AWS Launches Amazon Textract Insights for Verifiable Document Extraction

By The Indus Pulse Editorial Team4 min read

Amazon Web Services announced the general availability of Amazon Textract Insights, a generative artificial intelligence feature inside Amazon Textract accessed through the StartDocumentInsights API, according to an AWS blog post. Organizations across financial services, insurance, healthcare, and the public sector can describe target data fields in plain language and submit either individual files or mixed multi-document packets without training a custom model, AWS said. Unlike custom OCR-and-model pipelines that teams must operate themselves or general-purpose large language models that may omit source pages and calibrated scores, Textract Insights returns each value with a confidence score alongside page-level and bounding-box coordinates, according to AWS.

Blueprints, Multi-Document Packets, and Symbeo Evaluation

Users define extraction rules using a JSON Schema blueprint that specifies whether each field should be read verbatim as an explicit value or derived as an inferred value from document content, AWS said. AWS provides public blueprints for W-2 forms, Form 1040, Form 1098, passports, and birth certificates, while also allowing custom blueprints for specialized business records. Teams can submit a single file of up to 3,000 pages from an Amazon S3 bucket with splitting enabled so the service separates sub-documents, matches each to its blueprint, and writes results back to S3, AWS said. For instance, a 26-page mortgage PDF containing a W-2, payslips, bank statements, and handwritten loan officer notes can be processed in one StartDocumentInsights request, according to AWS. Because every extracted value links back to its page and coordinates, reviewers and auditors can verify results against the original document and route only lower-confidence fields for manual checks, AWS said. Vijay Agrawal, vice president of information systems at Symbeo, A CorVel Company, said in the AWS post that his team evaluated the service on varied medical bills and forms so staff can focus review on lower-confidence fields. In US West (Oregon), standard AnalyzeDocument pricing for one million pages is $0.015 per page for Queries alone and $0.070 per page when combining Tables, Forms, and Queries ($0.055 per page after one million pages), according to Amazon Web Services. Because Textract Insights bills per page, teams can control costs by removing blank pages upstream before submission and avoiding resubmitting unchanged files during confidence-threshold tuning, according to explainx.ai.

Supported Formats, Security, and Core Textract Updates

Amazon Textract Insights accepts PDF, PNG, JPEG, and TIFF files in English, French, German, Italian, Portuguese, and Spanish, and is billed per page rather than per token with no minimum commitment, AWS said. Projects and blueprints are encrypted at rest using AWS owned keys or customer-managed AWS Key Management Service keys specified via EncryptionConfiguration, while transient intermediate job data is deleted once processing finishes, according to AWS. Inference requests are automatically routed across AWS Regions within the same geographic area where the request originated, AWS said. Separately, AWS updated Amazon Textract Core's DetectDocumentText and AnalyzeDocument APIs to support 40 languages and expanded Core availability to Asia Pacific (Tokyo) and Europe (Zurich), bringing in-region Core coverage to 18 AWS Regions, though Textract Insights is not yet available in those two new Regions. Amazon Textract's 18 AWS Regions span US, Asia Pacific, Canada, Europe, and GovCloud locations, according to Amazon Web Services. On April 21, 2022, AWS added natural-language Queries support to Amazon Textract's AnalyzeDocument and GetDocumentAnalysis operations, according to Amazon Web Services. Because output objects persist under the customer's S3 retention policy even though intermediate job artifacts are transient, teams handling sensitive data are advised to pair Textract Insights with AWS KMS customer-managed keys, according to explainx.ai. Cross-Region inference routes compute within a geographic area without replacing data-residency contracts, leaving legal teams to verify which AWS Region stores output objects at rest, according to explainx.ai.

Sources & Citations

Spotted an error in this article? Tell us and we will check it.