Karus

The removal test leaves the manual underwriting desk the platform was measured against for twenty months. Proprietary machine learning models trained on tens of millions of loan outcomes evaluate thousands of data points and decide in real time at the point of sale, with separate model classes for underwriting, loan structuring and dealer level pricing, plus cash flow forecasting and a conversational layer over the suite. Pricing a loan in seconds against both a borrower's credit trajectory and a specific vehicle's depreciation is not achievable by score cutoffs.

Decisions are made in real time at the point of sale, described as instant decisioning, which is the requirement in a channel where a dealer routes the application to whoever answers first. No human checkpoint, referral threshold or override is described anywhere.

The operating portal provides what the company calls loan decision transparency, visualising underwriting data in real time, but that is visibility after the fact rather than a control before it, and nothing states what a lender's credit officer can adjust, when a case is escalated, or what happens to marginal applications.

The validation methodology is the strongest in this index and it is the one a model risk function would design itself: a champion challenger comparison run in production for twenty months, with the platform and a fifteen person underwriting team working the same dealers, the same borrowers and the same market conditions, results reported across twenty monthly cohorts and 80 million dollars of originations rather than in aggregate, and outcomes measured on losses, consistency and yield at controlled pricing.

Cohort level reporting matters because it shows performance held across changing conditions rather than in one favourable window. Models are trained on tens of millions of loan outcomes and the operating portal is built specifically to make real time underwriting data visible for decision transparency.

The parallel run is the evidence and it is the strongest of its kind in this index: the platform operated alongside a lender's fifteen person manual underwriting team for twenty months on the same loan population, the same dealers, the same borrowers and the same market conditions, across 80 million dollars of originations in twenty monthly cohorts, reporting lower losses, higher consistency and a yield advantage at identical pricing.

Three customers are named across the chain, a subprime lender, an independent dealer group and an automotive lending platform, with the most recent partnership announced in July 2026. A listed technology company acquired a fifth of the business at a ten million dollar consideration, and backing comes from a private credit investor active in asset based lending.

No boundary statement was located and the structure makes the question pointed. Models improve from loan outcomes across the customer base, and the company describes structural preference inside dealer routing channels giving its underwritten loans first look at quality flow, which means it observes and shapes market wide origination rather than sitting inside one lender.

Nothing states whether one lender's performance data trains models serving a competitor, who owns the resulting improvement, or how a lender's dealer relationships and pricing behaviour are protected.

No data protection agreement, retention schedule, subprocessor list or deletion commitment was located. The platform evaluates thousands of data points per applicant at the point of sale, holds borrower income and credit information, and retains loan level performance across portfolios, with training data drawn from tens of millions of historical loan outcomes. Nothing published describes provenance of that training corpus, what is retained on declined applicants, or how borrower data obtained through a dealer is handled.

No attestation, certification, trust centre or enumerated framework was located. Lenders and institutional investors are named buyers and both apply supplier assessment to any system holding borrower data or driving credit decisions, so the review has occurred privately while nothing is published for a prospective customer to examine.

No statute, regulator or rule is named, and the omission carries more weight here than in most profiles. Consumer auto lending sits under equal credit opportunity and fair lending obligations with adverse action notice requirements, dealer routed origination raises its own disclosure questions, and subprime auto specifically has been an area of sustained supervisory and enforcement attention. A platform performing underwriting, structuring and dealer level pricing touches all of it, and none appears in published material.

One published finding is the best inclusion evidence produced by any vendor in this index, because it is measured rather than asserted. Across the twenty month parallel run the platform delivered lower losses while underwriting borrowers with 1,354 dollars lower average monthly income than the human team, which the company summarises as a weaker credit profile on paper with better outcomes in the portfolio.

That is a demonstrated result on the same population under the same conditions, and it is precisely the claim every alternative data lender makes and almost none tests. Held at B rather than A because income is not a protected characteristic: no fair lending testing, disparate impact analysis or outcome breakdown across protected groups is published, and the segment served is one with a documented history of consumer harm.

No guarantee, indemnity or correction process was located. The lender is well served, with cohort level performance data and portfolio visibility giving it unusually clear feedback on whether the models are working. The borrower has nothing described, and their position is weak by the structure of the channel: they sit in a dealership, are decided on in seconds by a system they never chose and whose involvement they may not know about, and nothing states whether reasons are supplied on decline, how misread income is corrected, or how the pricing applied to them was determined.

Models are stated to be proprietary and trained on tens of millions of loan outcomes, which establishes ownership without identifying provenance: whose loans, from which lenders, over what period and covering which credit segments, all of which determine what the models can generalise to.

No credit bureau, vehicle valuation source or alternative data provider is named despite vehicle depreciation being central to the stated method, and the base model behind the conversational layer is not identified. No subprocessor list appears.

Integration into dealer routing channels is the commercially decisive one and the company describes it as structural rather than incidental, giving its underwritten loans first look at flow rather than what remains after other lenders have passed. That is where auto lending is actually won. An end to end operating portal handles programme management, and real time decisioning at the point of sale requires connection into dealer systems. No named dealer management system, loan origination platform, credit bureau or servicing system appears, and no developer documentation was located.

No hosting provider, region selection, residency commitment or private deployment option was located. Exposure is domestic. One arrangement raises a question published material does not answer: a partner intends to integrate the underwriting output into blockchain infrastructure to tokenise the resulting loan assets, which implies loan level model output leaving the platform into a different technical environment, and nothing describes what crosses that boundary.

No pricing, packaging or basis of charge was located. The commercial case is made through portfolio economics instead, since a yield advantage at the same annual rate and lower losses across a measured cohort tell a lender what the platform is worth without stating what it costs. Multiple buyer types are served across originators, dealers and investors, which would ordinarily price differently, and nothing describes how.

This is a deliberately narrow product: one asset class in consumer auto finance, one country, and a focus on the subprime and near prime end where credit scores work least well. Within that slice it reaches several roles along the chain, covering originators and lenders, dealers seeking finance for customers, institutions managing portfolios and investors buying loan flow, and it spans decisioning through pricing, structuring and portfolio monitoring. Specialisation is the strategy rather than a gap, and it remains narrow beside vendors here serving several institution types across markets.