MetricSign
EN|NLRequest Access
High severityexecution

Power BI Refresh Error:
DOCUMENT_AI_EXTRACTION_FAILED

What does this error mean?

The Snowflake Document AI AI_EXTRACT function failed to extract structured data from a document, usually because the document format is unsupported, the model build is not published, or the stage path is invalid.

Common causes

  • 1The Document AI model build has not been published; only published builds can be called via AI_EXTRACT
  • 2The document file format is not supported (Document AI requires PDF or supported image formats)
  • 3The Snowflake stage storing the documents uses a non-SSE encryption type, which is required for presigned URL generation
  • 4The role calling AI_EXTRACT does not have USAGE privileges on the Document AI model

How to fix it

  1. 1Step 1: In Snowsight, open the Document AI project and confirm the model build is in Published status.
  2. 2Step 2: Verify the input files are in a supported format (PDF, PNG, JPG, TIFF) and that the stage uses SNOWFLAKE_SSE encryption.
  3. 3Step 3: Confirm the calling role has USAGE on the Document AI model object: GRANT USAGE ON MODEL <name> TO ROLE <role>.
  4. 4Step 4: Test with a single, simple document to isolate whether the issue is format-related or model-related.
  5. 5Step 5: Review the Document AI troubleshooting guide in the Snowflake documentation for build-specific issues.

Frequently asked questions

How do I publish a Document AI model build?

In Snowsight, open the Document AI project, select a trained model build, review its accuracy metrics, and click Publish. Only one build can be the active published version at a time.

What document formats does Snowflake Document AI support?

Document AI supports PDF, PNG, JPG, JPEG, and TIFF. Multi-page PDFs are processed page by page. Password-protected PDFs are not supported.

Other execution errors