CarbonAI Raises $12M to Bring AI Agents to Commercial Real Estate Lending

CarbonAI has raised $12 million to automate the analytical work behind commercial real estate loans, putting AI agents to work on underwriting, due diligence, investment memos, and monitoring after a loan closes. The funding, announced October 15, is led by Cowboy Ventures, with participation from Timber Grove Ventures, Bling Capital, and Max Ventures.

The New York company is targeting work that often sits between a collection of documents and a consequential financial decision. A lender has to reconcile leases, rent rolls, financial statements, and legal terms before deciding what a property can support. CarbonAI’s pitch is that specialized agents can carry that preparation through to reviewable outputs, with the supporting evidence attached.

The announcement does not disclose a financing stage or valuation. CarbonAI says it reached seven figures in contracted revenue within six months of launching its first product suite in early 2025. Contracted revenue is distinct from recognized revenue or a disclosed annual recurring revenue figure.

Turning a Document Pile Into a Reviewable Loan Model

Commercial real estate underwriting depends on more than extracting numbers from a PDF. An analyst needs to understand how those numbers relate to each other: whether rental income matches lease terms, how expenses are treated, and which assumptions drive the resulting cash flow.

According to the announcement, CarbonAI’s agents extract and structure information from source documents, build underwriting models, draft investment memos, and flag risks. Outputs link back to the relevant documents, while decisions requiring human judgment are routed to the lending team.

That division of work is central to the product. Preparing a model and documenting an assumption can be automated without treating the resulting lending decision as automatic. A reviewer still needs to judge whether the assumptions make sense and whether the proposed loan fits the institution’s criteria.

CarbonAI calls these “long-horizon” agents: systems intended to complete connected analytical tasks rather than wait for a separate prompt at every step. In this setting, the useful outcome is a coherent package for review, rather than a sequence of isolated answers about individual documents.

Working With the Lender’s Existing Process

CarbonAI’s product website adds an important implementation detail: its underwriting workflow is designed around a lender’s own templates and lending criteria. The company also describes tracking outstanding diligence documents in the background and delivering surveillance reports in the firm’s preferred format.

That approach recognizes a practical constraint in enterprise AI adoption. An institution’s model is also a record of how it evaluates risk. Replacing it with a generic output can create reconciliation work for analysts, even if the new software generates that output quickly.

For example, a rent figure pulled from a lease may need to be associated with the correct tenant, period, and property before it belongs in a cash-flow model. Source links help a reviewer inspect the evidence, but the interpretation and placement of that figure still matter. The product’s value therefore depends on how well extraction, modeling, and review work together.

The website says every step is logged and figures are traceable to their sources. It also lists role-based access, single sign-on, encrypted U.S. hosting, and a policy against using customer data to train models. These are the company’s published product and security descriptions.

Extending Analysis Beyond Loan Origination

CarbonAI’s scope continues after a loan is made. The release describes continuous monitoring intended to surface risks before the next manual quarterly review, alongside agents that perform quarterly surveillance.

Origination and monitoring involve related information, but different questions. Before closing, a lender is testing a prospective investment. After closing, it needs to determine whether the property’s performance, documentation, or obligations have changed in ways that require attention.

Keeping the underlying records connected could make it easier to compare new information with the assumptions used when a loan was approved. It could also reduce the effort involved in repeatedly rebuilding an asset’s history. The announcement does not specify the refresh frequency or external data sources used for every monitoring workflow, so continuous monitoring should not be read as a guarantee of real-time coverage of every event.

Reported Savings, With Important Distinctions

CarbonAI says early customers saved as much as 90% of the time spent on selected core underwriting tasks, including lease review, financial statement analysis, and loan onboarding. The figure describes task-level time savings reported by the company, rather than a measured reduction in every customer’s complete lending process.

The release names Greystone as a customer and includes comments from co-president Mordecai Rosenberg about automating due diligence collection. CarbonAI also says its customers include debt funds, mortgage real estate investment trusts, agency multifamily lenders, and publicly traded financial institutions.

For lenders evaluating this class of software, speed is only part of the comparison. The relevant questions include whether the system preserves source evidence, identifies exceptions, and reduces review effort without obscuring uncertainty. These issues connect to the broader challenge of governing AI agents in enterprise workflows, where permissions and accountability must remain clear as systems take on more work.

Founders With Experience on Both Sides of the Workflow

Co-founders Jeff Saul, CarbonAI’s CEO, and Josh Kaufman previously worked in commercial real estate lending and loan management at firms including Starwood Capital, Morgan Stanley, and Guggenheim. They later built Nativ, a commercial real estate deal-management platform, and sold it to Built, where both held leadership roles.

That background helps explain the emphasis on existing spreadsheets, report formats, and lender-specific processes. CarbonAI is positioning domain knowledge and workflow execution as the product, with generative AI supporting the work rather than serving as the entire interface.

The funding backs a focused application of agentic AI: assembling and maintaining the analytical record behind a commercial real estate loan. Its practical test will be whether lending teams can review more work with confidence while retaining responsibility for the judgments that determine a deal’s risk.