Accend

The removal test leaves exactly the manual work the company measures against, analysts spending hours spreading and analysing statements by hand. Models parse full tax packages including individual, partnership and corporate returns with supporting schedules in minutes, standardise financials across three statement types, surface add-backs automatically, map data into cash flow models, generate credit memos and flag anomalies as they appear. Nothing beneath the models constitutes a product on its own.

The most complete oversight construction located in the commercial and small business underwriting lane, and each part is stated rather than implied. Every output is reviewed and validated by expert human analysts before the customer relies on it, so review is universal rather than exception-based.

The credit team can drill into sources behind any figure, override values, run their own ratios and leave notes, with every change tracked, so the human retains authority over the number and the audit trail records who changed what. Models are configured to the bank's own personal, business and global policies rather than the vendor's defaults, and the product is positioned to sit inside existing credit workflows rather than replace the decision.

The control set is strong and specific: structured output carries source traceability back to the originating document, data is described as audit-ready, every change made by a reviewer is tracked, and an anomaly detection feature flags issues as they appear rather than at review. Held at B for a precise reason.

The company guarantees 100 percent accuracy, but the mechanism delivering it is expert human review of every spread rather than model performance, and no error rate, correction frequency or accuracy figure for the artificial intelligence itself is published. A buyer therefore cannot tell how much human correction the guarantee is absorbing, which is the number that would determine whether the process scales.

Six customers are named directly, spanning a listed payments company, two large spend management platforms, a business banking provider, a chartered bank and a payables company, with a shared outcome figure of 80 percent reduction in application processing time. That is an unusually specific named customer set for a company at seed stage.

Funding is 3.2 million dollars from a venture syndicate including a well known accelerator and two established firms, with angels from three major fintechs, and the founders previously led product and engineering on the risk team at one of those customers and worked at two global banks.

No boundary statement was located. The platform holds borrower financials for lenders competing directly for the same commercial customers, and models that improve from spreading experience across that base raise the question of what one lender's document flow contributes to another's results. Nothing states whether client data is isolated, whether it informs model training, or what the expert reviewers see across accounts.

No data protection agreement, retention schedule, subprocessor list or deletion commitment was located, and the holdings are more personal than commercial credit usually implies. Small business underwriting routinely requires the owner's individual tax return alongside business filings, so the platform processes personal income, dependants and deductions for guarantors as well as company financials, and none of the handling terms are published.

No attestation, certification, trust centre or enumerated framework was located. Document collection is described as running through a secure link, which characterises one transfer step rather than the control environment. Several substantial fintechs and a chartered bank have completed supplier review before sending borrower financials through the platform, and none of that assurance is published.

No regulator, supervisory expectation or lending rule is named. The platform produces the credit analysis and memos that support commercial lending decisions and tracks covenants afterwards, all of which feed examination files at regulated banks, and nothing maps the output to the model risk or credit administration standards those banks are held to.

Borrowers are businesses rather than consumers, which reduces but does not remove the exposure. Spreading involves judgement about classification, normalisation and which add-backs are legitimate, and those judgements fall hardest on businesses with unconventional accounting, seasonal revenue or owner-operator structures where personal and business finances intermingle, which describes most small firms. No analysis of accuracy or classification consistency by business type, size or sector is published.

An explicit guarantee of 100 percent accuracy is published and repeated as the central commitment, which places this ahead of the great majority of vendors here, who offer nothing. It is also credibly constructed, since the company states the mechanism, expert review of every output, rather than asserting the models are simply correct.

Held at B because no remedy is described: nothing states what the customer receives if an error reaches a credit decision, and the borrower whose financials were misclassified has no stated route at all.

No base model, provider, hosting arrangement or subprocessor is identified. The human review layer is itself an undisclosed dependency, since the guarantee rests on it and nothing states whether reviewers are employees or contracted, where they work, or what they see. For a product whose output enters regulated banks' credit files, both the model and the review chain are provenance questions a buyer would raise.

Two intake paths are described, a secure link sent to borrowers for document upload with automated requests and reminders to chase completeness, and direct connection into accounting systems, which removes the collection friction the company identifies as where the process breaks down. Positioning is explicitly additive, sitting inside existing credit workflows so banks modernise without replacing systems. Held at B because no accounting package, core banking, loan origination or document management system is named.

No hosting provider, region selection, residency commitment or private deployment option was located. Commercial banks routing borrower tax returns and financial statements through an external platform would examine processing location during procurement, particularly where expert human reviewers form part of the service and their location is likewise undescribed.

No pricing, packaging or basis of charge was located. A free first financial spread is offered as an acquisition mechanism, which reveals the unit the service is delivered in without revealing what it costs, and for a product blending software with expert human review the split between licence and per-spread service fee is the question a buyer would need answered.

Buyers span commercial banks, commercial real estate lenders and fintech lenders, with insurers named as an affected segment, and the platform covers the underwriting chain end to end from borrower document collection through spreading, modelling and memo generation to post-close covenant tracking. Coverage is deliberately confined to business and commercial credit rather than consumer, and the tax form types named are United States specific, so no international footprint is evidenced.