Bizbaz

The scoring model is the entire product. There is no bureau file to fall back on for the population this serves, which is the founding premise, so the inference drawn from alternative signals is the only thing being sold. Remove the models and there is no data product, no workflow tool and no distribution asset underneath. Consistent with the whole alternative data credit cohort in this index, where Pave, Prism Data, Carrington Labs, credolab, Scienaptic, Zest AI, GiniMachine and Uplinq all hold A.

No oversight model published. The lender retains the lending decision, which is the standard arrangement in this category, but nothing describes what the score presents to a credit officer, whether the contributing factors are visible, whether a human can override an assessment, or what review applies when the model and a loan officer disagree.

For a product whose inputs include inferences a credit officer cannot independently check, such as an assessment of personality, the absence of any described review path is more consequential than for a conventional score built from verifiable financial facts.

The headline result is the strongest quantified claim in this pull and the least supported. A 50 percent reduction in bad debt defaults alongside a two to three times revenue increase would be a remarkable joint outcome, and it arrives with no named institution, no time period, no portfolio size, no baseline model and no vintage analysis.

The methods most in need of validation are the ones with none: no published accuracy exists for the personality based or voice based components, and these are not established credit techniques with a literature behind them, they are novel inferences being used to allocate credit. The company's framing that the current credit system is broken and obsolete is a fair description of the problem and does not establish that this is the solution to it.

One named institution with its chief executive on the record, which is the B bar. RCBC, a national retail bank in the Philippines, announced it would pilot the credit assessment tool for MSME, unbanked and underbanked lending, with president and chief executive Eugene Sering Acevedo quoted by name.

HSBC's venture investment arm is an investor, which is an independent party with money at stake, though a venture bet is a weaker signal than the operational commitments that earned bondIT an A. The quantified figures are real but unattributable: clients including an Indonesia based national bank are reported to have seen revenue rise two to three times with customer acquisition costs down 40 percent and bad debt defaults down 50 percent, and that bank is not named.

Two caveats belong on the record. The RCBC engagement was announced as a pilot in November 2023 and no confirmation of production rollout was found. And a 50 percent reduction in defaults alongside a two to three times revenue increase is a very large joint claim that no independent party has verified.

Nothing published on boundaries, retention or reuse. The structural question is specific to the business model: the same profiles that score an applicant for a lender also feed a product recommendation engine, and the company describes creating revenue for clients by upselling financial products to existing customers.

That means a single profile built from a person's health, lifestyle and social data serves both a risk decision and a marketing decision, and nothing states whether a profile assembled for one client's underwriting is reused for another client's targeting, or what happens to it when a lending relationship ends.

No privacy posture, retention schedule, consent framework or processing disclosure was found, and the categories of data involved make that omission unusually serious. The company's own description of its inputs includes health, lifestyle and social footprints alongside financials, and a 2022 master services agreement disclosed by its counterparty, the listed company Advanced Human Imaging, granted rights to integrate software development kits for body circumference measurement and face scan measurement into the Bizbaz platform.

Health information, biometric identifiers and voice recordings are special category data under the GDPR and under the data protection regimes now in force across several of the markets served, requiring an explicit lawful basis and heightened safeguards. This grade sits at the boundary of D and is held at C only because the axis convention reserves D for silence paired with a contrary claim, and no privacy claim was found here to contradict. A reviewer revisiting this vendor should consider whether D is the more accurate grade.

No SOC 2, ISO 27001, penetration testing programme or trust centre found. The gap is amplified by the data categories: biometric templates and voice recordings are not resettable in the way a password or account number is, so a compromise is permanent for the individual, and the affected people are low income borrowers in markets with limited practical remedies. A bank partner of RCBC's size will have conducted its own assessment, and as with several vendors in this pull, that evidence exists privately while nothing is published for anyone else to evaluate.

No licence, registration or supervisory relationship named in any market. The perimeter is genuinely complicated here rather than merely unaddressed, because the company operates across at least seven jurisdictions with divergent and rapidly changing rules on credit information, data protection and biometric processing, and several, including Indonesia, the Philippines and Vietnam, have introduced or tightened personal data legislation during the company's operating life.

Producing a credit assessment on an individual also engages credit reporting rules in several of these markets, which govern who may compile such assessments and what rights a subject has to see and dispute them.

The most invasive set of credit inputs recorded in this index, aimed at the population least able to contest an outcome, with no fairness testing published and an inclusion claim on top. The company's own account of the method is the evidence: risk profiles are built from financials, health, lifestyle and social footprints, and loan suitability is determined using personality based and voice based risk assessment. Each of those is a proxy problem in its own right.

Health status correlates with disability and age, both protected characteristics in most of the regimes involved. Voice carries gender, age, regional origin, ethnicity and speech difference, and speech systems are known to perform unevenly across exactly those dimensions. Social footprint analysis encodes community and network, which tracks ethnicity and class directly.

Personality inference is the pattern this index recorded at Psympl, where assigning a person a mindset they were never told about was graded D for the same reason. A counterparty's own shareholder disclosure adds face scan and body circumference measurement software development kits integrated into the platform, meaning physical appearance was contemplated as an input to a financial assessment. Proxy discrimination is a property of this method rather than a risk of poor implementation.

Against all of that the company positions itself on financial inclusion and broadening access for the unbanked, and no fair lending testing, outcome analysis across borrower groups, adverse action explanation or dispute mechanism was located. This is the FinBox pattern taken further: device metadata is invasive, but a borrower's face, voice, body and inferred personality is further still.

No warranty, service level or remedy published, and the recourse gap is at its widest here. A borrower assessed on inferred personality, voice characteristics, health signals and social footprint has no way to know which of those contributed, no way to verify them, and in most cases no way to correct them, because they are not facts held in a file but inferences generated about them.

The applicant never encounters this vendor, in the structural pattern recorded for Uplinq and Incandor, and here the gap is larger because the inputs are not documents the applicant supplied. The target population is by definition outside the formal financial system and therefore has the least access to complaint mechanisms, ombudsman schemes or legal remedy of any borrower group this index covers.

No model, provider or version named for the scoring engine, the voice assessment or the personality inference. One third party component is traceable, and only because the supplier disclosed it rather than the vendor: Advanced Human Imaging, a company listed in Australia and the United States, granted rights in 2022 for its body circumference and face scan measurement software development kits to be integrated into the Bizbaz platform.

That is a useful reminder that supply chain facts often surface from the counterparty's obligations rather than the vendor's disclosure. It also means a component assessing a borrower's physical characteristics was licensed from a third party whose own validation, accuracy across demographics and data handling terms the lending institution never sees.

Little integration detail published. The RCBC arrangement is described as the bank integrating the scoring and risk models, which implies API delivery into an existing origination flow, and the Advanced Human Imaging agreement describes software development kits for iOS and Android being embedded into the Bizbaz platform, which is the company consuming a component rather than integrating into a customer's stack.

No core banking platform, loan origination system, bureau or telecom data partner is named, and for a product distributed partly through e-commerce and telecom channels, how it reaches those systems is the load bearing question and is undescribed.

No deployment model, hosting arrangement, region or residency commitment published, and residency is a live legal question rather than a preference for this vendor. Several markets served have enacted data localisation or cross border transfer restrictions covering personal data, and the company's teams are distributed across Singapore, Israel, Vietnam, the Philippines and Malaysia, which raises the question of where personal, health and biometric data on borrowers in one country is processed and by staff in which other.

No pricing, model, unit or minimum published. Nothing indicates whether scoring is priced per assessment, per approved loan or by subscription, which matters for a product sold into thin margin lending in emerging markets where the cost per assessment determines which loan sizes are economic to originate at all.

Broad on both axes this measures, buyer type and market, and consistently described across several independent sources rather than only in company marketing. Buyers span four distinct categories: banks, fintech lenders, e-commerce platforms and telecommunications companies, the last two being genuine alternative distribution channels for credit assessment rather than incidental mentions.

Markets named include Singapore, Indonesia, the Philippines, Malaysia, Vietnam, Bangladesh and Thailand, with stated activity in Africa and Latin America, and teams in Singapore, Israel, Vietnam, the Philippines and Malaysia. The product line covers both individuals and micro and small enterprises through separate profiles, and extends beyond scoring into fraud detection, eKYC and product recommendation. That is wider functional and geographic coverage than most vendors in this category attempt.