Overview
Amazon Textract is AWS's document AI service – text extraction, form parsing, table extraction, invoice/receipt parsing, and ID document reading. The Automize AWS Textract connector covers the sync APIs (single-page) and async APIs (multi-page) for these workflows.
- Regions: Textract is regional with patchy availability – not every AWS region has it. Common: us-east-1, us-west-2, eu-west-1, eu-central-1. Document on S3 must be in the same region as the Textract call.
- Plans: Pay-as-you-go per page processed. Free tier: 1,000 pages/month for 3 months for new accounts.
Setting up the connection
Auth type: aws_sigv4.
- AWS Console → IAM → create user + access keys.
- Attach AmazonTextractFullAccess (or a scoped subset).
- If documents live in S3, also grant s3:GetObject on those buckets.
- In Automize, open Settings → Connectors → AWS Textract → Add connection. Paste keys + region.
Find your credentials at https://us-east-1.console.aws.amazon.com/iam/home.
Rate limits
Default 1 TPS for analysis APIs (10 TPS for detect-text), 25 active async jobs. Request increases via support. The connector retries on ProvisionedThroughputExceededException with exponential backoff.
Data model
Two API modes:
Sync APIs – single-page documents, return immediately
Async APIs – multi-page (PDFs >1 page), 2-step pattern:
Start → poll Get until status = SUCCEEDED → consume
blocks
Document input: either pass bytes (Bytes param) for files <5 MB, or
S3Object pointer for larger / async.Available operations
8 operations available. Click any row to jump to its detail.
| Operation | Category | What it does |
|---|---|---|
| Analyze Document | General | Analyze forms and tables (sync, single page) |
| Analyze Expense | General | Analyze invoices and receipts (sync) |
| Analyze ID | General | Analyze identity documents (sync) |
| Detect Text | General | Detect text in a document (sync, single page) |
| Get Analysis Result | General | Get async analysis result |
| Get Text Detection Result | General | Get async text detection result |
| Start Analysis (Async) | General | Start async document analysis |
| Start Text Detection (Async) | General | Start async text detection for multi-page docs |
Operations
Analyze Document
Analyze forms and tables (sync, single page)
No input parameters.
Tips
- More expensive – adds forms (KEY-VALUE pairs) and tables to the response.
- FeatureTypes: ['FORMS', 'TABLES', 'QUERIES', 'SIGNATURES']. QUERIES lets you ask natural-language questions about the doc.
Analyze Expense
Analyze invoices and receipts (sync)
No input parameters.
Tips
- Specialised for invoices + receipts. Returns ExpenseDocuments with normalised fields (TotalAmount, InvoiceDate, Vendor name).
Analyze ID
Analyze identity documents (sync)
No input parameters.
Tips
- Specialised for ID documents – passports, driver's licenses. Returns IdentityDocuments with normalised fields (FirstName, LastName, DocumentNumber, DateOfBirth).
Detect Text
Detect text in a document (sync, single page)
No input parameters.
Tips
- Cheapest, fastest – plain OCR. Returns words + lines with bounding boxes + confidence.
- Single-page sync. For multi-page PDFs, use Start Document Text Detection.
Get Analysis Result
Get async analysis result
No input parameters.
Tips
- Get the analyse blocks once SUCCEEDED.
Get Text Detection Result
Get async text detection result
No input parameters.
Tips
- JobId from Start. Returns the blocks (text + bounding boxes) for the full document.
- Use NextToken to paginate if >1,000 blocks.
Start Analysis (Async)
Start async document analysis
No input parameters.
Tips
- Async analyse – multi-page forms/tables/queries. Same start-then-poll pattern.
Start Text Detection (Async)
Start async text detection for multi-page docs
No input parameters.
Tips
- Async for multi-page docs. Document must be in S3.
- Returns JobId. Poll via Get Document Text Detection until JobStatus=SUCCEEDED.
- Or set NotificationChannel with SNS topic to get a push notification when done.
FAQ
- Sync vs async – which?
- Sync supports single-page only (PNG, JPG, or first page of a PDF up to 5MB). Async supports multi-page PDFs up to 10MB (PDF) or 5MB (image), or any S3-hosted doc up to 500MB. Use sync for receipts, async for contracts.
- Async – how do I know when it's done?
- Two options: poll Get every few seconds, OR pass NotificationChannel.SNSTopicArn so Textract publishes to SNS on completion. Polling is simpler; SNS scales better.
- Queries feature – when?
- Pass natural-language questions in the FeatureTypes=['QUERIES'] + Queries[] payload. e.g. 'What is the invoice total?' Returns answers extracted from the document. Costs extra; great for structured ingestion.
- Why does my form get parsed but lose context?
- Textract returns flat KEY-VALUE pairs without intrinsic ordering. Re-attach context via bounding box coordinates if order matters. Or use Queries to ask for specific fields.