Autyn, an AI-native mortgage operations platform for independent US mortgage brokers, deployed LandingAI's Agentic Document Extraction (ADE) to automate income reconstruction on mortgage loan files, where it now delivers 94–98% field-level accuracy on borrower documents, processes 500+ loan files and 3,000+ documents in production, and cuts end-to-end income extraction of a full borrower package from one to two hours down to one to three minutes per file.
Autyn's pain concentrated in four places: letters of explanation (free-form, borrower-written, no template), paystubs (dozens of formats across employers and payroll providers), state and government-issued identity and benefit documents (no uniform format), and personal tax returns (Form 1040) with Schedule C, Schedule E, and K-1 attachments buried inside mixed PDF bundles. "Very often a loan officer who gets a borrower a solid preapproval fastest earns the deal and the real estate agent's referrals. Reconstructing income is the hardest part, and Agentic Document Extraction is the cornerstone to getting it right and traceable. Accurate data upfront means we can get a cleaner loan file to the underwriter and get to CTC faster. Get it wrong and everything downstream gets affected," said Myra D'Souza, CEO of Autyn.
Autyn's earlier OCR-plus-LLM pipeline broke on these documents: template-based extraction failed on unseen layouts, LLM-only extraction hallucinated with no source link to a page or location, and the system had no confidence signal so low-accuracy fields passed silently. LandingAI's ADE was selected over alternatives because it extracts against a customer-defined schema (predictable typed output even when document format varies), returns each field with an explicit source tag and a numeric confidence score (low-signal fields route to human review rather than passing silently), and ships with enterprise security and compliance posture including HIPAA, SOC 2 Type II, and a guarantee that customer data is not used for training.
Autyn integrated ADE directly via its API inside the platform's AIML layer, slotting it in as a dedicated extraction component without any custom model training. Since deployment, loan officers save two to four hours of manual work per file, per-document parse and extract runs in five to thirty seconds depending on length, and new document types ship faster as the team extends schema-first extraction across the borrower document set. The roadmap tightens cost per file through document-level change tracking (re-extracting only the swapped pages when a broker updates a single document) and broadens coverage to edge-case filings beyond the standard borrower set.
Details