Banxware

Models carry the underwriting judgement, with risk assessment built on bank account level quantitative analysis, cash flow based underwriting and scoring that draws sales and transaction data from the platform through interfaces rather than relying on a credit bureau file. The company positions that explicitly as going beyond traditional credit scoring.

Against it, the product describes itself as an orchestration layer coordinating onboarding, decisioning, disbursement, servicing and collections, and that coordination has independent value to a bank whichever way the credit decision is reached. A stated use of the 2025 investment was developing artificial intelligence based underwriting models, which places part of the modelling capability in the funded roadmap rather than in what ships today.

The product is built for speed, with capital reaching a merchant within 24 hours on little or no paperwork, which necessarily means the credit decision is automated. Genuine control sits with the lender at design time, since decisioning rules are configurable and the company is explicit that banks and lenders retain ownership of products, capital and risk, so the institution sets the box the automation runs inside.

Nothing exists at decision time: no referral threshold, no manual review path for marginal applications, no confidence exposure, and no described treatment for a merchant whose platform data is thin or atypical. The company's own published musing on whether artificial intelligence will replace the human underwriter is posed as a question rather than answered as a position.

No accuracy figure, discrimination statistic, validation result or portfolio performance data was located, and precise risk assessment is asserted rather than shown. One indirect signal carries some weight: a major banking group examined the underwriting closely enough to invest twice and then to assume the entire loan book under a forward flow agreement, which is a sophisticated counterparty putting capital behind the models. That is diligence conducted in private rather than evidence published, and nothing describes how models are validated, monitored or governed as merchant behaviour shifts.

More than 50 platforms are live on the infrastructure and six are named publicly, spanning a major payments acquirer, a large food delivery group, a business banking provider, a card acceptance business, a cash flow management platform and a consumer subscription business, which is an unusually concrete customer list for a company of this size.

A white labelled lending product for a large German bank is live in production, offering businesses financing from a thousand euros to five million through a single integration and application flow. A major European banking group is both a strategic investor across two rounds and the funding partner behind the loan book, which is validation from an institution that examined the underwriting closely enough to fund it. What is absent is volume: no origination total, loan count or portfolio performance figure is published.

No data boundary statement was located. The question is real at this footprint because more than 50 platforms feed merchant sales and transaction data into shared underwriting models, and some of those platforms compete directly, so repayment outcomes observed through one marketplace improve the scoring applied to merchants trading on another.

Nothing states whether models are trained per platform or in common, whether a platform can decline to contribute performance data, or what happens to the accumulated behavioural data when a platform partnership ends.

No data protection agreement, retention schedule or subprocessor list was located. The payload is commercially revealing rather than personally sensitive: bank account transaction history plus sales and order data pulled directly from the platform a merchant trades on, which together constitute a fuller picture of a small business than that business has ever handed to a lender before.

Operating under European data protection law provides a statutory floor, and nothing published describes what the merchant is told about the platform sharing its sales data, how long that data persists after a declined application, or whether it is retained when a loan is repaid.

No attestation, certification, trust centre or enumerated framework was located. A major banking group has integrated the infrastructure deeply enough to run a branded product on it and to fund the resulting loan book, which implies security assessment was passed at a demanding standard, and none of that assurance is published. A new platform or lender evaluating the company therefore starts its own review from nothing.

The regulatory posture is articulated clearly and the structure supports it. Under the forward flow arrangement the bank partner assumes the full loan book, so the regulated institution is the lender holding the credit risk and carrying the licensing obligations, and the company states plainly that banks and lenders keep products, capital and risk ownership fully in their control. That is the correct position for a technology supplier and it is reasoned rather than assumed.

Named partners are all regulated entities, including a major European banking group, its German bank and two banking as a service platforms it owns. What is missing is specificity, with no supervisor, statute or lending regime named in any of the markets served.

The borrower is a business rather than a consumer, which lowers the protected characteristic exposure that drives this axis in consumer lending, and one published term genuinely favours the borrower: financing is offered without a personal guarantee, so a failed business does not follow the owner into personal insolvency, which is a material protection rarely stated this plainly. The unaddressed risk is platform dependency.

Judging a merchant on sales and transaction data drawn from one platform means creditworthiness is measured inside a single ecosystem, so a business is scored on performance shaped by that platform's own ranking, fee and visibility decisions, and a merchant disadvantaged by an algorithm change is downgraded as a borrower for reasons outside its control. Nothing published addresses fair lending testing, decline explanations or performance across merchant categories.

No guarantee, indemnity or falsifiable accuracy commitment was located. Two features work in the borrower's favour without being framed as recourse. Because the bank partner owns the loan book and the risk, the regulated lender carries the obligations that attach to a credit decision rather than an unregulated intermediary. And the absence of a personal guarantee means a rejected or failing borrower faces a bounded commercial outcome rather than a personal one. What is missing is any route to challenge: a merchant declined on the basis of its own platform sales data is not told which signals counted against it and has no described path to correction.

The funding and distribution chain is named at every significant point, covering the banking group behind the loan book, its German bank whose branded product runs on the platform, the two banking as a service platforms it partners with, and a lending marketplace collaboration, so an institution can see who sits on either side of a transaction. Data inputs are described by category, comprising bank account information and platform sales and transaction feeds. What is not disclosed is the model layer, with no provider named for the scoring components, and no subprocessor list or hosting arrangement located.

Integration is the entire product rather than an attribute of it. The orchestration layer connects the full lending chain, onboarding, credit decisioning, disbursement, servicing and collections, and a bank can take the whole stack or activate only the modules it lacks, which is the flexibility that determines whether an incumbent can adopt anything at all.

On the platform side, interfaces pull sales and transaction data directly from the merchant's trading environment, and more than 50 platforms are live, including large payments, delivery and business banking businesses. Two banking as a service platforms are named as partners, and a bank's own branded lending product runs on the infrastructure through a single integration and application flow spanning a thousand euros to five million.

No hosting provider, region selection, residency commitment or private deployment option was located. Operating from Germany and serving European institutions places the business inside a strict data protection regime by default, which sets a floor, but a bank outsourcing credit decisioning is expected by its supervisor to know where processing occurs and under what arrangements, and nothing published answers that.

No pricing, packaging or basis of charge is published for the platform itself. Loan parameters are stated clearly, with amounts running from a thousand euros to five million and a higher band available through the bank partnership, but those are the end product's terms rather than what a platform or a lender pays to use the infrastructure. Nothing indicates whether the model is a share of origination, a per decision fee, a platform licence or some combination.

Three distinct buyer types are served through one infrastructure, with digital platforms embedding credit for their merchants, banks and lenders reaching small businesses they cannot acquire directly, and a broker channel using the same rails to place bank backed loans on identical terms. Product coverage spans term loans and credit lines configured through modular workflows, with new products addable and distributed consistently across platforms.

Geographic reach is Germany and the Netherlands with stated European expansion. The limit is segment: this is small business lending specifically, and the merchant population reachable is defined by which platforms have integrated.