Understanding document processing

This article explains what happens to a document once it reaches FlowEngine: from the security check through recognition to review by your team and further processing.

The process at a glance#

1. Intake

   The document is uploaded or scanned with the mobile app.

2. Security check

   Size and malware are checked before processing starts.

3. Text recognition

   Text and tables are read from the file, scanned pages via OCR.

4. Recognition

   Document type, header data and line items are recognized and normalized.

5. Plausibility check

   Missing or contradictory information is flagged for review.

6. Review and approval

   Your team reviews, corrects and approves. Confident results can be approved automatically.

7. Further processing

   The document is assigned to a procurement process, exported or transferred.

Document intake#

Documents reach FlowEngine through two channels that are available in every installation:

• Upload in the browser: one or more files on the Upload Documents page.
• Scanning with the mobile app: paper documents are photographed and uploaded directly.

Further intake channels, such as email intake, depend on how your installation is set up. Every document belongs to your company and is not visible to other companies. Who within your company may see and edit a document depends on responsibility and role.

Security check#

Before a document is processed, FlowEngine checks:

• File size: your company has an upload limit between 1 and 100 MB, 20 MB by default.
• Malware: every file is scanned for viruses. If malware is found, FlowEngine rejects the file; it is not processed.

If the virus scan cannot be performed temporarily, the file is rejected by default as well. Your administrators can decide in the security settings that files are accepted in this case anyway.

Text recognition#

Depending on the file format, FlowEngine reads the content differently:

Word files can be uploaded and viewed but are not recognized automatically at the moment. Save Word documents as PDF before uploading them.

Recognizing document type and data#

From the text it has read, FlowEngine determines with AI support:

• Document type: request, offer, purchase order, order confirmation, delivery note, invoice or credit note. If the type cannot be determined, it stays unknown and is flagged for review.
• Header data: for example document number, date, buyer and supplier, references to other documents, payment and delivery terms, and totals.
• Line items: articles with description, quantity, unit and prices.

The values are normalized, for example date and number formats. Where possible, FlowEngine remembers the place in the document where each value was found, so you can later check directly where a value comes from.

Your company's administrators decide which AI model performs the recognition.

Plausibility check#

After recognition, FlowEngine checks the result. Review hints are raised, for example, when

• the document type remains unknown,
• buyer or supplier is missing or their addresses are incomplete,
• buyer and supplier point to the same company,
• no line items were recognized on a document that should have them,
• payment or delivery terms are missing on an offer,
• the document totals do not match the recognized line items,
• the overall recognition confidence is too low.

Review hints do not change any values. They show where your team should take a closer look.

Structured data#

The result is a structured record made up of document type, fields and line items. It is shown next to the preview of the original document and is the basis for all further steps: assignment to procurement processes, offer comparison, export and transfer to target systems. The original document is kept unchanged.

Review and approval#

Documents with review hints always go to the Work Queue. There, and in the detail view of a document, your team checks the recognized values, corrects them where needed and approves the document.

Your administrators can configure that documents without review hints are approved automatically when recognition reaches a defined minimum confidence. Your team reviews all other documents manually.

An approved document is read-only. To change it again, its approval has to be removed first.

Further processing#

After processing, FlowEngine assigns the document to a procurement process through its references. You can export the data as JSON, CSV or Excel, provided exports are allowed in your company, or – if set up for your company – transfer it to a target system through an integration.

Limits of automatic processing#

• Recognition can be wrong, especially with poorly legible scans or handwriting. Always check important values before approval.
• FlowEngine does not change values on its own after approval.
• Deviations between documents are shown but not corrected automatically.

How to upload and review a document step by step is described in Upload and review a document (sign-in required).