AI‑native LOS scaled fast

A loan origination system is the software lenders use to take an application, check documents, underwrite the file and move it to approval or denial. One AI-native setup said it processed 140 applications in 90 minutes on its first day by sending only exceptions to people instead of handing every file to staff. (techcrunch.com) (x.com) In lending, “debt-to-income ratio” means a borrower’s monthly debt payments divided by gross monthly income, and lenders use it to judge repayment capacity. Automating that math removes one of the routine checks that usually sits near the front of underwriting. (consumerfinance.gov) For conventional mortgages sold to Fannie Mae, the maximum total debt-to-income ratio is generally 36% for manual underwriting, up to 45% with qualifying compensating factors, and up to 50% for casefiles underwritten through Desktop Underwriter. That makes debt-to-income calculation a rules-heavy step that software can standardize before a human reviews edge cases. (selling-guide.fanniemae.com) The “exception-only” model means staff do not touch every application. Dark Matter Technologies, a long-established loan origination vendor, describes the same idea as exception-based process automation that focuses teams only where human attention is needed. (dmatter.com) The other piece is a unified front end and back end. Dark Matter says its platform combines the loan origination system and point of sale in one system and gives each user direct access to the same database, avoiding syncing and duplicate data entry. (dmatter.com) That architecture targets a known pain point in lending software. TechCrunch reported in March 2026 that traditional loan origination systems can take as long as a year to integrate and often come with expensive multi-year contracts, which helps explain why startups are pitching AI-native replacements around speed and lower operating cost. (techcrunch.com) The market is moving because the system of record in lending is hard to swap once it is installed. TechCrunch reported that Fuse had raised $25 million, had more than 100 customers, and was targeting credit unions still running older platforms from vendors including nCino and MeridianLink. (techcrunch.com) The claim in this case is not that humans disappear. It is that software handles the standard files, while people step in when income, debt, documents or policy checks fall outside preset rules — the same division of labor embedded in mainstream exception-based lending systems, but pushed much harder on throughput. (dmatter.com) (x.com) If that first-day result holds up across larger volumes, the pitch is straightforward: fewer manual touches per file, faster application handling, and underwriters spending more time on the loans that actually need judgment. That is the operating model AI-native lenders are now trying to sell into one of finance’s oldest software categories. (techcrunch.com) (x.com)