Non-QM lender Griffin Funding unveils AI-driven underwriting platform

Griffin Funding has launched an internal artificial intelligence-driven platform designed to help loan officers navigate the complex world of non-QM lending.

The tool, known internally as “LIA,” pulls hundreds of pages of program guidelines into a single system and uses AI to deliver tailored answers to loan qualification questions, according to Chloe Shubin, vice president of strategy for Griffin Funding.

Shubin said that development for LIA started in late 2024. The company partnered with AI engineering firm Cadre to build out the automated underwriting system (AUS), which has been in testing with Griffin’s top loan officers since March and was rolled out companywide in July.

The company used to be primarily focused on U.S. Department of Veterans Affairs (VA) loans, but non-QM loans have become the majority of Griffin’s pipeline since 2020 when Shubin joined the company. But these products come with lengthy, investor-specific requirements that can be difficult for new LOs, and even seasoned professionals, to master.

“We started thinking about how we can supplement this learning curve and get people up to speed faster. We have multiple different investors and multiple different takeouts, and all of them have a set of guidelines that are at least 100 pages long. And then on top of that, they’ll have maybe five different matrices for the different programs that they offer,” Shubin said during a live demonstration with HousingWire.

“Instead of flipping through 200 pages of guidelines or relying on tribal knowledge, LIA gives our staff a starting point with qualified answers and flags potential issues upfront,” she added.

The platform includes four chat modes: find a product, test a product, chat with a PDF (most likely a credit report, Shubin explained) and 3.4 MISMO upload.

Each mode allows users to search loan products, upload credit reports or integrate loan application files to test program eligibility. It also connects with lenders to do pricing checks and is being expanded to pull property profile data directly.

As for quality control and compliance, Shubin said that Griffin LOs who use the platform need to perform due diligence as they navigate a loan.

“If [LIA] is ever unsure, it’s going to give you a suggestion; it won’t give you a direct answer. And so that is your trigger to go in and look at the guidelines,” she explained. “Then, after finding the correct answer, we feed the right solution back to LIA so it learns.”

For now, LIA is being used internally as part of Griffin Funding University, the firm’s training program for new hires. But the company hasn’t ruled out eventually offering the product more broadly once it has been fully “hardened” against errors.

Shubin said that a key feature set to debut later this month is source citations, which will link LIA’s answers directly to the relevant section of investor guidelines. That, Shubin said, should give loan officers and underwriters added confidence in the system’s recommendations.