From Scan to Searchable: OCR and Intelligent Capture on IBM i

Back to Blog

Every document problem starts at capture. A document that arrives with no index data can't be found, routed, or retained properly — so someone has to add that data by hand, and everything downstream waits on them.

For years the answer was a person with a keyboard: read the invoice number off the page, type it in, read the date, type that in. It works, right up until the stack gets tall. The good news is that there are now several layers of automation that can do that reading for you, and the best results come from knowing which one fits which document.

The Hidden Cost of Manual Indexing

Manual indexing looks cheap because the cost is spread out. Nobody gets a bill for it. Instead it shows up as a backlog that never quite clears, as documents that can't be found because a digit was mistyped, and as skilled people spending their day copying values from one place to another.

It also creates a delay. A document that's been scanned but not indexed is effectively invisible until someone gets to it, which is exactly when your team needs it most — at month-end, at audit time, or when a customer is on the phone.

The Layers of Automated Capture

"OCR" gets used as a catch-all, but automated capture is really a toolbox. Each tool is suited to a different kind of document:

• Barcode recognition. If you control the form, a barcode is the most reliable way to carry data. The value is read at scan time and flows straight into the index fields, with no interpretation required.
• Full-page OCR. This converts the scanned image into searchable text, so a document can be found by any word on the page, even when you don't know its index values.
• Zone recognition. For fixed-layout forms like invoices, checks, and HR forms, you define the area where each field lives, and the system reads whatever is in that area. It works well on clean, consistent forms, but it has no way to judge whether what it read is right, so results deserve a spot-check.
• Form processing. Structured forms such as applications, survey sheets, and order forms are recognized and their data extracted based on template matching.
• Pattern matching. Rules that look for the shape of a value, such as an invoice number, an account code, or a date, anywhere in the document's text.
• Batch recognition. Recognition that runs in the background, so large volumes can be processed overnight and your team starts the day with indexed, searchable files.

Barcode or OCR? What We've Learned in Real Implementations

A customer moving to RVI from another capture product recently asked a set of questions we hear often: how do we split a stack of scanned pages into separate documents, how do we get index values onto them, and what happens when something goes wrong? Their instinct was to use OCR for all of it. Our answer was more nuanced, and the distinction is worth passing along.

OCR comes in two flavors. Full-text OCR reads the whole page and stores the recognized text so you can search it later by keyword. It's character recognition, not word recognition, so search results reflect exactly what the engine read. Zoned OCR is more targeted: you define a box on the page, and whatever text sits in that box becomes an index value.

For separator pages and index values, barcodes are more dependable. A barcode is an unambiguous code: it reads correctly or it doesn't. Zoned OCR reads printed characters, which makes it more sensitive to real-world scanning. In our own testing, a page that feeds through the scanner slightly skewed can throw off the read, and if any other text lands inside the zone, the engine can't tell it apart from the value you wanted. It doesn't flag a questionable read. It captures what it sees and moves on.

Neither method will tell you a separator was missed. If a separator isn't recognized, whether that's an unreadable barcode or an OCR value that isn't picked up, the page is treated as a regular page and grouped into the document already in progress. Nothing stops and nothing alerts. So the real question is where you want to catch the problem:

• With barcodes, manual scanning mode lets the operator control page and batch breaks live as they scan, so an unread barcode is corrected on the spot instead of quietly merging two documents.
• With OCR, you catch it after the fact, by reviewing each page's results in the review window before accepting and uploading.

Where AI Changes the Picture

Template and zone methods shine when documents look the same every time. They get harder to maintain when layouts vary, because every new layout needs its own setup.

That's the gap content-based AI is meant to close. Instead of depending on where something sits on the page, AI-assisted capture looks at what the document actually contains. That makes a few things possible: classifying a document by type as it arrives, suggesting index values based on its content, and routing it to the right place without a person sorting the pile first. Because the system learns from usage, those suggestions can improve the more your team works with it.

The practical effect is that the work shifts from typing values to confirming them, and for well-understood documents, not even that.

Capture Quality Still Matters

Recognition is only as good as the image it reads. Skewed pages, shadows, and faint text all make reading harder for any engine. That's why the quality of the scanning step matters as much as the recognition step. RVI works with Tungsten VRS and the AIPE engine to enhance images and read barcodes right at the scanner, so recognition starts from a clean image rather than trying to rescue a poor one.

Start With the Documents That Hurt Most

You don't need to automate everything at once. A sensible way to begin:

1. Pick one high-volume document type. AP invoices, purchase orders, and HR onboarding packets are common starting points.
2. Decide which index fields you actually need. Fewer fields means faster, more accurate capture.
3. Choose the most structured method available. Can you add a barcode? Is the layout fixed enough for zones?
4. Run a batch and review the exceptions. The first pass tells you where the rules need tuning.
5. Measure the time saved, then move to the next document type.

If you're still earlier in the journey, our post on how IBM i shops are going paperless covers the broader steps, and spool file capture handles the documents your IBM i already generates without any scanning at all.

Take the Manual Scanning Step Out Entirely

If your team's biggest pain is sitting at the viewer accepting and uploading each batch, RVI's Import Service removes that step. Despite the name, it works two ways. With File Import, your team scans into a designated network folder just as they do today, and the service picks the pages up, reads the barcodes, and applies the indexes. With Email Import, it monitors an inbox and does the same with attachments, such as vendor invoices. Both run continuously in the background with real-time monitoring, and both read Aztec, Codabar, and Code 128 barcodes. The Import Service manual page has the details.

Where RVI Fits

RVI brings these layers together in one platform. Advanced recognition covers barcode reading, full-page OCR, zone recognition, form processing, pattern matching, and batch recognition. RVI's AI features add automatic document classification, predictive indexing, and automatic import and indexing on top. Everything feeds the same searchable repository, and it runs alongside your existing IBM i environment.

Back to Blog