AI Audit Document Extraction Agent

How Cassidy automates using AI

Step 1: Trigger on PBC upload or request deadline

The Workflow activates when new documents arrive via PBC portal, email, SFTP, or VDR—or when request deadlines approach without client response.

Step 2: Ingest and normalize documents

Cassidy de-duplicates files, applies consistent naming conventions, runs OCR on PDFs and images, and extracts metadata including source system, report name, run date, and parameters.

Step 3: Classify and extract key fields

The agent classifies documents by workstream and type—invoices, bank statements, reconciliations, confirmations, contracts—then extracts dates, amounts, counterparties, control totals, and signature evidence.

Step 4: Build IPE completeness and accuracy evidence

Cassidy captures and stores report lineage, parameter screenshots, SQL definitions where available, and computes control totals with checksums—creating reproducibility packs ready for inspection.

Step 5: Perform tie-outs and reconciliations

The Workflow maps extracted data to the trial balance and general ledger, auto-performs footing and cross-footing, flags variances, and surfaces breaks between sub-ledger extracts and GL balances.

Step 6: Generate Excel-first workpapers

Cassidy produces lead schedules with cell-level links to immutable evidence copies, embedded tick-mark legends, automated WP indices, and formula protection—preserving full audit trail lineage.

Step 7: Deliver inspection-ready packages

Final binders export to your audit management system with checksums, timestamps, access logs, and an evidence matrix mapping each conclusion to procedures and supporting artifacts.