Skip to content
Dynamite Docs logoDynamite Docs
The handoff
Supplier invoice: Infer, correct, rememberInfer the fields. Correct the values. Remember the pattern. Illustrative example, not extracted customer data.FROM SOURCE FILE TO A RESULT YOU CAN USESupplierinvoiceReview workspaceInfer,correct,remember1Infer the fields2Correct thevalues3Remember thepattern
  1. 01Infer the fields
  2. 02Correct the values
  3. 03Remember the pattern
Saved corrections guide similar layouts. Illustrative example.
03Updated

Extraction is where Dynamite Docs earns its keep. It operates on the core loop of infer → correct → remember:

1

Infer

The AI reads the document, recognizes the layout, and infers a schema that identifies which columns matter and what each value is.

2

Correct

You review the table in the Data Studio and edit anything that is wrong. You can also add or remove columns to match your needs.

3

Remember

Every correction is saved as a reusable pattern. The next document with the same layout is extracted more accurately, with less review time.

Prerequisites

  • At least one document uploaded and visible in the Library sidebar.
  • Available Page-Equivalents (PE) in your monthly balance (1 PE per processed page) or a connected BYOK key.
  • Document page count within tier limit (Free: 5 pages; Starter: 50 pages; Pro: 500 pages; Scale: 1,000 pages).

Steps to run extraction:

1

Select document and configure routing

Click on the target document in the Library sidebar to open it in the document viewer. Check [AI Settings] in the top header to confirm your active model routing (Standard, Premium, or BYOK).

2

Execute extraction

Click the [Extract] button in the Data Studio toolbar or press Ctrl+Enter. The AI reads the page and populates the structured table below.

3

Run batch jobs for multiple documents

To extract several files at once, select them in the Library, click [Add to job] in the action toolbar, assign a job title, and click [Run job]. Results land in the [View Extractions] panel under the job folder.

Cost of extraction

Each successfully processed hosted page costs 1 page-equivalent (PE). Free and Starter BYOK pages use the same allowance. Failed extractions and Pro or Scale BYOK runs cost 0 PE.

Document size cap

Each plan allows a maximum number of pages per document (see Plans & limits). Documents over the cap must be split first.

Exact error messages & resolutions

Document exceeds plan limit of [X] pages

The document has more pages than allowed for a single extraction. Split the PDF into smaller page ranges or upgrade plan.

Monthly processing limit reached (0 PE remaining)

Your monthly hosted PE allowance is spent. Connect your own provider key (BYOK is unlimited on Pro and Scale) or upgrade your plan.

Extraction timed out (120s limit)

Dense multi-page scan took longer than the server lease timeout. Split the file or switch to a faster vision model in AI Settings.

Troubleshooting this task

Low confidence warning tags: Cells with low confidence flags show amber badges. Zoom in on the PDF viewer to verify values directly, correct them inline, and save the pattern.

Complex multi-table invoices: When an invoice contains multiple tables (e.g. services and reimbursable expenses), use [Add column] in Data Studio to shape the schema to capture both, or extract in separate sections.

Related workflows & guides

Dynamite Docs logoDynamite Docs
The review
Supplier invoice: Infer, correct, rememberExtraction works in three steps: infer (the AI reads the document and infers which columns matter), correct (you review and edit anything wrong in Data Studio), and remember (every correction becomes a reusable pattern).. Select one or more files and press Ctrl+Enter, or use the Extract button. Results appear as an editable table. For whole batches, select files and use Add to job to build a named Extraction job (Hobby and above), then press Run job. Hobby permits 25 files per job, 10 active jobs, and unlimited runs. Each document is extracted on its own into a View Extractions folder named after the job.. Each page successfully processed with a hosted model costs 1 page-equivalent (PE). Failed extractions cost 0 PE.. Illustrative example, not extracted customer data.SOURCE DOCUMENTSupplier invoiceInfer the fieldsExtraction works in thre…Correct the valuesSelect one or more files…Remember the patternEach page successfully…Review before exportInfer the fieldsExtraction works inthree steps: infer…Compare with the source on the leftAccept after reviewSaved corrections guide similarlayouts.Original stays…
Infer the fields
Extraction works in three steps: infer (the AI reads the document and infers which columns matter), correct (you review and edit anything wrong in Data Studio), and remember (every correction becomes a reusable pattern).
Correct the values
Select one or more files and press Ctrl+Enter, or use the Extract button. Results appear as an editable table. For whole batches, select files and use Add to job to build a named Extraction job (Hobby and above), then press Run job. Hobby permits 25 files per job, 10 active jobs, and unlimited runs. Each document is extracted on its own into a View Extractions folder named after the job.
Remember the pattern
Each page successfully processed with a hosted model costs 1 page-equivalent (PE). Failed extractions cost 0 PE.
Saved corrections guide similar layouts. Illustrative example.
Extracting data document extraction guide: reviewable document extractionReview the source document and extracted rows in one workspace.
Extracting data document extraction guide: table and line item extractionCheck quantities, descriptions, and prices row by row.
Extracting data document extraction guide: human-in-the-loop validationResolve flagged values before approving the extraction.
Extracting data document extraction guide: Excel, CSV and JSON exportDownload reviewed data as CSV, Excel, or JSON.
Previous chapterUploading documentsAll chaptersFull documentationNext chapterData Studio

Can we save optional cookies?

Essential cookies keep you signed in. With your permission, optional analytics cookies and masked session replays help us understand usage and diagnose errors. Privacy policy. Change your choice anytime in Cookie settings.