The Old Way: OCR Templates + Scripts
For decades, automating document-to-terminal workflows meant two separate tools: an OCR engine to extract text from scanned documents, and a terminal script to enter that text into IBM i screens.
The OCR engine requires templates—manually defined zones on each document type that tell the software where to look for specific fields. The terminal script requires hardcoded screen coordinates—exact row and column positions for every field on every screen in the workflow.
Both are brittle. A vendor changes their invoice layout? Rebuild the OCR template. An IT update adds a field to a screen? Rewrite the terminal script. Multiply this across dozens of document types and screen workflows, and you have a full-time maintenance burden.
The AI Vision Approach
AI vision models process documents and screens holistically. Instead of looking at predefined zones, they read the entire document—understanding layout, labels, values, tables, and even handwritten notes in context.
How It Works
Document understanding
The AI model sees the full document image and extracts structured data—field names paired with their values, regardless of where they appear on the page.
Screen understanding
The same AI reads the IBM i terminal screen, identifying field labels, input positions, and the current context (which program, which menu).
Intelligent mapping
The AI matches document values to screen fields by semantic meaning—“Invoice Total” on the document maps to “Total Amount” on the screen, even if the labels don’t match exactly.
Adaptive entry
Data is entered via TN5250, with the AI monitoring the screen response after each action and adjusting as needed.
Why This Matters for IBM i Organizations
No Template Maintenance
When a vendor changes their invoice format, AI vision still reads it correctly. It understands that “Amount Due” and “Total Payable” mean the same thing. No template rebuild, no downtime.
Handles Any Document
Typed forms, handwritten notes, multi-page reports, tables with variable columns—AI vision processes them all. Traditional OCR chokes on anything outside its template definitions.
Self-Improving
When a human operator corrects an entry, the system learns from that correction. Over time, accuracy improves without manual retraining or template updates.
End-to-End Automation
Instead of two separate tools (OCR + terminal script), AI vision handles the entire pipeline: document ingestion, field extraction, screen navigation, data entry, and error handling—in a single, integrated system.
The Bottom Line
AI vision isn’t an incremental improvement over OCR + scripting—it’s a fundamentally different approach. It eliminates template maintenance, adapts to document and screen variations, and handles the full workflow from document to terminal entry. For organizations running IBM i with high-volume data entry workflows, it represents a shift toward automation that requires minimal ongoing attention.
