Aloan

The company articulates the distinction this axis tests better than any vendor in the index: platforms built around document analysis and reasoning as the core capability, versus workflow systems that pre-date the shift and added AI features to defend an installed base.

Its own worked examples make it concrete, with spreading as the structured output of a model that has read the whole tax return rather than an optical character pass an analyst stitches together, and multi-entity K-1 tracing running as a reasoning task rather than a workflow tab the analyst opens. Remove the models and nothing remains but the manual process it replaces.

The oversight position is built into how the benchmarks are stated, since the saved time is described as converting analyst preparation into analyst review rather than removing the analyst, and eligibility work surfaces ownership structures, affiliate identification and cross-document inconsistencies for lender review rather than resolving them. Guidance published alongside the product is explicit about automating global cash flow rollup without losing human credit judgement. Held at B because no mandatory review step or approval gate is specified, and a memo produced in thirty minutes leaves the checkpoint placement to the institution.

Source attribution is the core design rather than a feature, with every calculated figure mapping to its originating document through click-to-source citation, which is precisely the evidence supervisors have begun asking for where models touch credit decisions. The company also identifies model documentation obligations as a reason institutions buy rather than build, showing awareness of what the validation function will require. Held at B because no extraction accuracy, error rate or validation result is published for a system reading tax returns and reconciling K-1s across related entities.

The operational benchmarks are unusually concrete and honestly framed, describing spreading falling from two to four hours of analyst time to ten to twenty minutes of analyst review, multi-entity tax packages that took half a day being ready in the morning, memo first drafts moving from three to five days to same day, and analyst capacity roughly doubling, with the company explicitly labelling these as changes credit teams report in production rather than vendor abstractions. Held at C because no institution is named, no deployment count is given, and the platform launched in March 2026, so the record is early whatever its quality.

No boundary statement was located. The platform processes borrower financials for community banks and credit unions lending into overlapping regional markets, and its service organisation channel means several institutions may sit under one deployment with standardised formats across them. Nothing states whether document understanding improves from customer material, or how one institution's borrower data is separated from another's.

No data protection agreement, retention schedule or subprocessor list was located. The platform ingests complete tax returns, personal financial statements and guarantor schedules, which is among the most sensitive document sets in commercial lending, and it operates a branded borrower portal collecting them directly from applicants, so borrower-facing handling terms matter and are not published.

No attestation, certification, trust centre or enumerated control set was located. That is a conspicuous gap for a platform positioned on examination readiness, since a bank preparing for its own examination will be asked about the control environment of a vendor holding complete borrower tax and financial records, and the company publishes examiner guidance on everything except itself.

The regulatory mapping is the most specific in the lending category and names both the supervisors and the individual guidance. Audit trails are built to hold under national bank, deposit insurance and state examination, with credit union supervision handled separately for the service organisation channel, and the company publishes an examiner readiness guide against model risk guidance and two named bulletins.

Government-guaranteed lending is covered at the level of the actual forms and programme distinctions, mapping use of proceeds between the two main loan programmes back to the specific application form. Small business lending data collection under the consumer bureau's rule is addressed directly, which no other vendor here has done.

Fair lending is addressed through a specific and unusual mechanism: the small business lending data collection rule requires consistent demographic and business data on every application, and the platform passes demographic fields through cleanly while producing standardised application and decision documentation, so that if a fair lending review is initiated the underwriting reasoning is reproducible rather than reconstructed.

Reproducibility is the property a fair lending examiner actually needs. Held at B because no disparity testing, approval analysis by borrower group or fairness evaluation of the models themselves is published.

No guarantee, indemnity or correction process was located. The borrower submits documents through a branded portal and is unaddressed thereafter: nothing describes whether an applicant learns their financials were spread by a model, how a misread figure is corrected, or what recourse exists where an automated analysis contributes to a decline. Source citation serves the credit committee and the examiner rather than the applicant.

No base model, provider, hosting arrangement or subprocessor is identified. The founding team's background in shipping AI at four major technology companies is disclosed, which speaks to capability rather than dependency, and an institution cannot document whose models read its borrowers' tax returns, which is exactly what third-party model guidance expects it to know.

Five specific core banking platforms are named across all three major United States providers, covering the flagship bank and credit union systems of each, and the output is described concretely as the spread, analysis and memo attaching to the credit file inside whichever system the institution already runs, with no core replacement. Interfaces and webhooks push extracted data, spreads, risk flags and finished memos into existing workflow. Naming the specific core products rather than the vendors is what distinguishes this, because it tells a bank whether its own installation is covered.

No hosting provider, region, residency commitment or private deployment option was located. Implementation speed is published, at two to four weeks alongside an existing origination system with the vendor handling setup, policy configuration and integrations, which answers how fast rather than where.

Pricing is characterised as built for community banks and contrasted with full origination platforms on cost, without any rate, unit or structure published. For a buyer whose alternative is adding an analyst, the comparison that matters is cost per credit against salary, and nothing supports it.

The target band is stated precisely rather than vaguely, covering United States community banks and credit unions between 500 million and 25 billion dollars in assets, alongside non-bank commercial lenders and credit union service organisations serving multiple institutions. Product coverage spans commercial and industrial, commercial real estate, government-guaranteed and agricultural lending. Held at B because coverage is one country and one institution band by design, with no presence outside it.