Worth AI

Models do real work in the scoring layer, resolving inconsistent business records probabilistically, weighting more than eleven hundred traditional and non traditional signals into a single credit score, and driving agentic decisioning that clears or routes a case. The foundation beneath is a data asset and a matching engine: a database of hundreds of millions of businesses joined to secretary of state filings, federal tax records and other public sources.

Apply the removal test and a business identity resolution and data service survives, which is a saleable product, so this sits with the data and orchestration vendors rather than the model native ones.

The automation boundary is stated in operational terms, which most vendors avoid: the system auto clears cases that match policy and routes edge cases to underwriters with the context already assembled, so a human sees exactly the decisions that need judgement and arrives with the file prepared. Outputs are confidence weighted, case management is configurable, and the company states that every decision is auditable and traceable for compliance purposes.

What is missing is assurance over the automated side, with no sampling or audit of auto cleared approvals described, no stated confidence threshold at which routing triggers, and no reconsideration path for a rejected business.

An explainable credit model is named as a platform component and decisions are described as auditable and traceable, which is the right intent for output that determines access to capital. None of it is evidenced. No accuracy figures, no error rates, no model documentation, no validation summary and no stated support for a lender's own validation were located, and the claim to the industry's most accurate business match rates carries no methodology or comparison basis. For a product whose central artifact is a proprietary score, the absence of any published performance measure is the gap.

Nothing here measures the product. A thirty million dollar funding round, a named credit bureau partnership and a database of hundreds of millions of businesses describe scale and backing, and the founding team's prior success in payments explains the confidence, but no financial institution is named as a customer, no customer count is published, and no outcome is quantified anywhere: no approval rate lift, no time to decision, no reduction in manual review measured at a named institution. The database figure is also inconsistent across sources, appearing as 242 million businesses in one place and more than 350 million in another.

One phrase deserves testing rather than accepting: the platform is described as delivering confidence weighted outputs that grow stronger with every decision, which states cross customer learning as a benefit while defining no boundary, no anonymisation and no opt out, so a bank cannot tell what its own decisioning contributes to scores served to competitors. A named credit bureau partnership indicates part of the data chain. What is absent is any evaluation of the scoring, any model provider disclosure, and any account of how the non traditional signals are selected or retired.

Although the subject is a business, the platform reaches individuals directly through customer identity verification and beneficial owner checks, and it draws bank and tax information alongside more than eleven hundred data points per business from traditional and non traditional sources. No published privacy framework, retention schedule, subprocessor list or statement on how owner level personal data is handled differently from entity data was located.

No trust centre, enumerated certification list, attestation scope or audit period was located in this pass. The platform holds business identity, beneficial owner, banking and credit data and sits inside onboarding decisions for regulated institutions, which is the profile that triggers a formal vendor risk assessment, so attestations have very likely been supplied privately to the banks and credit unions using it.

Worth supplies technology and holds no licence, the expected posture. Its regulatory grounding runs through the workflow rather than through claims: customer identification and beneficial ownership verification sit under anti money laundering obligations, business verification draws on federal tax records which implies a gated data arrangement rather than a scrape, and decisions are described as auditable and traceable to support regulatory requirements. Planned frameworks for verifying autonomous agents anticipate a regime that does not exist yet. No named supervisory instrument is identified as a design target.

Fairness is not incidental here, it is the founding pitch: the company launched on a promise to increase data transparency and fuel economic equity, and states that its platform eliminates biases by processing thousands of traditional and non traditional data sources. No fair lending testing, demographic performance analysis, disparate impact study, adverse action reason documentation or independent audit was located to support any of it.

Small business credit scoring built on non traditional signals is precisely where proxy variables hide, and lending data collection rules are extending supervisory attention to exactly this segment. Claiming to eliminate bias while publishing no fairness evidence is a weaker position than saying nothing, because it invites a buyer to rely on it.

One feature here is genuinely unusual in this index and pulls the grade up. The score is surfaced to the business it describes through a separate consumer facing brand, framed as helping a company understand its financial position and improve its creditworthiness, so the party being scored can actually see the number rather than merely suffering its consequences. Almost no scoring vendor assessed here does that.

What still does not exist is redress: no accuracy guarantee, no remediation term, no published error rate, and no described route for a business to dispute an underlying record or challenge a decline built on it.

Data provenance is described with more specificity than most vendors manage, naming secretary of state filings, federal tax records and web search sources among dozens of others, characterising the score as drawing on more than eleven hundred traditional and non traditional data points, and disclosing a partnership with a major credit bureau, so a buyer can identify significant parts of the chain feeding a decision. What is not published is the model layer or the fourth party register: no providers are named for the scoring, matching or agentic components, and no subprocessor list exists.

Delivery is designed for embedding rather than for a separate console, with software development kits and interfaces intended to fit an existing stack and a white labelled experience so the institution's customer never leaves its brand. The platform positions itself as orchestrating workflows across systems specifically to reduce the number of vendors a risk team manages, which is the right argument for a consolidation play.

What was not located is named connector detail: no core banking, loan origination or case management platforms are identified individually, and no public developer documentation or status page was found.

Delivery is cloud hosted software as a service. Global coverage is claimed for the business database and a one click global onboarding experience is described as the direction, while the identity resolution sources named are domestic state and federal records, so the practical footprint and the stated ambition differ. No hosting regions, residency options, transfer mechanisms or subprocessor locations were located.

No rates, tiers, billing unit or minimum were located. The platform is sold on consolidation, replacing several point tools for verification, fraud and credit with one system, which is a cost argument a buyer cannot evaluate without a price to set against the licences being displaced.

Buyers are named across the relevant institution types including banks, credit unions, fintechs and payment providers, and naming credit unions separately matters in a lane where most vendors ignore them. The functional scope is one workflow, small business onboarding and underwriting, covered end to end from identity through credit and fraud to continuous portfolio monitoring. Coverage is domestic in substance despite global language, since the crosswalking sources are United States state and federal records.