or expensive data, and the dashboard helped one bank partner find it was losing a third of its female business owner applicants. Facial recognition, liveness and fraud flags reach first time borrowers with no accuracy breakdown by population, gender or document type.

Flagged applications, for fraud or mismatched details, go to manual review, so a wrong flag gets a human look before the outcome is final. A rejected applicant is not stated to receive a reason, the privacy policy's correction rights cover only Boost's own loans, and no guarantee or accuracy commitment is published.

VerifyIQ is the company's own document processing system, reading government IDs such as PhilSys and bank statements from BDO and GCash among others. Its video annotation model, the money laundering databases checked and the hosting provider are unnamed.

A development kit, an API or a white labeled app can drop into an institution's own channels. Named connections are the Instafin core banking system through Oradian and Brankas for open banking, though public API documentation is not linked.

Distributed infrastructure chosen for reliability is the only hosting description. Provider, region and residency go unstated across five countries, several expecting customer data to stay in country.

Delivery comes in several forms, which usually implies a per component price, yet no price, packaging or basis of charge is published. Engagement runs through contact.

Banks, digital banks and microfinance institutions are served along with e wallets, across Southeast Asia from a Singapore base. One funnel reaches loans, savings and payments, but only small business lending in the Philippines has dedicated material.