Use caseMortgage lending
A couple sends twelve documents through their broker on Friday night. An auto loan hides in the third statement, and Monday goes into retyping.
Incomes and debits are read on arrival, with their source lineThe undisclosed loan is found before the broker promises an answerThe analyst decides, within his lending authority
Emily: $2,650 a month, stable over three months. Ryan: $2,180 base pay, the October $1,500 bonus set aside as non-recurring.
source · 6 pay stubs, tax return 2025A figure needs a person · the AI says which, with its line
Emily and Ryan Parker's bank statements show the same debit three times, $212 to a consumer credit company on the 5th of each month, and their mortgage application lists no current loan. It may well be an oversight. The analyst simply reaches that line after retyping six pay stubs into a spreadsheet, by which time the broker has promised the couple an answer.
Now the documents are read on arrival. The AI extracts incomes and regular debits the moment the file lands, sets them against your lending policy and names what does not match. The analyst starts from figures and their sources, and keeps the credit decision.
An analyst knows statements, figures and queues by heart. The Parkers' loan goes through because they reach him in the right order: the documents are read on Friday night, the missing schedule is asked for on Monday morning, and the decision is taken on verified figures. The acts below stay on LN-26-20417 until the decision reaches the lending system.
The broker submits on Friday evening. By Monday the incomes and the commitments are extracted and checked.
Emily and Ryan Parker are buying a two-bedroom condo in Charlotte for $310,000. Their broker submits the application and twelve documents through the portal on Nov 6. The portal calls the repository API: a loan case is created, and each document is filed under its type and its borrower.
The AI extracts net pay, employer, employment type and start date from each pay stub, then checks the household income against the 2025 tax return. Ryan's October pay stub carries a one-time bonus of $1,500.
Your lending policy counts only recurring income, so the extracted figure for Ryan is his base pay of $2,180, and the bonus is set aside with the line it comes from.
The AI reads three months of statements for regular debits, rejected payments and overdrafts, and compares the debits that look like a loan with the debts disclosed in the application.
A debit of $212 goes to a consumer credit company on the 5th of each month. The application lists no current loan, so the status written on the case is Attention required, with the three statement lines as evidence.
Nathan Cole picks up the file on Monday. One repayment schedule is all that stands between him and a decision.
Nathan asks the assistant to summarize the application. It gives the purchase, the qualifying income, the disclosed and found debts, and the scoring result your lending system returned through its API, each line with its source.
Nathan asks the assistant for a request to the broker. The draft names the Finauto debit, asks for the repayment schedule and the outstanding balance, and says why the end date matters. He sends it, and the case moves to Waiting for documents.
The broker uploads the schedule the next day: an auto loan of $10,500 signed in 2023, outstanding balance $3,220, last payment on Mar 5, 2028. It is filed on the same case. The AI checks it against the statements and updates the commitments, and the scoring engine is called again.
The file is complete, and the credit decision belongs to the analyst.
Before deciding, Nathan opens the schedule in the viewer next to the case. He asks the assistant how many payments are left, and the answer points to the line on page 2. He drops a note on it for the committee file.
Nathan approves the loan at a 34.6 percent debt-to-income ratio and records his reason: the auto loan ends in 2028 and the household saves $400 a month. The routing rule checks his lending authority. Above the 36 percent ceiling, the process would have opened a credit committee review as a child case, with every document and field.
On approval, the workflow passes the decision to your lending system through its API, where the commitment letter is produced. The loan case keeps its history: every figure the AI extracted, every change Nathan made, every document version, dated and attributed.
Credit stays a human call. What the AI returns is the file read against your lending policy, in one of three states Nathan can act on.
Incomes consistent, debts disclosed, documents current. The case keeps one sourced line per check.
After that: the analyst checks the figures and decides.An undisclosed debit, a bonus counted as salary. The analysis says which, with the lines behind it.
After that: the analyst asks the broker; the check runs again when the document arrives.A ratio above the ceiling, a job still in its probation period. The AI lists what passes, then names the rule that fails.
After that: the file goes to the credit committee as a child case, whole.It grants nothing. It denies nobody. It computes no score of its own. Across the nine steps on this page, the credit decision is taken by the analyst, within his lending authority and with his own trace.
The AI reads, checks, drafts and summarizes. The control is in how the process is written, with its authority levels and its committee path.
Counted on the Parkers' file, twelve documents and one hidden auto loan. Your policy and your brokers will produce other counts.
Where these numbers come from. They are counted on the scenario described on this page, with example data. They are not a customer result.
The analyst signs, the credit manager answers for the portfolio, the CIO wires the systems. Each of them weighs this differently.
Reading documents takes the week, and judging risk gets Friday afternoon.
A figure in a credit decision is hard to defend when nobody can say where it came from.
Broker portal, scoring engine, lending system: three integrations before anything works.
With one document type. Take the bank statements, where undisclosed loans hide, and read them on arrival. A month of applications will show how many loans you were missing.
Your broker portal, your scoring engine and your lending system stay in charge of what they do today. The loan case connects them.
See it live
Bring a few anonymized loan files and your lending policy. We will read one with you on arrival and leave the credit call to your analyst.