Key Takeaways
- Free for Current Subscribers: PaperStream Capture 6.0 is available at no additional charge on an existing subscription.
- Background Processing: OCR extraction now runs in the background, so operators no longer wait for processing before moving to the next batch.
- Mailroom Separation: Ultrasonic sensors detect envelopes in the stack, so a full feeder of mixed mail separates automatically.
- Zero-Shot Extraction: PaperStream AI reads unclassified documents with no templates and no training beforehand.
- Fixed Forms Stay Free: Fixed form processing carries no page volume count and no click charges. You pay for the cloud only when a job needs it.
- AI Needs Good Input: About 75% of attendees in our live poll already use generative AI, and extraction accuracy depends on the image quality it receives.
Jump to a section:
The short answer
PaperStream Capture 6.0 is the sixth full release of PFU's capture software, adding background OCR processing, envelope-based mailroom separation, and configurable QA filtering. PaperStream AI is a separate paid cloud add-on that extends Capture Pro and Pro Premium into intelligent document processing, handling semi-structured and unclassified documents without templates.
The details below come from a live webinar hosted by the PFU team, including working demonstrations of each feature. This article covers what was discussed. The full webinar is available at the end.
Why capture quality determines AI accuracy
In a live poll during this webinar, attendees were asked whether their company uses a generative AI solution such as Microsoft Copilot or Google Gemini. About 75% said yes. Another 10% said they are planning or reviewing one. That shifts what document capture is for.
Two points tend to hold across expert opinion on AI. The first is that accuracy requires a large volume of information. The second is that the information has to be correct, because inaccurate data creates problems downstream. In agentic workflows, where agents act without close human supervision, data quality matters more than it did when a person reviewed every step.
That makes image quality a precondition rather than a detail. Poor input degrades recognition, and recognition errors propagate into whatever system consumes the data.
The manual scanning process this replaces
Without capture software, a knowledge worker runs roughly six steps for every batch. Walk to the scanner and open the scan application. Load the paper and scan. Review the images and rescan anything that failed. Name the file and choose a location. Upload it to the content management system. Add index values and metadata.
Using a multifunction printer adds more. Navigate the control panel, scan to file, email the file to yourself, walk back to your desk, open the email, and save the attachment before the upload and indexing even start.
Two costs follow. Manual steps vary between operators, which introduces errors that automation does not. And on a production scanner running a back file conversion, every one of those steps is time the scanner sits idle.
There is a security cost as well. Each route creates shadow copies. The desktop path leaves a file on the local machine before upload. The MFP path leaves one on the device hard drive, one in the email system, and one on the desktop. Capture Pro releases directly into the content system instead, so those intermediate copies never exist.
What's new in PaperStream Capture 6.0
Background processing
Previously, scanning and OCR extraction ran as one blocking sequence. In 6.0, extraction moves to a background process that runs on the local station or offloads entirely to the storage server in a multi-station environment. It is multi-threaded, so it uses multiple cores. Operators complete a batch and move on, then return to documents that are already indexed and ready for review.
Mailroom support
With hybrid and remote staff, mail often arrives at a headquarters office where the recipient is not present. In legal and finance that mail is frequently time sensitive.
Capture 6.0 can separate scanned mail automatically by detecting envelopes through the ultrasonic sensors on fi Series scanners. Load envelopes and their contents together into the feeder and the software separates each item. On the fi-8950, with a 750-sheet ADF, that means an entire morning's mail at production speed. Version 6.0 supports separating on an envelope at either the start or the end of each item. Paired with PaperStream AI, the addressee information can be extracted from the envelope itself and passed into the content system to drive distribution.
QA filtering and per-field confidence
Extracted index values are checked against a confidence threshold, which defaults to 80%. Anything below it is flagged for the QA operator. New in 6.0, that threshold is configurable per index field, so a value where errors are expensive, such as an invoice number or a shipping tracking number, can be held to a higher standard than the rest of the form.
QA operators can also filter the view to show only the flagged fields rather than reviewing everything.
Interface and import improvements
When an operator zooms into a document and tabs through index fields, the image now follows along, highlighting the active field in a red rectangle. Capture Pro Premium adds an Outlook plugin that launches the software directly from an email attachment, without Capture Pro needing to be open first.
See the demonstrationsEvery feature here was demonstrated live, including envelope separation running on an fi-8950 and zero-shot extraction on four unseen documents. Link: Request access to the full webinar.
Definition: the three types of document processing
| Document type | What it means | What handles it |
|---|---|---|
| Fixed forms | The same form every time, with different data. Credit applications, healthcare admissions forms, DMV forms. | Capture Pro and Pro Premium, using known field locations |
| Semi-structured | The same information every time, in different places. Invoices from three suppliers all carry an invoice number, but not in the same spot. | PaperStream AI, which locates the known field |
| Unstructured and unclassified | Disparate documents with nothing declared in advance. The system is told nothing and trained on nothing. | PaperStream AI, which identifies the document and decides what to extract |
Traditionally, even semi-structured processing required setting up templates by form ID and fixing the geolocation of each field, followed by training on sample documents. Zero-shot processing removes both steps.
What PaperStream AI does
In the live demonstration, four unrelated documents were imported with no setup. PaperStream AI identified each one, an automobile claim report, an invoice, a dental plan document, and a car accident report, then determined which fields to extract from each and produced both the field name and its value.
Table data comes through as well. On an invoice, that means part number, description, quantity, unit price, and amount, which is what makes automated invoice-to-purchase-order matching possible.
Extraction can be steered with natural language expressions. Rather than configuring a field definition, you write an instruction such as asking it to extract the policy number and, where several numbers appear, to pick the one associated with the insurance company or agreement. Results from a zero-shot batch can also be saved as a reusable scan job profile.
How the editions compare
| Capability | Capture Pro | Pro Premium | Plus PaperStream AI |
|---|---|---|---|
| Fixed forms, barcodes, separation | Yes | Yes | Yes |
| Redaction and after scan correction | Yes | Yes | Yes |
| File import and Outlook attachments | No | Yes | Yes |
| A3 production scanner support | No | Yes | Yes |
| Semi-structured and unstructured documents | No | No | Yes |
| AI learning from corrections | No | No | Yes |
Capabilities that carry across every edition
Any TWAIN scanner is supported. Ricoh document scanners deliver the image quality and paper handling the software is built around, but third-party devices work too. In PFU testing, those devices produced better throughput through Capture Pro than through their own manufacturers' scan applications.
Advanced file separation runs on form ID, patch code, barcode, zonal OCR, blank page, or page count. That lets operators load mixed multi-page documents into the feeder in one stack rather than separating by hand.
Redaction is burned into the image. Many content systems apply redaction as a layer, which can be reversed. Here it is permanent once released, in either a black or white rectangle, and you choose whether redacted values are passed into the content system or the export file. Version 5.2 added free form redaction for areas that are not predefined fields, such as a routing number partway down a page.
After scan correction fixes image problems without rescanning. The operator is shown nine thumbnail variations, picks the closest, and gets nine more from that starting point. A home button restarts the process and a back button steps back one choice. It requires no expertise in image enhancement settings and can be run at a different station entirely.
Two-way ECM integration logs into the content platform during setup, so folders, libraries, and index fields are pulled directly from the user's own account rather than recreated by hand. Demonstrated integrations included DocuWare, Laserfiche, Hyland via XML DIP, SharePoint, and EtherFAX, which turns an fi-8170 into a 70 page per minute fax device. Systems without a dedicated integration can take images plus XML data.
Everything collapses into one button. Scanner settings, PaperStream IP enhancement profiles, separation rules, QC profiles, file naming, and destination all live in the scan job. The operator walks up, loads documents, and clicks the job.
Watch the full webinarScott Francis demonstrates background processing, redaction, after scan correction, and zero-shot AI extraction, followed by audience questions on security, pricing, and accuracy. Fill out the form to watch in full. Access the webinar.