Commercial Mortgage Underwriting
Before analysts could evaluate a commercial mortgage, they had to manually extract information from rent rolls, operating statements, financial statements, and other borrower documents, then re-enter it into a standardized pro forma. We built an agentic AI product that prepares the pro forma, flags missing information, and gives analysts more time to evaluate the deal.
Users
Brokers, analysts & underwriting leaders
Before
manual re-keying, deal by deal
After
auto-filled pro forma, ready to decide
When a broker brought in a potential deal, an analyst received a collection of borrower documents: rent rolls, operating statements, construction budgets, financial statements, and more.
Before evaluating whether the property could support the requested loan, the analyst had to identify each document, find the relevant financial information, normalize inconsistent formats, enter the values into the correct pro forma, and search external listings and past appraisals for market comparables. The process varied by loan type, document format, and analyst.
The analyst's judgment was valuable. Re-keying borrower documents wasn't.
An analyst uploads the borrower's documents to a deal. The system classifies each document, extracts the relevant values, and maps them into the correct pro forma based on the loan type. The analyst then reviews the extracted information, corrects any values, completes missing fields, and moves directly into evaluating the deal.
01
Rent rolls, operating statements, construction budgets, and other borrower documents.
02
Identify each document and determine which information should be extracted.
03
Map the extracted values into the correct purchase, refinance, or construction pro forma.
04
Highlight missing or uncertain fields for the analyst to verify before recommending.
01
Purchase, refinance, and construction deals need different documents and pro forma templates. The system identifies the loan type first, then routes information into the right workflow.
02
Every extracted value stays editable. Anything the system can't confidently extract is left blank and flagged instead of presented as fact.
03
The product prepares the financial model. The analyst still assesses the property, evaluates the borrower, and recommends whether the loan should proceed.
The underwriting workflow surfaced a second problem: deals were tracked in individual broker spreadsheets. Leadership had no way to see live deal flow, expected closings, or analyst capacity without asking brokers for updates and consolidating everything by hand. We designed a shared deal dashboard where brokers create and monitor deals, underwriting leaders assign analysts, and every deal moves through intake, underwriting, review, and decision in one place.
The copilot accelerated individual deals. The dashboard made the entire pipeline visible.
Underwriting answers one question: should we issue this loan? Once a loan is issued, the lender has to run recurring reviews to make sure it stays compliant, which meant analysts tracking deadlines in spreadsheets, interpreting each loan's requirements, drafting request letters, chasing borrowers, and validating whatever came back. We extended the product into an Annual Review workflow that handles that cycle automatically.
The goal wasn't only to automate document review. It was to automate the compliance treadmill around it.