The company says the system automatically handles more than 30% of credit-risk account reviews that had previously been entirely manual.
The pipeline standardizes borrower financial statements for downstream credit decisions. It extracts line items and reporting periods, then maps them to Ramp’s semantic definitions, including industry-specific exceptions. For digital PDFs and spreadsheets, Ramp says it reads underlying calculation logic; for scans and images, it uses multiple OCR passes. Deterministic code recalculates metrics such as EBITDA, checks for double counting, reconciles duplicates, and flags fallback values when inputs are missing.
Ramp evaluated seven frontier models on 200 income statements and balance sheets previously reviewed by credit analysts. Its reported target measures were precision and recall within a 5% tolerance. The company says the pipeline produced median recall gains of 3.7 percentage points on income statements and 5.9 points on balance sheets.
The post does not provide post-deployment error rates, credit-loss outcomes, the exact automation decision boundaries, or customer access to the system.