Mortgage follow-up calls have long been a bottleneck that loan officers absorb one conversation at a time. Better.com is routing that call volume through AI. Betsy, the company’s generative artificial intelligence loan agent, handled nearly 100,000 mortgage-related calls per month in 2025 and resolved 35.5% of borrower inquiries without any human involvement, according to ElevenLabs, the AI audio company whose voice technology powers the system.
New Pipeline for Mortgage Conversations
Betsy runs on a modular stack combining speech-to-text, a large language model and text-to-speech built on ElevenLabs Agents. Better had initially used a speech-to-speech model but switched architectures to get lower latency and tighter compliance controls, the case study noted. The shift matters in mortgage, where a single borrower conversation can trigger dozens of back-end system calls spanning pricing engines, eligibility checks and investor guidelines.
In 2025 alone, Betsy placed 1.89 million inbound and outbound calls, saving Better’s loan officers more than 1,666 hours of human time each month, ElevenLabs reported. Better CEO Vishal Garg said the system lets the company move routine interactions out of manual workflows while offering borrowers 24/7 support. HousingWire reported that Garg described Betsy as capable of performing rate calculations across more than 26,000 product and investor configurations.
The economics moved with it. AI voice automation delivered a 41% reduction in the average cost to originate and doubled lead-to-lock conversion in 2025, the case study found. That’s a meaningful shift in an industry where margins are thin. Freddie Mac found that in recent years, loan origination costs rose 35% and the average retail-only lender was losing approximately $600 per loan.
Voice as the Front End
What Better has built isn’t a call center chatbot. It’s a voice agent that conducts eligibility checks, discusses pricing and walks borrowers through rate locks. A key factor was selecting platform that is stable under high tool-call volume, since a single borrower conversation can trigger dozens of internal system calls.
That reliability requirement separates financial services deployments from lower-stakes use cases. ElevenLabs Head of Agents Growth Lauren Rothwell said supporting that volume in a regulated space requires accuracy and full compliance, not just a capable voice model.
Advertisement: Scroll to Continue
PYMNTS reported that voice is viewed as the connective layer between large language models and consumer transactions, the interface fast enough and natural enough to compress multistep financial workflows into a single spoken exchange. Voice technology spent a decade as a novelty that consistently hit a ceiling at the point of transaction. That ceiling is moving.
Where the Labor Goes Next
Better’s licensed consultants haven’t been replaced. They’ve been redirected. Betsy handles eligibility screening and appointment scheduling at scale and the humans take calls that require judgment, borrowers in complex financial situations, edge-case underwriting, relationship-sensitive conversations an AI agent isn’t built to close.
That reallocation model is what makes the economics work. PYMNTS Intelligence found that only 60% of U.S. consumers use voice assistants, down from previous years, with trust declining across age groups. It’s a signal that earlier versions of the technology fell short on reliability. Betsy’s architecture is a direct response to that credibility gap, a purpose-built system tuned for one domain rather than a general assistant asked to handle mortgage conversations.
ElevenLabs raised $500 million at an $11 billion valuation earlier this year, a funding round that reflects investor conviction that voice infrastructure is becoming foundational in high-stakes workflows.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
