Complete AI Training
Start for Free

Your job's AI kit

Get your AI kit

Tell us who you are and what you do. We show you your kit right away and email you the link: skills, prompts, AI agents, MCP servers and courses for your job.

500+ jobs ready, and we make a kit for any other job. No payment needed to look.

Share

AI news ·

Amazon Textract Insights brings verifiable generative AI to document processing

Amazon Textract Insights extracts fields from mixed documents using natural language descriptions, returning each value with a confidence score and page-level grounding. The service processes up to 3,000-page files and spans 40 languages for core text detection.

Amazon has released Amazon Textract Insights, a new generative AI feature now generally available through the Amazon Textract API. The service lets organizations describe the fields they want in natural language and receive structured, auditable results-each carrying a confidence score and page-level source grounding-with no model training required.

The feature targets a persistent problem in financial services, insurance, healthcare, and the public sector: extracting reliable data from mixed document packets. A single mortgage file, for example, often arrives as one scan containing a W-2, payslips, bank statements, and handwritten notes. Claims files and onboarding packages present similar challenges, mixing typed forms, dense tables, receipts, and free-text letters in no fixed order.

How the workflow changes

Teams have typically chosen between two imperfect paths. One involves building a custom pipeline that strings together optical character recognition, document splitting, classification, a foundation model prompt, and a manual review step. The team must integrate, scale, and operate every component. The other path uses a general-purpose large language model that returns fluent answers but does not reliably cite the source page or provide a confidence score. Without traceable evidence and a calibrated score, deciding which values to accept automatically becomes guesswork.

Amazon Textract Insights collapses this workflow into a single asynchronous API call. Users define extraction requirements in a blueprint-a JSON Schema where each field carries a natural language description and an inference type that specifies whether the value is read verbatim from the page or inferred from the content. AWS provides public blueprints for common documents such as W-2s, Form 1040, Form 1098, passports, and birth certificates. Teams can also author their own blueprints for business-specific documents.

For a mixed packet, users add the relevant blueprints to a project, submit a file of up to 3,000 pages from their Amazon S3 bucket, and enable splitting. The service splits the packet into logical sub-documents, matches each to its blueprint, extracts the requested fields, and writes results-each with a confidence score and bounding-box coordinates-back to the customer's S3 bucket.

Confidence and grounding for auditable automation

Amazon Textract Insights returns two field types. Explicit fields are read directly from the page. Inferred fields are derived from the document's content. Both carry a confidence score and grounding that identifies the exact page and coordinates. An auditor or reviewer can confirm any result against the original document. This traceability lets teams extend automation to documents that previously required manual review, while keeping people in the loop where judgment is needed.

Vijay Agrawal, Vice President of Information Systems at Symbeo, A CorVel Company, said the service's approach fits their operational reality. "Our medical bills and forms are received from a wide variety of sources and often have different layouts and formats. As a result, extraction rules developed for one source do not always transfer effectively to another," Agrawal said. "With Amazon Textract Insights, we can describe the fields we need in plain language, and the service identifies and extracts the corresponding data. Each extracted value is returned with a confidence score, allowing our team to focus their review efforts only on the fields where the service has lower confidence."

Engine updates and regional expansion

The release also includes an update to the underlying Amazon Textract Core reading engine. Accuracy has improved, handling of dense tables and handwriting is stronger, and language coverage now spans 40 languages for the DetectDocumentText and AnalyzeDocument APIs. These improvements apply to existing customers with no API changes required. Amazon Textract Core is now generally available in two additional AWS Regions-Asia Pacific (Tokyo) and Europe (Zurich)-bringing in-region processing to 18 Regions total. Amazon Textract Insights is not yet available in those two new Regions.

The service processes documents in English, French, German, Italian, Portuguese, and Spanish. It supports PDF, PNG, JPEG, and TIFF inputs and is priced per page rather than per token, with no minimum commitment.

Why this matters for operations and finance teams

For teams in insurance, mortgage processing, claims management, and healthcare administration, the combination of natural language field descriptions and auditable confidence scores changes the automation calculus. Rather than accepting a black-box extraction or maintaining a brittle custom pipeline, a team can set a confidence threshold calibrated on its own documents. Values above the threshold flow through automatically; values below it go to a reviewer who can verify the result against the exact page and bounding box the service cites. The machine handles the volume. The human handles the judgment calls-and has the evidence to make them quickly.

Those looking to build the skills to design blueprints and evaluate confidence outputs may find relevant grounding in Generative AI Courses. Teams in regulated sectors can also explore AI for Insurance Courses to connect extraction workflows to compliance and audit requirements.

Share