Auto ML & Precedent
Fingerprinting
The system learns what your documents are.
Document classification shouldn’t require manual tagging or rigid keyword rules. Filing Cab’s fingerprinting engine uses machine learning trained on your organization’s actual filing history to classify every incoming document — and it gets smarter with every decision your team makes.
150+ Document Type Taxonomy
A comprehensive classification system spanning 18 categories and over 150 specific document types — from W-2s and 1099s to board resolutions, lease agreements, shipping manifests, and medical records. Each type has defined identification signals, expected metadata fields, and routing dispositions. The taxonomy is extensible: add your own types as your needs evolve.
Visual Signal Analysis
Documents have visual fingerprints — logos, layouts, form structures, table patterns, signature blocks. The fingerprinting engine analyzes visual features to identify documents even when text content is ambiguous. A W-2 looks like a W-2 regardless of the employer. An invoice has a recognizable structure regardless of the vendor.
Textual Signal Analysis
Beyond visual patterns, the engine analyzes text content for classification signals — header text, field labels, regulatory language, standardized form numbers, and domain-specific terminology. Combined with visual analysis, this dual-signal approach achieves classification accuracy that exceeds what either method could deliver alone.
Historical Precedent & Continuous Learning
Every time a human confirms, corrects, or overrides a classification, the system learns. Documents from recurring senders converge on correct classifications faster. Seasonal patterns — tax season, annual renewals, quarterly reports — are recognized and anticipated. The model is scoped to your organization’s data, not a generic one-size-fits-all classifier.
Confidence Scoring
Every classification comes with a confidence score. High-confidence matches are filed automatically. Low-confidence matches are routed to human reviewers with the top candidate types and reasoning. You set the thresholds — aggressive automation for high-volume low-risk documents, cautious review for sensitive or unusual filings.