OCR (Optical Character Recognition)
OCR is technology that converts an image of text — a scanned document, a photo, a PDF page — into machine-readable text, the first step in extracting data from documents that aren't already digital.
OCR reads the pixels of a scanned page or photograph and outputs the text it recognizes. On its own, that output is just text — it doesn't know that the third number in a row is a subtotal versus a tax line, or that a checked box means a form field is true. That structural understanding is a separate, harder problem layered on top of raw OCR.
Accuracy depends heavily on image quality, layout complexity and language — a clean, single-column invoice scans far more reliably than a multi-page statement with merged table cells or handwriting. Production document pipelines benchmark OCR accuracy against real sample documents rather than assuming a headline accuracy number will hold, and flag low-confidence extractions for human review instead of silently accepting a possibly-wrong value.
OCR is the entry point for most document intelligence work — invoice processing, form extraction, digitizing paper records — but the real deliverable is almost always the structured data downstream of it, not the raw recognized text.