A San Francisco startup whose AI agents handle the first phone call a mortgage shopper gets has raised $21 million, and it says six of the 10 largest independent mortgage banks now run the software in production.
Sela announced the funding on Sept. 22. The total spans a seed round and a Series A, led by Costanoa with participation from Emergence Capital. The company said its agents help loan officers originate more than $1 billion in new mortgages a month and that it crossed $10 million in annualized run-rate revenue in 18 months.
The product is narrow by design. Sela’s agents take the sales call β answering questions, explaining loan options, and handing the borrower to a human loan officer once the borrower is engaged enough to move forward. It is the top of the funnel, not underwriting or servicing.
Why lenders are buying it now
Origination economics are the reason, and Sela’s lead investor said so plainly.
“Every lender’s P&L comes down to the same two numbers: conversion and cost per funded loan, and both are under more pressure right now than at any point in a decade,” said David Cheng, a partner at Costanoa. “What convinced us to lead was that Nate and Vahe fundamentally sell the ability to do more funded loans.”
That pressure is not abstract. Lennar’s chief executive told investors on Sept. 16 that the 30-year mortgage rate was approximately 6.8 percent at the end of the builder’s quarter and “even higher since,” with buyers slowing their decisions as affordability tightened. On our reading, that is the sales case: when fewer applications are chasing the same fixed cost of a staffed sales desk, software that works leads at a marginal cost gets a hearing it would not have had in a refinance boom.
The numbers the company put on the table
Sela disclosed two A/B tests to support its claims. In one, run across more than 10,000 borrowers, the company said its agents produced a 9 percent increase in lead-to-lock rates and 40 percent higher profit. In a second, across more than 7,000 leads, it said its agents outperformed a competing voice AI by 41 percent on a lead-to-lock basis.
Those are the company’s own figures from its own tests; no independent evaluation was released alongside them. They are also the metric lenders care about β whether a lead becomes a locked loan β rather than a measure of whether the agent’s answers were correct.
That distinction matters. RealtyWire has reported on a test in which AI mortgage assistants got nearly one in four answers wrong, a reminder that conversion and accuracy are separate problems in a product that is explaining federally regulated consumer credit to first-time borrowers.
Who built it
Sela was founded by Nate Becker, its chief executive, who previously co-founded VoiceOps and worked as a data scientist at LinkedIn, and Vahe Tshitoyan, a former senior machine learning engineer and tech lead at Google.
“At my last company, I watched the strongest salespeople outperform the average by three or four times, and how hard it was to coach an entire team to that level,” Becker said. “Sela’s agents take the best performing sales behaviors, learned across tens of millions of calls, and employ them consistently across every customer interaction.”
He added that automating the mortgage process “doesn’t actually help anyone unless it improves the experience for consumers and makes the process more effective for lenders.”
The company has 17 full-time employees and said it plans to reach about 50 within a year, hiring in product, engineering and go-to-market roles. It said it will use the money to broaden its agents beyond mortgage sales into the wider consumer finance journey.
Voice AI on lender phone lines is no longer a novelty. New American Funding has deployed AI to field customer calls across its mortgage business, and vendors have been pushing into every stage of the loan file. What is newer is a funded company publishing conversion numbers and inviting lenders to judge it on them. More coverage of lending technology is on RealtyWire’s Mortgage page.
