How the scan works
Documents go in, defensible findings come out. The pipeline reads every page with OCR, uses AI to extract line items and tie them to the lease, then runs deterministic checks that either surface a measurable discrepancy or do not. A typical scan returns findings in under 15 minutes.
The pipeline, stage by stage
OCR per document
Each uploaded statement, lease, and ledger runs through Textract OCR in parallel. Scanned pages, tables, and line items become structured, machine-readable text.
AI extraction and normalization
An AI model reads the OCR output, pulls every billed line item, ties each expense to the lease clause that governs it, and normalizes the figures into a consistent structure. This is the only place AI touches the audit.
20 deterministic detection checks
The detection engine runs 20 checks across four pillars: math accuracy, classification correctness, GL and financial integrity, and compliance and advisory. Every calculation is deterministic Python. No model guesses at a number.
Branded findings and dispute draft
Findings compile into a branded report and a correction package, each grounded in the documented line items and the lease clauses reviewed, ready under your firm's brand.
Why AI never touches the math
AI is good at reading documents and bad at arithmetic you can be sued over. So the line is drawn hard: AI handles extraction and classification, reading the lease and labeling line items. Every dollar figure, allocation, cap test, and reconciliation is computed by deterministic Python. Run the same documents twice and the numbers are identical, which is exactly what a defensible finding requires.