AWS brings Amazon Textract Insights, a generative AI feature for document processing, to general availability
Amazon Textract Insights is a new generative-AI feature of the Amazon Textract service. It lets you describe the required fields in natural language and extract data from mixed documents of up to 3000 pages with confidence scores and page-level references, without training a model.
Amazon has announced the general availability (GA) of Amazon Textract Insights, a new generative AI feature within the Amazon Textract service, available via a standalone API. The user describes the required fields in natural language and submits a document or a mixed package of multiple documents; the service returns structured data, with each value assigned a confidence score and a reference to the specific page (page-level grounding). According to the company, none of the features require model training.
Extraction is defined using a so-called blueprint, which is a JSON schema where each field contains a natural-language description and an inference type that determines whether the value is read directly from the text or inferred from the document's content. Public AWS blueprints are available for common document types (W-2, Form 1040, Form 1098, passport, birth certificate), and custom blueprints can also be created for company documents. For mixed document packages (e.g., a mortgage application containing a W-2, pay stubs, bank statements, and handwritten notes), multiple blueprints can be added to a single project and the splitting feature enabled — the service then splits a package of up to 3000 pages from Amazon S3 into individual logical documents, assigns each to the correct blueprint, and returns the extracted fields with confidence scores and grounding data.
The source cites a reference from Symbeo (a CorVel Company), according to which the new feature allows manual review to be focused only on fields with lower confidence, and the company also expects benefits from the multi-document splitting feature. At the same time, AWS updated the core text engine Amazon Textract Core — improving accuracy, handling of dense tables and handwritten text, and expanding support to 40 languages for the DetectDocumentText and AnalyzeDocument APIs, while existing APIs remain unchanged. Textract Core is now also generally available in the Asia Pacific (Tokyo) and Europe (Zurich) regions, bringing the total to 18 AWS regions. The rest of the article was not available; for details, see the source article.
Why it matters
Teams processing documents (mortgages, insurance claims, onboarding) have so far faced a dilemma between a custom pipeline (OCR, classification, prompting, review) and a general-purpose LLM, which doesn't provide reliable source references or confidence scores. According to AWS, Textract Insights offers both in a single API — the ability to automatically approve high-confidence values and route only uncertain items for review, reducing the volume of manual checking while maintaining auditability.
Two audiences, two different impacts
What this means
For individuals
Developers working with AWS can replace a custom OCR/classification pipeline with a single API call and define extraction fields in natural language without training a model.
For a business
Companies processing large volumes of mixed documents (financial services, insurance, healthcare) can reduce manual review costs by extending automation even to documents with lower extraction confidence, thanks to confidence scores and page-level grounding for auditing.
ProcessesCheck the original
Event sources
clearly official source · 1 publisher, 0 independent. We count feeds from the same owner only once.