Extract loan packages to structured JSON

Loan packages bundle dozens of document types into a single PDF: applications, pay stubs, bank statements, tax returns, and disclosures. Sensible classifies, splits, and extracts each document for automated underwriting, compliance checks, and loan boarding.

Quick links:

Financial services solutions

API documentation

Extracting data from mortgage portfolios

Schedule a demo

Why loan packages demand document intelligence

Hundreds of pages. 20+ document types. No table of contents. Extraction starts with classification.

Document Classification

No table of contents. No consistent ordering. A 200-page file mixes tax returns, pay stubs, W-2s, bank statements, and disclosures. Classification identifies each document type, then the appropriate extraction config takes over.

Mixed Document Quality

A single loan package mixes clean digital PDFs with faxed copies, mobile photos of documents, and degraded scans. Some pages need layout-based extraction; others need LLM parsing to read through noise. Sensible's hybrid engine and configurable OCR engines handle the full quality range without manual sorting.

Consistent Schema Across Document Types

Pay stubs, W-2s, bank statements, and disclosures each get their own extraction config, but all output validated JSON against your target schema. When the borrower's name, income, and property address appear across ten document types, your pipeline can cross-check them programmatically because every extraction is structured the same way.

Fields we extract

Fields span all document types in the package. The schema unifies borrower, loan, and compliance data.

Borrower information

Borrower name, SSN (masked), address, employer, income, assets, liabilities (from application and verification docs)

Loan terms

Loan amount, interest rate, loan type, term, amortization, LTV, DTI, property type, property address, appraised value

Compliance and conditions

Closing costs, APR, TRID tolerance checks, conditions to close, title exceptions, insurance requirements


{ /* SenseML: loan package extraction */
  "fields": [
    {
      "method": {
        "id": "queryGroup",
        "queries": [
          {
            // Borrower name
            "id": "borrower_name",
            "description": "borrower name, applicant name, borrower"
          },
          {
            // Loan amount
            "id": "loan_amount",
            "description": "loan amount, principal amount, amount financed",
            "type": { "id": "currency" }
          },
          {
            // Interest rate
            "id": "interest_rate",
            "description": "interest rate, note rate, APR",
            "type": {
              "id": "custom",
              "pattern": "[0-9]{1,2}\\.?[0-9]{0,3}%"
            }
          },
          {
            // Property address
            "id": "property_address",
            "description": "property address, subject property, collateral address"
          }
          // Additional fields for loan type, term, LTV, borrower income, etc.
        ]
      }
    }
  ]
}
Commercial Loan Package

Business lending package including financials, guarantor docs, and collateral.

SBA Loan Package

Small Business Administration loan documentation with SBA-specific forms.

Residential Mortgage Package

Complete loan file with application, disclosures, verifications, and title documents.

Supported loan package types

Sensible handles conventional, FHA, VA, USDA, jumbo, and commercial loan packages. The classifier detects 20+ document types, and hybrid extraction ensures each gets the right configuration.

By loan type

Conventional, FHA, VA, USDA, jumbo, HELOC, commercial, SBA, construction

By workflow

Origination, underwriting, closing, post-closing QC, due diligence, securitization

Trusted by operations and engineering teams at

Common Questions

Answers about document classification, portfolio processing, and multi-document extraction.

Can Sensible process multiple loan files in a portfolio?

Yes. You can submit multiple loan package PDFs via the API and receive structured data for each. This supports portfolio due diligence, bulk onboarding, and audit workflows.

Is there a page limit for loan package processing?

Sensible handles loan packages of any page count. Packages of 200+ pages are common in mortgage workflows. Processing time scales linearly with page count.

What document types can Sensible identify within loan packages?

Sensible classifies each document within a loan package by type: tax returns, pay stubs, bank statements, W-2s, 1099s, VOEs, closing disclosures, appraisals, title reports, and more. You can add custom document types as needed.

How does Sensible handle loan packages with many document types?

Sensible classifies each document within a loan package PDF, splitting it into individual documents like pay stubs, tax returns, bank statements, and closing disclosures. Each is then extracted with the appropriate configuration.

Do you support webhooks?

Yes. Sensible sends extraction results to your webhook endpoint when processing completes. You can also poll the API for status.

Does Sensible support human review?

Yes. Sensible flags extractions with low confidence for human review. You can configure review thresholds and workflows.

What security certifications does Sensible have?

Sensible is SOC 2 Type II certified and HIPAA compliant. Data is encrypted in transit and at rest.

How long is document data retained?

Document data is stored indefinitely by default. Custom retention policies are available and can be configured for same-day deletion if needed.

Is there a free trial?

Yes. Sensible offers a 14-day free trial on the Growth plan. No credit card required to start.

How is pricing structured?

Sensible uses per-document pricing for predictable costs. No token-based billing or usage surprises. Volume discounts are available for higher throughput.

How do I integrate with Sensible?

Sensible provides REST APIs and SDKs for Python and Node.js. Most integrations take a few hours. Webhooks, Zapier, and direct API calls are all supported.

What file formats does Sensible support?

Sensible processes PDFs (native or scanned), Microsoft Word (DOC, DOCX), spreadsheets (XLSX, XLS, CSV), single-page images (JPEG, PNG), multi-page images (TIFF), and email bodies with attachments.

How accurate is the extraction?

Accuracy depends on document quality and configuration. Most production deployments achieve 95%+ accuracy with proper validation rules and confidence signals.

How fast is document processing?

Processing speed depends on document size, page count, OCR requirements, and which extraction methods are used. Simple single-page documents process in seconds. Larger or more complex documents that use LLM-based extraction take longer.

Resources

Documentation Links

Docs: Insurance template quick startDocs: Validation and human reviewDocs: API and delivery options

Related insurance documents

Docs: Insurance template quick startDocs: Validation and human reviewDocs: API and delivery options