Document Processing and Extraction

Document processing turns a stored file into usable workspace context. Eigenn classifies supported files, produces searchable text and summaries, adds tags, extracts structured financial fields when possible, and proposes transaction matches for eligible documents.

Processing is intentionally reviewable. The file status, extraction status, confidence, field evidence, corrections, and matching decisions remain visible instead of being hidden behind a single “AI complete” state.

Prerequisites

Before relying on processed data:

• upload the file to the correct workspace under Vault
• keep the source document available for comparison
• use a role with write access to edit extracted fields or resolve matches
• connect transaction data for transaction suggestions
• allow background processing time after the file upload finishes

Find Processing Details

1. Select Vault in the primary sidebar.
2. Open a file, or open its action menu and select View details.
3. Review File Information → Status and Progress.
4. Review Summary, Extracted Data, Suggested matches, Tags, and Related Files when those sections appear.
5. For a cross-document queue, select Vault → Workspace and review Review queue, Suggested matches, Status center, and Audit trail.

Processing Lifecycle

The file-processing state and the financial-extraction state are related, but they answer different questions.

File-processing status

While a job is running, Progress can show a percentage and current stage. If detailed progress is not available, the interface still shows a pulsing Processing indicator.

Extraction status

Use the Extraction status Vault filter to distinguish:

• Pending
• Extracted
• Verified
• Edited
• Failed
• Skipped
• Needs review

A file can be Ready at the file level while structured extraction is Skipped or Needs review. Ready means the file is usable; it does not guarantee that every financial field was extracted.

What General Processing Adds

For supported files, general processing can add:

• a generated title
• a summary
• searchable text
• language and document-type context
• automatically generated tags
• semantic-search indexing

Supported processing formats include:

• PDF
• common image formats
• CSV, plain text, and Markdown
• Microsoft Word, Excel, and PowerPoint formats
• OpenDocument text, spreadsheet, and presentation formats
• RTF

HEIC images are converted for image processing. Other image formats are routed through image classification. Formats outside the supported processing set can remain stored in the Vault without generated context.

Structured Financial Extraction

For recognized financial documents, Extracted Data can include:

• document type
• counterparty
• amount and currency
• tax amount
• issue date
• due date
• statement or service period
• invoice or receipt line items
• statement transactions and balances

Not every field appears on every document. Empty or irrelevant fields stay out of the view.

The document-type badge identifies the detected category. Before human verification, the panel can show an overall confidence percentage. Individual fields can also include evidence such as confidence, validation state, validation issues, source path, and extraction source.

Review warnings

Needs review can be triggered by conditions such as:

• low text signal
• no readable text sample
• OCR being required
• an unsupported extraction format
• a MIME type that does not match the file extension
• validation issues in extracted totals or fields

These warnings do not automatically mean the stored file is unusable. They mean a person should compare the source and add or correct details.

Verify or correct fields

1. Open Extracted Data.
2. Select Verify / Edit.
3. Review Counterparty, Amount, Currency, Tax amount, Issue date, and Due date.
4. Correct the values that do not match the source.
5. Select Save.

Saving a correction marks the extraction as edited and verified by the current user, records the changed fields, and refreshes the document’s search representation.

The panel can show:

• Activity for extraction and correction events
• Versions for recent document versions
• field-level before-and-after details for recorded corrections

Review Line Items and Statement Rows

When the extraction contains invoice or receipt line items, review:

• Description
• Qty
• Unit price
• Amount

When it contains statement transactions, review:

• Date
• Description
• Amount
• Balance

Compare totals with the source document before using the data in a decision. A high confidence value is supporting evidence, not a substitute for source verification.

Resolve Transaction Suggestions

Eligible invoices, receipts, and expenses can receive suggested transaction matches when an amount and issue date are available.

1. Open Suggested matches.
2. Compare the transaction name, date, amount, currency, and confidence.
3. Select Link for the correct suggestion.
4. Select Dismiss for an incorrect suggestion.

For queue-based review, open Vault → Workspace → Suggested matches. You can resolve one suggestion at a time or use Link visible and Dismiss visible after reviewing every item in the visible set.

Work the Review Queue

When extracted data needs attention, Vault → Workspace → Review queue lists the visible documents and their status, document type, and confidence.

• Select Repair failed to retry the visible failed or needs-review extractions that are eligible for repair.
• Select Verify visible to mark the visible queue entries verified.
• Use Audit trail to confirm that changes were recorded.
• Use Export CSV to download extracted rows in Extracted, Verified, Edited, or Needs review states.

Use bulk verification only after inspecting the visible set. It is an operating shortcut, not an automatic validation step.

Verify Success

Processing is complete enough for use when:

• File Information → Status shows Ready
• the preview opens and the file name, size, and type are correct
• the summary and tags describe the source accurately, when present
• Extracted Data agrees with the source or has been corrected and shows Verified
• a resolved suggestion shows Linked or disappears from the pending match queue
• Activity or Audit trail records the human correction or bulk action

For a non-financial or unsupported file, success can simply mean the file is Ready, previewable, searchable by its stored metadata, and available to download. Structured fields are not required for every file.

Limits and Edge Cases

• Upload completion and processing completion are separate. The success toast confirms storage before every background result is guaranteed.
• Vault accepts up to 25 files per batch and 100 MB per file; an empty file is rejected.
• An unsupported processing format can be stored without a summary, extracted fields, or generated tags.
• Structured extraction is intended for recognized financial documents with usable text. Non-financial documents can be marked Skipped and show no Extracted Data panel.
• Image classification and structured financial extraction are separate capabilities. An image can receive a title or summary without structured fields.
• Scanned or low-text files can require review even when the preview looks correct.
• Suggested matches require enough extracted context and available transaction data.
• Verify visible affects every document currently presented in the review section.
• Link visible and Dismiss visible affect every suggestion currently presented in the match section.
• Correcting extracted fields does not modify the pixels or text in the original uploaded file.

Troubleshooting

Status remains Pending or Processing

Refresh the file details and check whether Progress changes. Confirm that the upload itself completed and that the file is not empty. If the status remains unchanged, copy the Storage Path and include the file name and upload time when contacting support.

Status is Failed

Open Vault → Workspace → Review queue. Use Repair failed when the document appears there, then inspect the individual document again. If repair does not resolve it, convert the source to PDF or a supported text-bearing format and upload a new copy.

File is Ready but Extracted Data is missing

The file may be non-financial, unsupported for structured extraction, image-only, or intentionally skipped. Check Summary, Tags, File Information → Document Type, and the Extraction status filters. Use the original file as the source of truth when no structured panel appears.

Extracted Data says Needs review

Read the review reasons, compare each displayed field with the source, select Verify / Edit, and save corrections. For a scanned PDF with no readable text, a cleaner text-based PDF may produce a better result.

A field has low confidence

Read the evidence detail beneath the field. Check validation issues, the cited source path, line-item totals, and currency. Correct the field rather than relying on the overall confidence badge.

No transaction suggestions appear

Confirm that the document is an invoice, receipt, or expense with an amount and issue date, and that the workspace has relevant transaction data. You can still reconcile receipt work manually from Workspace Inbox and Approvals.

Related Pages

• Document Vault
• Organize and Share Documents
• Workspace Inbox and Approvals
• Transactions
• Support