Sela Builds Voice AI for Mortgage Borrower Outreach

Sela is a San Francisco artificial intelligence company building voice agents for mortgage lenders. Founded in 2024 by CEO Nate Becker and CTO Vahe Tshitoyan, the company focuses on a specific point in the mortgage funnel: the gap between a borrower showing interest and a licensed loan officer having a useful conversation.

The platform handles inbound lead qualification, proactive outreach, objection handling, and follow-up across voice, text, and email. When a borrower is ready, Sela transfers the conversation to a loan officer with context rather than handing over a cold name and phone number.

That focus matters because mortgage sales runs on timing, consistency, and trust. Sela reports 15M+ monthly calls, $6B+ in originations supported, six of the 10 largest independent U.S. mortgage banks as customers, and three of the five largest aggregator lead buyers. Those are company-reported figures, but they show why voice AI is moving from a demo category into production infrastructure for regulated consumer finance.

About Sela

Sela grew from Becker and Tshitoyan's different views of the same operating problem. Becker previously co-founded VoiceOps, a sales coaching and training company, and worked as a data scientist at LinkedIn. He saw that top salespeople could outperform average performers by three or four times, while the behaviors behind that performance remained difficult to reproduce across a team.

Tshitoyan previously worked as a senior machine-learning engineer and tech lead at Google. He holds a PhD in physics from the University of Cambridge. His background brought the production challenge into focus: a useful system has to perform reliably across millions of interactions, not just sound impressive in a controlled test.

Sela's stated goal is to make strong sales execution repeatable. Its agents learn from tens of millions of calls, then improve through simulations, evaluations, optimizations, and A/B tests. The company sells the outcome of productive borrower conversations rather than treating voice minutes as the product.

How Sela Fits Into Mortgage Distribution

Mortgage lenders lose opportunities when new leads wait too long for a response or when old contacts sit untouched in a customer relationship management system. Sela connects to lead providers, mortgage CRMs, loan-origination systems, and contact-center tools so an agent can respond after a trigger rather than waiting for a person to work a list.

In consumer-direct lending, the trigger might be a new lead, a purchase inquiry, an expiring preapproval, or an inbound call. Sela's agent makes contact, asks qualification questions, handles common objections, and completes a live warm transfer. In retail outreach, the trigger might be a rate change, credit activity, or a scheduled mortgage review. The agent re-engages the borrower and books a qualified opportunity with the loan officer.

The distinction is important. Sela is not positioning the AI as a licensed loan officer or as a replacement for the regulated advice surrounding a mortgage. The agent works the repetitive outreach and qualification layer, while the licensed professional takes over when the conversation reaches the point where judgment, disclosure, and accountability matter most.

Why Voice AI Has to Be Operationally Boring

Human-sounding speech gets the attention. The hard part lives underneath it.

A production mortgage agent needs low latency, reliable telephony, clear consent controls, accurate context, call recording, audit trails, safe action boundaries, and a clean human handoff. Sela says its platform includes TCPA-compliant dialing, SOC 2 Type II controls, and 24/7 infrastructure monitoring. The FCC has confirmed that AI-generated voices fall under the TCPA's artificial or prerecorded voice rules, which makes consent, identification, disclosure, and opt-out design operating requirements rather than legal footnotes.

Sela describes one apparent agent as a set of specialist systems that can change context, tools, and instructions between turns. The company tests those systems with turn-level evaluations, complete call simulations, simulated borrowers, and model-based judges. That machinery is less cinematic than a perfect voice demo, but it is where a regulated lender decides whether the system can be trusted with real customers.

Funding and Market Momentum

Sela has raised more than $21M across Seed and Series A financings from Emergence Capital and Costanoa Ventures. The most recent financing was a $15M Series A in June 2026. The company has not publicly disclosed its valuation, round terms, or a detailed use-of-proceeds plan.

The capital arrives as lenders search for operating leverage in a large and uneven market. The Mortgage Bankers Association forecast $2.2T in single-family mortgage originations and 5.8M loans for 2026, but a larger market does not make borrower acquisition simpler. Lenders still have to respond quickly, manage contact rules, preserve context, and move qualified demand to licensed staff without wasting expensive human time.

Sela's narrow mortgage focus is its strategic bet. General-purpose voice platforms can provide components. A mortgage-native system can build workflows around lead sources, loan officer handoffs, refinance timing, borrower objections, and the compliance expectations of lenders. The durability of that advantage will depend on conversion quality, reliability, and whether integrations become harder to replace as the product handles more real operating volume.

Leadership, Culture, and Hiring Signal

Sela's careers page describes a small San Francisco team working on real-time infrastructure, agent orchestration, simulation, evaluation, security, privacy, and customer-facing product delivery. The company says it hires for intelligence, communication, and the ability to make an individual contribution strengthen the whole team.

The open roles span engineering, product, recruiting, and strategy and operations. That hiring mix is a market signal. Sela is not only adding model talent. It is building the systems, customer translation, and organizational capacity required to move from early product-market fit into a larger enterprise footprint.

The culture is explicit about intensity and ownership, including four days each week in the San Francisco office. That will not fit every candidate, and the clarity is useful. Startups operating in regulated, high-volume AI need people who can connect model behavior to telephony, customer requirements, safety constraints, and measurable outcomes without treating any one layer as somebody else's problem.

What Sela Signals for Consumer Finance

Sela shows how applied AI companies can build leverage by choosing a narrow workflow before chasing a horizontal platform story. Mortgage outreach has high consideration, measurable conversion events, clear handoffs, and enough repetition to justify deep automation. It also has the regulatory and emotional weight to expose weak systems quickly.

If Sela succeeds, the lasting product will not be a voice that sounds human. It will be an operating layer that knows when to call, what to ask, what it may do, how to preserve context, and when a licensed person should take over. That pattern can extend into other high-consideration consumer categories, but mortgage is the proving ground.

The company is building for the few seconds when borrower intent is still warm. Its larger test is whether speed, context, compliance, and human judgment can arrive in the same conversation.