Uplinq

The removal test decides it. Take the models out and what is left is connectivity to more than ten thousand small business data sources, which is a data feed rather than a decision, and a lender cannot underwrite from it. Turning market, community and environmental signals into a credit assessment that a lender can act on is model work throughout, and both the company and its card network partner describe machine learning as the mechanism. The data connectivity is a genuine moat and it is the input to the product, not the product.

The division of labour is stated plainly and repeatedly, which is worth more than most oversight language in this index. The founder's line is that the company does not lend and works with lenders to help them say yes more often, and the platform is positioned as designed to complement an existing credit assessment process rather than to replace it, so the credit decision and the risk stay with the institution. What is not described is where the platform's output enters the workflow, whether a lender can override a score, or whether any deployment runs it as a straight through decision.

The approval uplift claim is the whole model risk question and it is published without its other half. Approving five to fifteen times more applications is only a good outcome if losses on the newly approved loans behave, and no default rate, delinquency comparison, vintage curve, backtest or validation summary appears in any source located. Nothing describes model documentation for a lender's own validation function, which is what a supervised institution has to produce when it buys a third party model.

The strongest item is a joint case study published with a global card network, reporting a fifty percent reduction in underwriting cost, which is co authored evidence rather than a vendor claim. Read the headline figure carefully: the one point four trillion dollars is described as loans the technology has served as a foundation for over its full fifteen year life, not volume Uplinq itself has scored, and the approval uplift claims of five to fifteen times more approvals, and one interview headline of ninety five percent rejections becoming seventy percent approvals, are founder statements with no institution attached. Industry recognition includes a lending category award. No lender is named as a customer anywhere.

The commercial logic points at pooled learning, since a scoring platform serving many lenders improves as more outcomes flow back, and lenders competing in the same market would be contributing that performance data. Nothing published states whether a lender's application and repayment outcomes train models serving other lenders, whether that is optional, or what boundary applies. The same silence covers the alternative data itself, where nothing describes how the ten thousand sources were licensed.

A platform assembling billions of alternative data points about small businesses across a hundred and fifty countries raises questions this axis exists to test, and none are answered publicly. Nothing describes what is collected about the business owner as distinct from the business, which matters because small business credit routinely turns on the owner's personal profile and that pulls consumer credit reporting duties into scope. No lawful basis, no consent position, no retention schedule and no statement on cross border transfer were located.

Searched the company site, its press releases and partner announcements for an enumerated certification, an attestation, a penetration testing statement or a trust centre, and found none. A supplier handling applicant data for regulated lenders would ordinarily surface at least a service organisation control attestation during procurement, so the absence is a gap in public disclosure rather than evidence of absent controls.

No licence is required for a decisioning support supplier and none is claimed, which is the correct posture and is not penalised here. What is graded is clarity, and the available statement is a broad assertion that the underlying data sets have met every regulatory requirement in their local region, with no regime, statute, supervisor or register named anywhere. A vendor operating across a hundred and fifty jurisdictions asserting universal compliance without naming one requirement is making a claim a buyer cannot check.

Fairness is the company's stated purpose rather than an afterthought, with the mission framed around minority owned and protected class borrowers and the case study claiming the technology removes bias, and that engagement is the reason this is not the D given elsewhere for silence. It is still a claim rather than a disclosure.

No disparate impact testing, no approval rate comparison across groups, no search for a less discriminatory alternative and no account of how principal reason codes are produced when a score contributes to a decline. Scores feeding credit decisions sit inside equal credit opportunity duties, and the adverse action notice is exactly where an alternative data model becomes hard to explain.

The applicant never meets this vendor. A small business owner declined on a score that drew on community and environmental conditions has no way to know an external assessment contributed, no route to see what it said and no correction path, and nothing published describes one. On the institutional side there is no warranty, service level or stated allocation of responsibility if the model performs differently from the approval uplift the marketing describes. The recourse burden sits entirely with the lender.

One useful fact is stated and the rest is absent. The founder describes the technology as having been in market for over fifteen years before being repackaged under this brand, which means the models predate the company and were built for an earlier generation of lending, and a buyer should ask when they were last retrained.

Beyond that, nothing names the model components, any third party provider, the infrastructure the platform runs on, or which of the ten thousand data feeds are single source dependencies whose loss would degrade a score.

Two different kinds of integration and only one is documented. Data side connectivity is genuinely deep and specific, more than ten thousand direct connections into small business data sources across a hundred and fifty countries, which is the harder half to build.

Institution side integration is undescribed: no loan origination system, core banking platform or decisioning engine is named, and for a product that must sit inside an existing underwriting process that is the connection a buyer needs to see.

Nothing published describes hosting, tenancy or region. The residency question is sharper here than for most vendors on this axis because the platform draws data from more than a hundred and fifty countries and serves lenders across several of them, so where assessment happens and where the data rests is a live procurement question rather than a formality.

No pricing, no rate card, no billing basis and no indication of whether the platform is priced per assessment, per funded loan or by subscription. For a product sold as a supplement to an existing underwriting process, cost per assessment is the number that decides whether it pays for itself, and it is not published in any form.

The segment is narrow by design, small business lenders, and the geography is unusually wide for a vendor this size, with data coverage claimed across more than a hundred and fifty countries and a card network referring lenders in the United States and Asia Pacific to the platform. The founder describes the underlying technology as having served both very large and very small institutions. Held at B because that history belongs to the predecessor technology and no current institution is named.