Sensible Blog | Document Extraction Engineering

FEATURED

What is intelligent document processing?

How intelligent document processing works, where it differs from OCR and RPA, and why production accuracy is hard.

Apr 29, 2026

Read article →

Tutorial

2026-04-01

Delivery Orders & PODs

Logistics

Tutorial

How to extract data from delivery orders with Sensible

Learn how to extract vessel details, port routing, and cargo data from delivery orders using Sensible's SenseML. Handles carrier format variability with a generalized LLM config and OOCL-specific deterministic layout template through a single API.

Apr 1, 2026

Tutorial

2026-04-01

Insurance Quotes

Insurance

Tutorial

How to extract data from cyber insurance quotes with Sensible

Pull limits, sublimits, retentions, and premium from cyber insurance quotes, with configs built for multi-page carrier formats.

Apr 1, 2026

Tutorial

2026-04-01

Declaration Pages

Insurance

Tutorial

How to extract data from insurance declaration pages with Sensible

Learn how to extract named insured, coverage details, liability limits, and vehicle data from insurance declaration pages using Sensible's SenseML. Handles carrier format variability with a generalized LLM config and carrier-specific layout templates through a single API.

Apr 1, 2026

Tutorial

2026-03-12

Invoices

Cross-industry

Tutorial

How to extract data from invoices with Sensible

Extract vendor details, line items, and payment data from invoices of any format using Sensible's hybrid architecture. Schema-validated output with confidence signals per field.

Mar 12, 2026

Case Study

2026-03-04

EOBs

Healthcare

Insurance

Case Study

How a healthcare benefits company replaced a failing vendor and went live with 30 configurations in weeks

A healthcare benefits company processing Explanation of Benefits (EOB) documents from 20–40 insurance carriers needed a fast, accurate way to extract and normalize data across all those formats. By using Sensible's layout-based extraction methods alongside a postprocessing layer to enforce a consistent output schema, they went from signed contract to 30 live configurations in under two months. The implementation succeeded by matching deterministic extraction techniques to well-structured documents, proving that deterministic methods, applied thoughtfully, outperform more complex LLM approaches when the documents don't require them.

Mar 4, 2026

Case Study

2026-03-04

Payroll Reports

Construction

Case Study

How a construction compliance company tackled "impossible" payroll report extraction

A construction compliance software company had given up on automating payroll report data extraction after every vendor they tested failed to handle the documents' OCR challenges, format inconsistencies, and dense tabular layouts. Sensible solved the problem by combining deterministic layout-based methods with targeted LLM extraction for trouble spots, along with aggressive text preprocessing and customer-specific configurations rather than one-size-fits-all templates. The key lesson: hybrid approaches and "good enough" automation with some manual review beat either holding out for a perfect universal solution or resigning yourself to pure manual entry.

Mar 4, 2026

Case Study

2026-02-24

Corporate Filings

Legal & Compliance

Case Study

How a Spanish compliance startup automated extraction from 80-page corporate documents

A Spanish compliance startup faced a daunting data extraction problem: pulling structured shareholder and transaction data from dense, 80-page corporate deeds written in multiple languages. Their solution combined deterministic document slicing with LLM-based extraction and intelligent name-matching to merge scattered information into unified records. The result cut an hours-long manual process down to minutes, with roughly 50 data points reliably extracted per document.

Feb 24, 2026

Case Study

2026-02-24

Void Checks

Financial Services

Case Study

How a point-of-sale lender tackled the messiest document type in financial services

A Canadian point-of-sale lender needed a faster, cheaper way to extract banking data from void checks—documents that routinely arrive as blurry photos, screenshots, and degraded scans. Their solution layers template-based extraction for major banks, regex parsing for standardized MICR data, and LLM-based extraction for unstructured fields, with validation logic that returns null rather than a plausible-looking wrong answer. The result replaced a slow, expensive human-review process with one that completes in seconds—at significantly lower cost and with broader bank coverage.

Feb 24, 2026

Thought Leadership

2026-02-17

No items found.

Cross-industry

Thought Leadership

The "prompt janitor" problem: Why hybrid document processing is the future

LLM-based document extraction systems create a hidden "prompt janitor" problem where teams spend $140K-$210K annually maintaining prompts, handling model deprecations, and managing validations—costs that far exceed initial development estimates. This article argues that pure LLM approaches should be replaced with hybrid architectures combining LLM reasoning with deterministic validation to ensure reliability and governance. It announces a white paper with more details about buy v.s. build decisions for hybrid approaches.

Feb 17, 2026

Case Study

2026-01-27

Medical Records

Healthcare

Case Study

How two healthcare organizations automated document processing

Two healthcare organizations automated their document processing using Sensible. Company A, processing cardiac device reports for organ donation screening, built 50+ extraction configurations entirely on their own through self-service, reaching production faster than almost any other customer. Company B, a pathology lab handling handwritten requisition forms, partnered closely with Sensible's team to build configurations with sophisticated normalization logic that could handle handwriting and variable scan quality. Both chose deterministic, layout-based extraction methods (Company A for transparency and independence, Company B for consistency and accuracy), while Company B also built a generalized LLM extraction configuration for edge cases.

Jan 27, 2026

Case Study

2026-01-27

Utility bills

Telecom Bills

Cross-industry

Case Study

How two companies tackled utility and telecom bill processing

Two companies tackled utility and telecom bill processing with Sensible, but their implementations couldn't look more different. One processes thousands of bills monthly across 2,000+ carriers, requiring complex conditional logic to handle schema variations—like associating charges with meters even when they appear in separate sections of the document. The other processes documents stretching to tens of thousands of pages, using custom engineering to intelligently split and reassemble massive files while preserving data integrity. Both chose deterministic, layout-based extraction over LLMs because their documents were too complex and their schema requirements too strict for probabilistic approaches.

Jan 27, 2026

Case Study

2026-01-08

Bank Statements

Financial Services

Case Study

How two financial companies automated bank statement processing

Two financial companies automated bank statement processing using Sensible's platform. A UK financial forensics firm required sophisticated data normalization to handle messy, complex documents from legal cases, while a Canadian lending platform prioritized speed and cost-efficiency for processing high volumes of cleaner statements.

Jan 8, 2026