Stacc

The sixth consecutive centrality C off this analyst roster, and the most explicit self disposal encountered anywhere in the sweep. The vendor's own homepage states the position in four words, credit native rather than AI native, and describes the product as an AI assisted credit platform on a modular, highly automated and deterministic backbone. Every element of that phrasing places the learned components beside the system rather than inside it.

Its own guidance material describes a platform approach in which AI is embedded into discrete, structured parts of the process, which is an accurate account of an accelerant rather than an architecture. Strip every model and a complete cloud native credit platform remains, which is what a tier one bank is buying when it replaces a mortgage service. The AI line is nonetheless substantial and specific, which is why this is a build at C rather than a rejection.

Better articulated than most on this roster, and the architecture itself carries part of the argument. The platform is described as running on a deterministic backbone with a complete audit trail on every case, and the AI capabilities are consistently scoped to support rather than decide: decision support highlights household debt concerns, unused collateral and outdated valuations for an adviser to weigh, document intelligence flags discrepancies rather than resolving them, and cross sell suggestions are described as rule driven.

The agentic case advisor is explicitly bounded, guiding applicants toward options that sit within the bank's own credit policy and business rules, which is a published constraint on what the agent may propose. Held at B rather than A because the constraint is asserted at the level of architecture without being specified: no confidence threshold, no escalation route, no stated review band, and nothing describing what happens at an automated decline, which in a mortgage workflow is the decision that matters most to the applicant.

Nothing published about the models themselves. For a document intelligence capability whose stated function is extracting income and purchase price from employment contracts and tax returns, the decisive metric is extraction accuracy, and no error rate, confidence score, human review rate or benchmark appears anywhere. Nor is there validation evidence for the adverse media screening, the decision support signals or the conversational agent.

The deterministic backbone claim is an architectural statement about the surrounding system, not a measurement of the learned components sitting on it, and the two should not be confused. This is the standing document extraction question already banked against Oxane Partners, now recurring in a second pocket.

A named tier one customer with a named executive on the record. DNB, Norway's largest bank, selected the platform for a new digital mortgage service covering DNB and Sbanken, and DNB's Division Director for Mortgage Digitalisation is quoted by name describing the ambition to shorten approval and payout timelines, with refinancing decisions potentially completed within hours rather than days. The chief executive is separately on record on the Nordic market shift.

Independent recognition is specific and checkable: the vendor is named in the Gartner Hype Cycle for Bank Lending 2025 with the analyst and publication date stated, and the certifications page enumerates further awards by year. Held at B because the headline figure is an ambition for a platform still being rolled out rather than an achieved result, and the customer testimonials carried on the site are unattributed.

Nothing states whether customer data trains the vendor's models. The platform processes mortgage applications, income documentation and credit files for more than a hundred financial institutions including direct competitors in the same national markets, which makes the pooling question sharper here than for a single tenant deployment.

No exclusion of client data from any training corpus, no separation commitment between institutions, and no statement of controls around the conversational agent or the document pipeline. Against the reference set of Mortgage Capital Trading, Needl and AlphaSense, all of which answer this plainly, the silence is a choice rather than a constraint.

No privacy posture is published in the material reviewed. Nothing addresses retention of applicant financial data, purge on contract termination, the handling of the income documents, tax returns and employment contracts the document intelligence capability necessarily ingests, or the processing of adverse media screening results about identifiable individuals. Queued check: the real time trust centre is the likely location for this and was not opened, so the grade rests on an absence in indexed material rather than a confirmed silence.

The strongest security presentation on this roster, and the precision of the language is itself the evidence. The vendor states that it is SOC 2 Type 2 attested and ISO 27001 and ISO 9001 certified, using the correct noun for each credential in a single sentence: an attestation for the SOC 2 examination, certification for the two ISO standards, with the SOC 2 type stated. It also operates an audited real time trust centre and claims a complete audit trail on every case.

A dedicated certifications page enumerates the two ISO standards alongside separately labelled analyst recognitions and awards, so credentials and recognitions are not blended into one undifferentiated badge row. Queued check that would confirm or qualify this grade: the trust centre itself was not opened, so certificate numbers, the accrediting body and report access remain unverified.

A software supplier with no licence, no authorisation and no supervisory relationship of its own. The compliance language is about the product rather than the company: the platform is described as digitising compliance and as built to meet regulatory requirements natively, and the company describes serving one of the most digital and regulated financial markets. Those are claims about what the software helps a bank do, not statements of standing. Under the standing bar a technical or product compliance claim does not read across as regulatory status, and no enrolment, sandbox participation or registration was found.

No fairness position of any kind. No disparate impact testing, no protected characteristic handling, no model inventory, no named governance framework, and no reference to the EU AI Act despite a European buyer base and a product that supports creditworthiness decisions, which the Act treats as a high risk use.

The concrete exposure is residential mortgage lending across the Nordics: decision support that surfaces household debt concerns and collateral adequacy shapes which applicants an adviser scrutinises, and a screening layer over unstructured adverse media makes judgements about named individuals from press coverage of uneven quality. Neither carries any published fairness or accuracy position.

No recourse position is published. Nothing states what happens when a document is misread, an adverse media hit is wrong, or a decision support signal misdirects an adviser, whether the bank or the supplier carries the consequence, or whether a declined applicant is told that automated processing contributed.

The exposure follows the standing pattern for lending software in this index: the bank is the regulated party answerable to its supervisor and its customer, while the supplier that built the extraction and screening layer sits outside the conduct perimeter. The adverse media element sharpens it, because a false match against a named person is a harm to someone who is not the vendor's customer and has no relationship with the vendor at all.

No model, family, version or provider is named for any of the AI capabilities. The conversational lending agent, the document intelligence extraction, the adverse media screening and the decision support signals are all described by function and none by dependency, and there is no statement of whether the models are built in house or licensed, no version policy and no per capability breakdown.

The contrast with the company's own engineering posture is the notable part: it publishes an open source framework for workflow and decision automation on the BPMN and DMN standards in a public repository, so it is demonstrably willing to be specific about its architecture. Openness about its own orchestration sits beside silence about its model dependencies.

A composable, cloud native architecture positioned as the credit platform of record rather than a layer beside one, covering origination, management and back office in a single modular system. The DNB engagement is the evidence of depth: replacing the mortgage service end to end for a tier one bank and its digital subsidiary requires integration with core banking, collateral, valuation and disbursement infrastructure.

The company also publishes an open source framework for workflow and decision automation built on the BPMN and DMN standards, which tells an integrator what the orchestration layer actually is rather than asking them to take it on trust. Held at B because no core banking platform is named as a certified integration and no connector catalogue is published, so depth is inferred from deployments rather than read off a stated list.

Cloud native delivery is stated repeatedly and is clearly the architecture, but nothing beyond that is published. No cloud provider is named, no regions are listed, no residency commitment is given, no on premise or private deployment option is described, and there is no statement of where customer data is processed or stored. That gap is worth noting for a supplier whose buyers are EEA banks under DORA and GDPR and whose largest customer is a systemically important Norwegian institution. Queued check: the real time trust centre referenced on the homepage is the obvious place this would be answered and was not opened.

No price is published: no plan structure, no unit or per case basis, no seat cost, no implementation estimate and no indicative contract size. One piece of genuine commercial information does exist and is credited in this note rather than in the grade, because it tells a buyer about the shape of the deal and not its cost: the investor material states that the business runs on a multi year annual recurring revenue subscription model for integrated enterprise software. That is more than most vendors on this roster disclose, and it is still not an answer to what the platform costs.

Broad across institution type, product line and geography. The platform serves banks, financing companies, invoice management firms and financial and investment managers, and the functional footprint spans mortgages, asset finance, consumer lending, commercial lending, small business lending, deposits, invoice management, financial management and KYC and anti money laundering case handling.

Five offices across the Nordics, customers in five countries, and forty five percent of 2025 revenue earned outside the home market, which is a real measure of coverage beyond a single jurisdiction. Nordic tier one banks are among the customers, including DNB and its digital subsidiary Sbanken. Recorded data conflict: the vendor's about and investor pages say more than 115 financial institutions while its guidance material and trade coverage say more than 280. The conservative figure is used and the discrepancy is noted rather than resolved.