Casca

The removal test leaves the ninety day manual close the founder describes as the problem he entered the market to solve. Agents are embedded across the origination chain rather than added at one step, automating more than 100 manual actions, reading tax returns, bank statements, financial statements and rent rolls in minutes, and running over 40 credit and know your business checks. The company describes itself as an AI-native origination platform replacing legacy infrastructure rather than accelerating it.

Human involvement is stated in the independent award citation rather than only in marketing, with agents automating the process while keeping people in the loop, and the company frames its approach as responsible artificial intelligence with bank-grade underwriting. The borrower-facing assistant answers questions and sends reminders rather than deciding, and lenders receive structured information and a centralised view of each applicant, which positions output as decision support. Held at B because automating up to 90 percent of lending workflows leaves the boundary undescribed, and no threshold, escalation rule or approval checkpoint is published.

Verification breadth is the substantive control, with over 40 credit and know your business checks performed per application and more than 30 native integrations to external data providers, so conclusions rest on corroborated data rather than document reading alone. Human review remains in the process and the company positions its underwriting as bank-grade.

Held at B because no extraction accuracy, decision error rate or validation result is published for a system reading thousands of documents and automating up to 90 percent of the workflow at institutions whose examiners will ask exactly that question.

The customer set is close to definitive for its market: the country's leading Small Business Administration lender by dollar volume and the largest originator by loan volume both run the platform, alongside the first customer bank and a further specialist lender, and all three flagship banks invested in the company rather than merely buying from it. The 29 million dollar Series A was led by a venture firm backed by more than 70 financial institutions.

Outcome figures come through an independent award citation rather than only from the company, covering up to 90 percent workflow automation, processing cut from months to one to four days and a 312 percent increase in lead conversion. Three separate industry awards were won in 2026.

No boundary statement was located, and the shareholder structure sharpens it. The two largest lenders in the same government programme are both customers and investors, competing directly for the same borrowers, and the platform observes application flow, pricing and outcomes across both. Nothing states whether data is isolated per institution, whether models learn across the customer base, or what an investor bank can see that a non-investor bank cannot.

No data protection agreement, retention schedule, subprocessor list or deletion commitment was located. Small business lending requires the owner's personal tax returns and bank statements alongside business records, and with more than 30 data provider integrations the platform also distributes applicant information outward, none of which is described.

An independent industry guide states the company is certified against the recognised service organisation control standard, which is an attestation held rather than a standard aimed at, and that is the assurance a bank's supplier review begins from.

Held at B rather than higher because the certification is reported by a third party rather than published by the company itself, no report type or scope is stated, and no trust centre, penetration testing summary or enumerated control set was located, which a community bank conducting its own first vendor assessment would need.

The platform operates inside a named federal programme with its own eligibility, documentation and guarantee requirements, and its customers are FDIC-insured institutions whose examiners review origination files, so compliance is structural rather than incidental. Know your business checks are built in at volume.

Held at B because no regulator, rule or programme standard is mapped explicitly in published material, and nothing describes how the platform keeps pace with programme rule changes that determine whether a guaranteed loan remains guaranteed.

The access argument is economic and precisely stated rather than asserted: smaller loans require nearly the same document collection, verification and underwriting work as large ones, which makes them inefficient to offer and pushes owners toward faster, higher-cost alternatives, so removing that fixed cost makes small-dollar lending viable for banks and keeps borrowers out of expensive substitutes.

Evidence supports it, with one named bank originating 56 million dollars through its small-dollar programme in a quarter and projecting annual originations above 750 million. Held at B because government-guaranteed small business lending has documented disparities in approval rates across borrower groups, and nothing published addresses whether automated origination narrows those gaps or reproduces them at speed.

No guarantee, indemnity or correction process was located. The small business owner is the affected party and is not addressed: an applicant declined after automated document analysis and dozens of external checks has no stated route to learn which check failed, to correct inaccurate third-party data, or to reach a human, and the fifteen minute application that makes access easier also makes the decline faster.

The data chain is characterised by count rather than content, with more than 30 native integrations to unnamed best-in-class providers, and no bureau, verification service, banking data aggregator or model provider is identified. For a platform whose decisions rest on external checks, which providers supply them determines both coverage and failure modes, and a count alone does not let a buyer assess either.

More than 30 native integrations with external data providers are claimed, which is a substantial connected footprint and the mechanism behind both the verification breadth and the automatic pre-fill that lifts conversion. The borrower-facing application is a first-class part of the product rather than a portal bolted on, completing in under fifteen minutes with pre-filled applicant information. Held at B because not one integration partner, core banking system or loan servicing platform is named, so the count is asserted rather than evidenced.

An independent industry guide states that borrower data is held in the United States, which is a residency commitment in substance and the answer a federally supervised lender needs before placing guaranteed loan files with an external platform.

Held at B because no hosting provider, region detail or private deployment option is named, the commitment is reported by a third party rather than published by the company, and the location of the more than thirty external data integrations in the decisioning path is not addressed.

No pricing, packaging or basis of charge was located. The company's own argument is that unit economics determine whether small loans get made at all, which makes its own cost structure directly relevant to the claim, and nothing indicates whether charge follows applications processed, loans funded or institutional licence.

Buyers span FDIC-insured community banks, regional banks, nationally leading government-programme lenders and non-bank lenders, with named institutions ranging from a 3.4 billion dollar asset bank to a 15.3 billion dollar one, so the platform is evidenced across genuinely different institution sizes. Coverage is deliberately concentrated on small business and government-guaranteed lending in a single country, with commercial lending adjacent, so depth is bought at the cost of breadth.