Proximitty

The removal test leaves the manual servicing teams the company exists to shrink. Agents autonomously monitor portfolios, chase and ingest borrower documents, parse formats down to blurry scans and handwritten notes, reconcile discrepancies with borrowers directly, spread financials to each institution's own conventions and draft credit memos. Learning an individual lender's edge cases and assumptions rather than applying a fixed template is model work, and none of the lifecycle survives without it.

The oversight answer is unusually direct because the company answers the question explicitly rather than leaving it implied. Reliability rests on a governance layer where every agent action is observed, logged and reviewable, with human review configurable at any step rather than at points the vendor chose.

Agents run under scoped tools, approved templates and policy bound parameters, so the boundary is enforced by what an agent can reach rather than by instruction, and the whole is auditable specifically for model risk management functions and examiners. Naming the examiner as the audience is the tell that this was designed for a supervised institution rather than retrofitted.

The company builds for this axis deliberately, describing outputs as audit ready, agents as operating under approved templates and policy bound parameters, and the governance layer as supporting model risk management and examiner review, which means it anticipates the validation a bank must perform rather than resisting it. Spreading to the institution's own business logic keeps the analytical convention under the lender's control.

What is absent is measurement: no extraction accuracy, spreading error rate or covenant detection rate is published, and for a product parsing handwritten notes and blurry scans into credit memos, extraction accuracy is the number that matters most.

For a company founded in November 2025 with two employees the traction described is substantial: five bank and fintech customers with more than a billion dollars in delinquent loans on the platform, adoption claimed among publicly traded fintechs and banks, and a presentation slot at a major industry conference.

The outcome claim is specific and unusually well constructed, with one fintech reducing servicing staff from fifteen to two and redeploying the team to close twenty million dollars of new originations in a single quarter, which frames the saving as capacity released rather than headcount cut. Founders bring relevant background in fintech operations and financial data infrastructure. No customer is named, funding is around a million dollars, and an independent reviewer notes the traction claim needs validating.

No data boundary statement was located, and the product's central promise makes the question pointed. Agents learn the servicing rules, assumptions and edge cases specific to each institution, which is exactly the encoded institutional knowledge a lender would least want generalised to a competitor, and the platform serves banks, credit unions and fintechs lending into overlapping markets. Nothing states whether captured rules remain tenant specific, who owns them, or what happens to them when a customer leaves.

No data protection agreement, retention schedule, subprocessor list or deletion commitment was located. The material handled is among the most sensitive in commercial banking, comprising borrower tax returns, financial statements, rent rolls and debt schedules, frequently for privately held businesses whose owners are personally identifiable in them, and agents contact borrowers directly to obtain more. None of its handling is described.

No attestation, certification, trust centre or enumerated framework was located. The company is months old with two employees, so the absence is expected rather than negligent, and it is also the most immediate obstacle to the bank segment it targets, since core system integration triggers the most demanding supplier assessment any vendor faces.

Regulatory awareness is embedded in how the product is described rather than claimed generically. Deposit insured banks are named as the buyer, government guaranteed small business lending is supported as a distinct loan class with its own programme rules, and the governance layer is stated to support model risk management and examiner review, which is the framework United States banking supervisors apply to any model influencing credit decisions. Held at B because no supervisor, statute or guidance is named directly, and servicing and collections carry consumer and commercial contact rules that go unaddressed.

Borrowers here are businesses rather than consumers, which lowers the fair lending exposure without removing it, and the sharper question is treatment during distress. The platform performs early warning, covenant breach escalation and recoveries, so its judgements determine which struggling businesses get chased, restructured or pursued, and automated consistency in that work is double edged: it removes the arbitrariness of which officer noticed, and it also removes the discretion an officer might exercise for a borrower having a bad quarter. No analysis of outcomes across borrower types and no escalation policy is published.

No guarantee, indemnity or correction process was located. The lender is well served by comprehensive logging and configurable review, which gives it the record to reconstruct any agent action. The borrower is the party contacted by agents, asked for documents, assessed against covenants and escalated on breach, and nothing describes whether they know they are dealing with an automated system, how a misread financial statement is corrected, or what happens when an early warning flag is wrong.

No model provider, version or hosting arrangement is named, and no subprocessor list was located. Downstream systems are enumerated in unusual detail, so a buyer knows exactly what the platform connects to while knowing nothing about what powers it, which is an odd asymmetry for a product whose governance pitch rests on examiner grade auditability, since model identity and change control are part of what an examiner reviews.

This is the most specific integration disclosure in the index. Ten core banking systems are named individually across the major providers, spanning two platforms each from three of the largest vendors plus four others including an international core, and three loan origination and servicing platforms are named alongside them, with most integrations stated to go live in under two weeks.

Naming particular core versions rather than claiming broad compatibility is a signal only a vendor that has actually connected to them can produce, and it directly addresses why servicing automation usually stalls, since the loan data lives in systems that are decades old and mutually incompatible.

No hosting provider, region selection, residency commitment or private deployment option was located. Exposure is domestic, which narrows the question, and deposit insured institutions connecting a platform directly into their core banking system will require the processing arrangement documented as part of third party risk review regardless.

No pricing, packaging or basis of charge was located. The return case is stated more concretely than most, through the staffing reduction and redeployment figure and through integration timelines of under two weeks, so a buyer can estimate benefit without knowing cost. Nothing indicates whether charge falls per loan serviced, per portfolio, per seat or as a platform fee.

Three buyer types are addressed, banks including deposit insured institutions, credit unions and fintech lenders, and loan coverage spans commercial and industrial, commercial real estate and government guaranteed small business lending, which are genuinely different in documentation, covenant structure and regulatory treatment. Lifecycle coverage is complete from document collection through spreading, covenant monitoring and early warning to recoveries. Scope is domestic and deliberately confined to servicing rather than origination or credit decisioning.