All case studies

Case study · Financial services · Credit operations

A leading lender marketplace

A document extraction agent that reads what analysts used to retype.

Every application arrives with a stack of financial documents, from complex packets to photos of forms, that someone has to turn into a spread before a lender can make a decision. We built an agent that extracts the financials, spreads the data into the marketplace's format, and flags issues for lender review, so underwriting starts from a judgment call instead of hours of data entry.

Industry

Financial services

Function

Credit operations

Workflow

Financial spreading

Engagement

Custom agent build

The problem

Borrowers upload financial statements in every imaginable format: scanned PDFs, spreadsheets, statements from a dozen accounting systems. Before any lender on the marketplace could evaluate a deal, someone had to manually extract the numbers and build the spread.

Hours of manual extraction before the credit decision could even start.

Spreading is the definition of an expensive, mission critical queue: high volume, error sensitive, and squarely in the path of revenue. Exactly the kind of workflow custom AI should own.

Uploaded financials · deal #4172

6 docs

FY24-income-statement.pdf

scanned

balance-sheet-Q4.xlsx

export

tax-return-2024.pdf

142 pages

AR-aging-detail.pdf

scanned

every format an accountant has ever invented

What we shipped

A financial spreading agent, built around the marketplace's own formats and review standards.

FY24-income-statement.pdf

Revenue

$12.4M

COGS

$5.1M

EBITDA

$2.2M

Document extraction

Reads financial documents in whatever shape they arrive: complex packets, scanned statements, exports, even photos of forms.

Spread · FY22–FY24

BUILT

Net revenue

9.8

11.1

12.4

Gross margin

54%

57%

59%

DSCR

1.4

1.6

1.7

Automated spreading

Normalizes the extracted financials into the marketplace's spreading format, line by line.

Flags · for lender review

Related party revenue · 8%

Inventory jump vs. FY23

Issue flagging

Surfaces anomalies, gaps, and inconsistencies for lender review instead of letting them hide in the spread.

Outcomes

What changed after the rollout.

Spread on arrival

Prebuilt · sources linked

Analysts review and approve instead of extracting and retyping.

Less manual intervention

Spreads arrive prebuilt with the source documents linked. Analysts review and approve instead of extracting and retyping.

Issues surfaced

up front

Flagged with evidence attached, before the credit call.

Cleaner lender review

Issues are flagged up front with the evidence attached, so lenders start from the judgment call, not the data entry.

Your workflow could be next.

Thirty minutes. A straight conversation about your systems and where custom AI would pay back first.