SecureLend

The removal test leaves nothing, since both the product and its distribution assume language models. Agents ingest decks, data room contents, emails and documents, structure them, run underwriting pre-checks against the institution's own rubric, perform quantitative analysis including burn multiples, debt service coverage and loss ratios, and draft investment and credit memos.

The origination system is described as language model native rather than a conventional platform with models attached, and the company reports processing roughly ten times faster than manual workflows on pilot averages.

Output is positioned as work a professional finishes rather than decisions taken: pre-checks run against the institution's own rubric rather than a vendor standard, memos are described as drafted, and the marketplace qualifies and routes borrowers toward underwriting rather than approving them. Memos are cited, so a reviewer can follow a claim to its source. Held at B because no review requirement or escalation rule is published, and the borrower-facing qualification step inside an assistant conversation happens without any described human involvement.

Two things support verification. Memos and analyses are described as cited, meaning a claim can be traced to its source document, which is precisely what supervisors have begun asking for where models touch credit decisions. And the company qualifies its own headline figure as based on pilot averages rather than presenting it as established, which is a small honesty that most vendors here do not extend. Held at B because no accuracy, extraction error rate or validation result is published, and the pilot base is not sized.

No customer, lender or institution is named anywhere, the single performance figure is explicitly qualified as based on pilot averages, and the strongest apparent third-party endorsement, a comparison naming this platform the only self-serve option in its category, was written by the company's own founder and is therefore not independent. What is verifiable is presence rather than adoption: a live listing in a major assistant's app store, a published open-source integration server, an accelerator place and a product launch.

No boundary statement was located. The marketplace qualifies borrowers and routes them across a database of more than 200 lenders, so the company observes demand, qualification outcomes and conversion across competing institutions simultaneously, which is commercially valuable and unaddressed. Nothing states whether borrower or deal material submitted to the agents informs models, or how long it is retained.

No data protection agreement, retention schedule, subprocessor list or consent framework was located. The platform ingests borrower files, data rooms and financial documents, and its marketplace captures borrower intent expressed inside third-party assistant conversations, which raises an additional question about what the assistant operator sees and retains that is not addressed anywhere.

No attestation, certification, trust centre or enumerated control set was located. The open-source integration layer allows technical inspection of the connection surface, which is a partial substitute for disclosure and not the same as an audited control environment, and a community bank placing borrower files with the platform would require the latter.

No regulator, statute or rule is named. Compliance checks and compliance file generation are listed as capabilities without indicating which requirements they address, and the marketplace model raises questions about borrower disclosure and referral arrangements in consumer and small business lending that published material does not reach.

The routing layer is where the exposure sits and it is undisclosed. A borrower expressing intent inside an assistant conversation is qualified by a model and shown a subset of more than 200 lenders, and nothing describes how that subset is chosen, whether commercial arrangements influence ordering, or whether qualification performs consistently across borrower types.

Because the interaction happens inside a conversational interface rather than an application form, the borrower is least likely of all to recognise that a consequential filtering decision has occurred.

No guarantee, indemnity or correction process was located. The borrower qualified and routed inside an assistant conversation has no described route to learn why particular lenders were shown, to correct information the model inferred from the conversation, or to reach a person, and the informality of the channel makes it least likely that any of this is apparent at the time.

Model agnosticism is the stated architectural principle and is presented explicitly as preventing vendor lock-in, so a buyer can change the underlying model rather than inheriting the vendor's choice, which is the property supervisory guidance on third-party models effectively rewards. Cloud services are named and the integration server is open source and therefore inspectable.

Held at B rather than higher because, unlike the one vendor in this index that names its model providers outright, the agnosticism is asserted without identifying which models are actually in use by default.

The integration layer is unusually open. An open-source server implements standardised schemas for loans, banking, cards and insurance and exposes twenty tools, ten prompts and thirty-two resources, so the connection surface is publicly inspectable rather than described, and agents reach into existing workflows through that protocol. More than 200 lender integrations sit behind the platform and deep integration into legacy core banking systems is claimed. Held at B because not one core banking, origination or servicing system is named individually, so the legacy integration claim rests on assertion.

The infrastructure is published rather than described, with an architecture diagram naming the cloud provider and its individual services for storage, compute, interface management and object storage, which is more specificity than nearly any vendor in this index offers. Held at B because no region, residency commitment or private deployment option is stated, and processing that occurs inside third-party assistant platforms sits outside that architecture entirely.

The pricing structure is disclosed in more detail than almost anywhere else in this index, covering a free allowance of fifty deck pre-checks a month, per-check pricing beyond it, per-page document intelligence, per-run quantitative analysis, per-output memo generation, a seven day free trial requiring no card, and consumption pricing on funded volume for the marketplace. That a buyer can begin without a sales conversation is itself commercial transparency. Held at B because the structure is published without the amounts, so a lender still cannot calculate cost at volume.

Buyer coverage is unusually varied for a company this size, spanning venture investors, commercial lenders, private credit funds, reinsurers and community banks, with the origination system aimed specifically at community institutions and the agent products launched first for venture underwriting. A database of more than 200 lenders sits behind the routing marketplace. Held at B because breadth across such different buyers is stated rather than evidenced, and no geographic footprint beyond a United States base appears.