Fintor

The removal test leaves the manual workflows and document chasing the company describes as the industry's current state. Agents are described as operating like employees with memory, planning and reasoning, working across whatever software a lender already runs rather than executing scripted automation, and covering intake, verification, document handling, compliance, closing coordination and post-close completeness. Capacity forecasting extends the same modelling to workforce planning.

Oversight is treated as a design commitment rather than a caveat. The company states that agents work around the clock while keeping humans in the loop for oversight and control, and human-in-the-loop mortgage operations appears as a named product surface alongside document processing and verification screens, meaning review is a place in the interface rather than a promise in copy. Output is described as fully auditable.

Held at B because no threshold, escalation rule or description of which decisions require sign off is published, and the agents are simultaneously described as handling routine processing autonomously.

The observability layer is the strongest part of the offering and it is unusual to ship as product: lenders can monitor agent performance, trace every decision, apply quality scoring, run comparison testing between approaches and observe continuous learning, all without code. Comparison testing of agent behaviour is the same experimental control found in the strongest lending platforms in this index, here applied to autonomous agents.

Adoption follows a stated sequence of connect, configure, learn and then scale based on measured results. Held at B because the headline quality claim, superhuman accuracy, carries no figure, baseline or validation of any kind.

No customer, deployment, volume or outcome figure was located. Total funding is around 8.7 million dollars with the last disclosed round in October 2022, meaning four years without a new round while the company repositioned from its original consumer real estate product to mortgage operations, and its own directory profile remains unclaimed. The product is clearly built and specified in detail, and nothing published demonstrates that a lender is running it.

No boundary statement was located, and the product description raises the question directly by promising that agents learn a lender's specific workflows and improve continuously. Nothing states whether that learning stays within one institution's tenant or whether patterns observed across lenders inform agents elsewhere, which matters because a lender's processing methods and exception handling are competitive practice rather than generic procedure.

No data protection agreement, retention schedule, subprocessor list or deletion commitment was located. Agents handle complete mortgage files containing income, asset, employment and identity documentation, and the platform is described as connecting to the origination system, customer system and contact centre, which means it touches borrower communications as well as documents. None of the handling terms are published.

No attestation, certification, trust centre or enumerated framework was located. Agents that operate inside a lender's origination system, customer records and contact centre hold effectively the same access as staff, which makes credential handling, permission scoping and activity logging central security questions, and none of the controls are published.

Compliance appears as a function agents perform and no regulator, statute or rule is named anywhere. Mortgage origination in the United States carries disclosure timing, fair lending and record keeping obligations that determine what an automated process may and may not do, and a platform whose agents execute compliance steps would ordinarily identify which requirements it is executing against.

No credit decision is claimed and the adapted exposure runs through processing. It is still real: verification agents assess whether submitted documentation is adequate, and applicants with irregular income, self employment, recent immigration or non standard documentation are the ones whose files most often fail automated checks and require exception handling.

If routine files clear automatically while unusual ones queue for human attention, service quality diverges along exactly those lines. No analysis of exception rates or error distribution by borrower type is published.

No guarantee, indemnity or correction process was located. The lender is well served by decision tracing and auditability, which allow any agent action to be reconstructed. The borrower has nothing described: someone whose documents are rejected by a verification agent, or whose file stalls in an automated step, is not stated to be told an automated system made that determination, has no described route to a person, and no means to challenge an agent's assessment of their paperwork.

No base model, provider, hosting arrangement or subprocessor is identified for agents described as having memory, planning and reasoning, which for a company of this size almost certainly rest on a third party model interface. That omission is more consequential here than for a scoring product, because borrower documents would traverse whichever provider sits underneath, and peers in this market have publicly described negotiating specifically over whether a model provider may train on customer data.

Integration targets are named by category and they are the right three, covering the loan origination system, the customer relationship system and the contact centre, which together hold the file, the borrower relationship and the communications an agent must act across.

The stated adoption terms are explicit about not requiring migration or replacement, and agents are described as able to work with any software and workflow rather than needing an interface built for them, which is what allows the claim of going live in days. No individual origination or servicing system is named.

No hosting provider, region selection, residency commitment or private deployment option was located. Exposure is domestic, and the platform processes complete borrower files while connecting into a lender's own systems, so whether agents run in the lender's environment or the vendor's is a question a bank's technology review would ask before the product could go live at all.

No pricing, packaging or basis of charge was located. The product is framed as a workforce rather than software, which usually implies pricing per agent or per unit of work completed rather than per seat, and nothing indicates which, nor how the observability and forecasting components are treated commercially.

Buyers are lenders and servicers in United States residential mortgage, which is one product in one country, and no institution type breadth beyond that is described. Functional coverage within that scope is genuinely wide, running from client intake through verification, documents, closing coordination and funding to post-close completeness, and extending into volume forecasting and operational performance tracking, which is broader than most single stage tools.