Vesta, the US startup building AI agents for mortgage origination, has closed a $30 million funding round led by Conversion Capital, lifting its total raised to roughly $85 million — about €79.9 million — since it was founded in 2020. Announced in October 2026, the round also counts Citi Ventures, Andreessen Horowitz (a16z), Pennymac, New American Funding, nbkc bank, First American, FirstKey Mortgage, Zigg and Navitas among its backers. The fact that two of the largest lenders in the country are writing cheques alongside the venture funds is the detail that matters most.
- Funding: a $30M round led by Conversion Capital, bringing total capital raised to approximately $85M.
- Traction: revenue is up 12x year over year, with customer count and headcount both up more than 50%.
- Technology: agents built on Anthropic’s Claude Sonnet 4.5 handle document collection, income verification and condition clearing.
- Risk: lenders, not the software vendor, remain legally responsible for underwriting decisions as regulators tighten scrutiny.
The economics of a 40-day, $11,000 loan
The problem Vesta is attacking has a brutally simple metric attached to it. Closing a mortgage in the United States takes an average of 40 days and costs the lender around $11,000 per loan in operational expenses. The wider market moves more than $1.5 trillion annually, yet fewer than 5% of loans are originated with any form of AI assistance, according to figures the company shares with its investors.
That gap is where the company sees its opening. Vesta does not try to simplify the process for the borrower; it targets the repetitive, document-heavy work that consumes an originator’s day. Agents gather paperwork, verify income, review loan conditions and manage communication with buyers and sellers. Crucially, the architecture does not replace the human originator — the agent runs in parallel, shrinking the number of manual tasks each originator must complete per file.
The company says roughly 40% of tasks completed on Vesta are now handled by AI agents and automated workflows, and that share is still climbing. Getting there was not quick: Vesta spent two years writing a core platform from scratch before converting some of the country’s largest lenders, and executives describe the legacy systems of record underpinning the industry as more than two decades old and poorly suited to agentic workloads.
Lenders as investors, and the cost case behind the bet
The composition of the cap table is unusual for a fintech round. Pennymac ranks among the largest mortgage lenders in the United States, while New American Funding is one of the biggest independent mortgage companies in the country. Their participation follows a straightforward logic: if the system works, they are first in line to benefit — and first to push it internally. Pennymac has already reported 25% back-office efficiency gains, with a roadmap to automate 80% of its production workflows by the end of 2027.
Citi Ventures adds institutional weight. For a bank running high-volume mortgage portfolios, automation across the origination cycle translates directly into margin. Vesta positions itself between legacy players such as ICE Mortgage Technology and AI-native rival Xpanse — the former has spent years digitising the process without leaning on agents for operational work, while the latter is racing towards the same target from a different starting point.
Mortgage origination already has verification loops and QC processes built in, and the largest buyers of loans mandate that a percentage of closed loans be audited — creating a clear quality metric to monitor as agents are rolled out.
Every action an agent takes and each piece of reasoning behind it is logged for compliance and audit purposes, with legal responsibility for the call resting on the lender. A personal-assistant product for mortgage issuers, handling task execution and workflow tracking, is reportedly in development.
Regulation is the real bottleneck
The Consumer Financial Protection Bureau has intensified its supervision of AI systems involved in credit decisions, and the burden of proving a model does not introduce bias falls on the lender rather than the technology provider. The 12x revenue multiple is eye-catching, but it comes without a disclosed ARR figure — a multiple on a small base tells you far less than an absolute number would.
At current burn, $30 million should fund between 18 and 24 months of operations. The central question is whether Vesta can convert its relationships with Pennymac and New American Funding into recurring revenue at scale, while absorbing the very real API and compute costs that scale with every agent call. If the model holds up in production and the investors of this calibre imply it has survived serious due diligence, US mortgage lending may prove to be one of the first heavily regulated markets where AI agents graduate from pilot projects to standard operating procedure.