An AI-powered underwriting system developed by Better.com can evaluate mortgage loans in a matter of minutes — and in some cases, seconds — a warp speed that may significantly compress one of the most time-consuming parts of the lending process.
Better announced the launch of a conversational credit decision engine on March 5 for mortgages and home equity loans, built in partnership with OpenAI and available in ChatGPT. The new application allows lending teams to connect directly to Better’s Tinman underwriting engine from within ChatGPT, generating loan evaluations that help lenders approve loans faster and streamline underwriting workflows.
With the release, Better is expanding access to its Tinman AI platform across the financial services industry through a custom Model Context Protocol (MCP) connector developed by the company’s engineering and AI teams in collaboration with OpenAI.
Humans in the loop
According to the company, the Tinman AI platform currently manages more than 1,200 distinct workflows across the mortgage process, including underwriting. In many cases, the system can complete the entire underwriting review end to end. When the AI detects irregularities or scenarios that require professional judgment, licensed underwriters step in.
“Licensed professionals are pulled in to review variances detected and for final reviews on any files that require a licensed underwriter prior to funding,” Vishal Garg, CEO and founder of Better.com, told Inman via email.
Vishal Garg | Photo credit: Better.com
Those human reviews also serve another purpose: the feedback is used to continuously improve the platform’s machine-learning models.
The 47-second underwriting claim
Better says the system can underwrite some loans in as little as 47 seconds, a figure that reflects highly standardized loan scenarios. “While there are loans that have been underwritten in as little as 47 seconds, the median timeline to underwrite a typical loan on the Tinman Credit Decision Engine app is 2 minutes and 24 seconds,” Garg said.
That compares with the traditional mortgage underwriting process, which often takes 15 to 21 days. Conventional loans made to highly qualified W-2 borrowers tend to process the fastest, according to the company. More complex loan files may take several minutes or require a final review by a licensed underwriter.
“Some loans take up to five minutes or require a handoff to a licensed underwriter for final review, judgment, and sign-off,” Garg said. “This is made easier by the AI doing all the manual work.”
Mortgage companies, fintechs, and loan officer teams that are enterprise clients of the Tinman AI platform will gain access to the Tinman Credit Decision Engine app if they have access to ChatGPT Enterprise.
A tool that ‘supercharges’ loan officers
If underwriting timelines continue to shrink, the role of loan officers could gradually shift away from paperwork-heavy processing work and toward advisory work with borrowers. Better argues that automation allows loan officers to focus on helping buyers navigate one of the biggest financial decisions of their lives.
“Buying, owning and maintaining a home is a fundamentally human process,” Garg said. “This tool frees up our Loan Officer teams to engage in that process where it is needed and not be bothered by paperwork complexities that are better left to machines.”
Better’s loan officers, processors and operations teams are already using the integration to reduce underwriting escalations, prevent rework by validating key process steps before taking action, and accelerate onboarding by giving new hires immediate access to Tinman’s institutional knowledge.
“This is a tool that supercharges loan officers, and allows them to spend more time on higher value tasks — helping people find a home,” Garg said.
The company also points to productivity gains from earlier versions of its AI platform. In a case study involving lender NEO, Better said the Tinman system led to more than a 90 percent improvement in loans funded per loan officer within six months of deployment.
“This is massive in the context of retail mortgage origination, a part of the mortgage industry with an average cost to manufacture a loan of $11,700 per loan and with near-zero productivity improvement in loans per loan officer for the last 10 years,” Garg told Inman. “With the addition of the Tinman AI app in ChatGPT, we expect even further improvement.”
’30x better than human underwriters’
Rapid underwriting decisions inevitably raise questions about accuracy and regulatory compliance. Better says the Tinman platform is trained on nearly a decade of institutional mortgage intelligence, powering its machine-learning-driven rules engine.
The Tinman platform also has a “context graph,” which has mapped individual mortgage processing and underwriting roles, with tasks, actions, rules, and decisions on more than $110 billion in funded loans, over 12 million recorded customer calls, and billions of pages of credit, income, home, and asset-level documentation.
According to the company, the system has achieved a critical defect rate of 6 basis points, which the company says is far superior to the industry average of 1.8 percent.
“That means the platform is 30x better than human processors and underwriters in terms of loans that are accurately underwritten to investor guidelines across over 40 investors across a broad range of products that all meet the compliance and regulatory conditions that every mortgage needs to comply with,” Garg said.
The ‘scary part’ for mortgage professionals
Better’s Tinman AI platform illustrates the growing shift toward automation in mortgage underwriting, and the company’s claims are backed by compelling data. Some industry observers remain skeptical, though.
Colin Robertson, who runs the mortgage news site The Truth About Mortgage, said the headline-grabbing claim of 47-second approvals may be more about marketing than fundamentally transforming the lending timeline.
“Given the company highlighted 47-second loan approvals tells you it’s mostly a publicity stunt,” Robertson told Inman. “Who cares if it takes less than a minute to find out you’re [conditionally] approved when the entire process still takes 30 to 45 days?”
Still, Robertson said tools like Better’s conversational underwriting engine represent a step toward a more automated mortgage pipeline. “The other side of the coin is the more companies integrate AI, the closer we get to that automated, streamlined mortgage loan experience from start to finish,” he said.
Robertson noted that the shift could have significant implications for the mortgage workforce if automation continues to expand into underwriting, loan processing and document verification.
“The scary part, if you work in the industry, is that the goal is clearly to replace humans with automated workflows, no different than Waymo ditching drivers,” Robertson added.
