Email
Copy Link.png)
Jeff's first job out of college was on the CMBS desk at the Royal Bank of Scotland, reconciling securitization data tapes until 3 AM. Josh started in underwriting and asset management at Morgan Stanley, where hours of every day went to wrangling borrower documents just to report on the health of loans each quarter. We spent roughly the next decade underwriting, originating, and asset managing CRE loans at firms like Starwood, Argentic, Bayview Asset Management, and Guggenheim. The work got more sophisticated. The problems didn't. No two deals looked alike. The rent rolls, cashflows, and other deal artifacts came in a different format every time, and every deal had its own nuance. As our MDs used to remind us, a $5 million loan takes as much work as a $50 million one. In those years, outside of Excel, neither of us can name a piece of technology that changed how we handled those nuances and absorbed the work.
So we built one. Nativ was a deal management platform for CRE lenders. It worked because it plugged into the models and reports lenders already used instead of asking them to switch to a foreign template and framework. More than 30 lenders and over $300B of AUM ran on it before Built acquired the company. We then rebuilt it inside Built and went on to lead its lender business. That's two times building software for the same workflow we used to do by hand, and it put us in a small group: people who’ve worked loans from broker package to payoff and have shipped software lenders actually use. We learned what lenders will and won't adopt. We also learned the limit of pre-AI software: it could organize the work, but it couldn't do the work. Someone still had to read the documents, validate the numbers, and type the same figures into multiple places (model, memos. etc) a dozen different times. Almost all of the work was still manual.
Then the technology caught up, one step at a time. In 2024, models learned to read messy deal documents. In 2025, well-tuned models paired with domain-specific application layers could get it right on deals they’d never seen, with work you could check. In 2026, agents learned to run the whole job, from the first document to the finished memo. Carbon is the third time we’ve built for this problem, and the next step in a journey we started almost ten years ago with Nativ.
The need is real. CRE teams across the industry spend an estimated $50 billion a year to underwrite and monitor investments. A loan gets underwritten again and again over its life: at origination, every quarter, at extension, at refi. With roughly $1 trillion in commercial mortgages maturing over the next year, teams can't hire their way through it. Lenders end up choosing between slower decisions and thinner reviews. It's a labor problem, not a software problem.
General-purpose AI doesn’t solve it. It’s a chat tool you prompt deal by deal, and it poses risk questions no lender can live with. When it gets a number wrong, how do you know? How do you trace where it came from? If it nails deal one, will it nail deal two? If one analyst finally gets it working after an afternoon of prompting, will the analyst next to them get the same answer? In lending, everything has to tick and tie, and every piece of data has to be explainable to a credit committee. Lenders don’t need another chat tool. They need agents that already know the domain and do the work the same way every time, across every kind of deal that shows up in the wild.
So we started where the work starts: getting the right data out of the documents every deal runs on, like rent rolls, operating statements, leases, and loan agreements. That sequence was deliberate. Nothing built on bad data can be trusted, and in lending, a number you can’t trace is a number you can’t defend. Pulling numbers or text off a single page is mostly solved. Doing it reliably across every format, then interpreting, reconciling, and validating what comes out, is not. That’s where we spent our first 40,000+ documents, and it’s where you earn the right to do any other analytical work for a lender.
From that foundation, Carbon grew into domain-trained agents that run the analytical work end to end across the life of the loan: underwriting, due diligence, the investment memo, and post-close surveillance. They work inside each lender’s own models and memo templates and follow its lending programs and parameters. They trace every figure back to the page it came from and hand back only the decisions that need a person. And they learn from every deal, building up the firm’s institutional knowledge and way of working over time. Lending teams stop triaging documents and data and spend their time on what matters most: the deal. The result is that you can make and manage more loans with the same team and materially mitigate risk with higher fidelity information and upleveling where your team spends their time.
Our first customers trusted us with real deals before we had a launch to announce, and some have cut time on core underwriting tasks by as much as 90%. Today we work with debt funds, mortgage REITs, agency lenders, third party diligence providers and publicly traded financial institutions. to expand their capacity to make and manage CRE loans with.
For years, software could help organize this work. It never did it. Carbon is the tool we wish we’d had as analysts, grinding through deals late at night and we are only getting started.
Built by lenders. Run by agents. Ready for you.