Worth AI

Probabilistic matching resolves inconsistent business records, and a scoring layer weights a large signal set into one credit score. A business database joined to public filings sits beneath and would sell as an identity and data service without the models.

Sampling of automatic approvals and a reconsideration route for a declined business are not described, but the rest of the control structure is. Applications that meet the institution's thresholds are approved or declined automatically, and edge cases route to manual review with the context already surfaced. The institution sets approval thresholds, required data points, escalation triggers and routing without engineering, and every agent acts only inside the authority it configures. Each decision carries reason codes, agent decisions are traceable and overridable, and CODI lets underwriters question the reasoning behind an agent's work.

An explainable credit model is named as a component, but no error rates, model documentation or validation summary are published. A 98 percent match rate carries no method or comparison basis.

PatientFI says Worth's merchant onboarding and monitoring tool has cut its onboarding time by nearly 50 percent, and Priority's head of operational risk credits it with faster, more reliable merchant onboarding. Across customers, Worth states average gains of more than 37 percent in approval rates and 43 percent less abandonment, over a business database of more than 350 million profiles.

Outputs are said to grow stronger with every decision, which presents cross customer learning as a benefit with no boundary, anonymization step or opt out. A bank cannot tell what its decisioning contributes to scores served to competitors.

The privacy policy effective January 22, 2024 covers website visitors and customers in general terms, keeping service data for the life of the relationship. Beneficial owner and applicant data, bank records and tax filings moving through onboarding fall outside what it describes.

Worth states SOC 2 Type II compliance, audited against the Trust Services Criteria for security, availability, confidentiality, integrity and privacy, and offers the report on request. The auditor and audit period are unnamed. Its program follows the National Institute of Standards and Technology Cybersecurity Framework, with third party penetration testing, multifactor authentication and logged, time stamped platform activity, and its controls are mapped to ISO 27001 without a certificate.

Worth names no supervisory instrument as a design target. Customer identification and beneficial ownership checks sit under money laundering obligations, and IRS tax data arrives through a gated arrangement rather than a scrape.

More than 1,200 data points per business feed Worth's scoring and underwriting agents, which is where proxy variables hide. Fair lending testing, disparate impact study and adverse action documentation are all unpublished, and an economic equity mission has no evaluation behind it.

A declined business has no described route to dispute an underlying record, and Worth publishes no guarantee or remediation term. What a lender does get is a case file with the evidence behind it and reason codes on every approve, decline or review outcome, which give it grounds to challenge an outcome.

Secretary of state filings, IRS records and web sources feed the platform, with Equifax, GIACT and LexisNexis among named data sources and AWS and Google in the privacy policy. Providers for the scoring, matching and agent components are unnamed, and subprocessors are unlisted.

Worth names no lender core banking, origination or case system among its connections, which include QuickBooks, Xero and Alloy. Public application programming interface (API) documentation covers onboarding, scores, ownership verification and webhooks, and software development kits put a white labeled experience on top.

Storage in the US is stated for website and relationship data, not for the platform itself. Hosting regions, residency options and transfer mechanisms for onboarding data are undescribed.

Consolidation is the pitch, replacing several verification, fraud and credit tools with one system, yet no price is published to set against the licenses displaced.

Banks, credit unions, lenders, fintechs, software vendors and payment processors each get their own treatment, buying one workflow from identity through credit to portfolio monitoring. An international platform extends onboarding, identity verification and fraud checks to more than 200 countries and territories, though the deepest sources, Secretary of State filings and IRS records, are American.