Amazon Textract
by AWS · Invoice & Document OCR
Cloud OCR and document data extraction API for forms, tables, and IDs
- Works with
- Amazon S3, AWS Lambda, Amazon A2I (Augmented AI), Amazon Comprehend
- Deployment
- Cloud
- Company size
- SMB, Mid-market, Enterprise
- Pricing
- Pay-as-you-go per page
Overview
Amazon Textract is a fully managed machine learning service from AWS that extracts printed text, handwriting, layout elements, and structured data from scanned documents and images. It goes beyond traditional optical character recognition by identifying the relationships between document elements, returning form key-value pairs, table rows and columns, and natural-language query answers rather than a flat stream of characters. Every extracted element is returned with a confidence score and bounding-box geometry, which lets downstream applications route low-confidence results to human review.
The service is delivered entirely as an API with no servers to provision and no ML model to train. Synchronous operations process single-page images or short documents in real time, while asynchronous operations handle multi-page PDFs stored in Amazon S3. Purpose-built APIs add document-type awareness on top of the general OCR engine: AnalyzeExpense parses invoices and receipts, AnalyzeID reads passports and US driver's licenses, and AnalyzeLending classifies and extracts mortgage packages. A Custom Queries adapter capability lets teams tune extraction for their own document formats without building a model from scratch.
Textract is priced per page on a pay-as-you-go basis with a three-month free tier for new AWS accounts, and it inherits the AWS compliance and security posture, including HIPAA eligibility and FedRAMP authorization. It is most often adopted by teams already operating on AWS who want to embed document extraction into a larger pipeline built from S3, Lambda, and other AWS services, rather than buy a packaged end-to-end document-processing application.
Screenshots & demo
Demo video from the vendor's YouTube channel.
Features & capabilities
Text and handwriting detection
General-purpose OCR via the DetectDocumentText API.
- Printed text extraction across varied fonts and layouts
- Handwriting recognition (English)
- Line and word-level output
- Per-element confidence scores
- Bounding-box geometry for every detected item
- Reading-order-aware output
Document analysis (AnalyzeDocument)
Structured extraction beyond raw OCR.
- Form extraction as key-value pairs with preserved context
- Table extraction preserving row and column relationships
- Selection element (checkbox / radio button) detection
- Signature detection with location and confidence
- Layout extraction of titles, paragraphs, headers, footers, and lists
- Combine multiple feature types in a single call
Natural-language queries
Ask questions of a document instead of mapping field coordinates.
- Plain-language queries such as 'What is the customer name?'
- Pre-trained on paystubs, bank statements, W-2s, loan forms, and insurance cards
- Up to 15 queries per page (synchronous) or 30 per page (asynchronous)
- Custom Queries adapters to tune extraction for proprietary formats
Specialized document APIs
Document-type-aware extraction endpoints.
- AnalyzeExpense for invoices and receipts (vendor, totals, line items, payment terms)
- AnalyzeID for passports and US driver's licenses
- AnalyzeLending for automated mortgage and loan document classification and extraction
- Normalized field outputs for common expense and identity fields
Processing, integration, and review
How extraction runs and connects into a pipeline.
- Synchronous API for real-time single/short documents
- Asynchronous API for multi-page PDFs in Amazon S3
- Output as JSON, plus CSV export for forms and tables in the console
- Native integration with AWS Lambda, Amazon Comprehend, and SageMaker
- Human-in-the-loop review via Amazon Augmented AI (A2I)
- API activity logging through AWS CloudTrail
Security and governance
Data protection and access controls.
- Encryption of content in transit and at rest
- VPC endpoints via AWS PrivateLink
- IAM-based access control
- Opt-out of data use for service improvement via AWS Organizations policy
- Customer-requested content deletion through AWS Support
Common use cases
- Automating invoice and receipt capture for accounts payable
- Extracting fields from mortgage and loan application packages
- Processing health intake forms and insurance claims
- Reading identity documents during customer onboarding / KYC
- Digitizing government, tax, and compliance forms
- Indexing scanned document archives for search
- Feeding extracted data into custom document-processing pipelines on AWS
Strengths & considerations
Strengths
- Delivered purely as an API with no model training or infrastructure to manage
- Purpose-built endpoints for invoices, IDs, and lending on top of general OCR
- Returns confidence scores and bounding boxes on every element for downstream routing
- Tight integration with the broader AWS data and ML stack (S3, Lambda, Comprehend, A2I)
- Pay-per-page pricing with no upfront commitment or minimum fees
- Inherits AWS compliance posture including HIPAA eligibility and FedRAMP authorization
Considerations
- Handwriting, invoices, receipts, identity documents, and Queries support English only
- It is a building-block API, not a packaged end-to-end document workflow application
- Per-page costs can become significant at very high volume without custom pricing
- AnalyzeID supports US driver's licenses and passports rather than all global ID types
- Realizing full value typically assumes building on AWS (S3, Lambda, IAM)
ERP integrations
Source documents and asynchronous output stored in S3 buckets.
Common orchestration layer to call Textract and route results.
Human-in-the-loop review of low-confidence extractions.
Extracted text passed to Comprehend for entity / classification analysis.
Pricing
Three-month free tier for new AWS customers (e.g. 1,000 pages/month DetectDocumentText; 100 pages/month Forms/Tables/Layout/Queries; 100 pages/month AnalyzeExpense and AnalyzeID; 2,000 pages/month AnalyzeLending). Paid per-page rates vary by API: DetectDocumentText from $0.0015/page; AnalyzeDocument Forms $0.05/page, Tables $0.015/page, Queries $0.015/page, Signatures $0.0035/page; AnalyzeExpense $0.01/page; AnalyzeID $0.025/page; AnalyzeLending $0.07/page. Volume tiers and custom pricing available; prices vary by region. Get an independent shortlist with pricing guidance below.
Technical & security
- Hosting
- AWS-hosted managed service
- Compliance
- HIPAA eligible, SOC, PCI DSS, ISO, FedRAMP (Moderate and High), GDPR
- Languages
- English, German, French, Spanish, Italian, Portuguese
About the vendor
- Ownership
- Subsidiary of Amazon.com, Inc.
- Notable customers
- Change Healthcare, Roche, Elevance Health, Pennymac, Cox Automotive, The Washington Post, PitchBook
Alternatives to Amazon Textract in Invoice & Document OCR
Amazon Textract — frequently asked questions
What file formats does Amazon Textract support?
Textract accepts PNG, JPEG, TIFF, and PDF. Synchronous operations accept images as S3 objects or raw byte arrays, while asynchronous operations require documents stored in Amazon S3. For PDFs each page is billed separately; single images count as one page.
How is Amazon Textract priced?
Pricing is pay-as-you-go per page with no upfront commitment, and rates differ by API (for example DetectDocumentText starts around $0.0015 per page, while Forms extraction is $0.05 per page). New AWS accounts get a three-month free tier, and high-volume customers can request custom pricing.
Is Amazon Textract HIPAA compliant?
Textract is HIPAA eligible. With an executed Business Associate Agreement (BAA) with AWS, it can be used to process protected health information. It is also FedRAMP authorized and compliant with PCI, ISO, and SOC.
What languages does Amazon Textract support?
General text detection supports English, German, French, Spanish, Italian, and Portuguese. Handwriting, invoices, receipts, identity documents, and Queries are English-only.
Does Amazon Textract require training a model?
No. Textract is a pre-trained, fully managed API, so there is no model to build or infrastructure to provision. The optional Custom Queries adapter feature lets you tune extraction for proprietary document formats without building a model from scratch.
Evaluating Invoice & Document OCR?
Tell us your ERP and requirements and we'll send an independent shortlist — including Amazon Textract and the best-fit alternatives — with honest pros and cons.