The Scanned Contract That Nobody Can Search
A regional property management firm has 4,200 vendor contracts stored as scanned PDFs from a 2011-era multifunction copier. Each file is 8-14 MB, image-only, and rotated 90 degrees on roughly a third of the pages. The legal team wants full-text search and the ability to copy clause numbers into an Excel tracker. They have Adobe Acrobat Pro on 12 desktops and a shared network drive; no server-side automation budget this quarter. A paralegal proposes running OCR on all 4,200 files overnight on her laptop. The IT lead notes that the originals are the only copies and must remain byte-for-byte unchanged for audit purposes.
1. Evaluate the paralegal's overnight OCR plan. What will actually happen, and what should change?
Read the worked answer
The plan fails on three counts.
- Scale vs. throughput. 4,200 files at 8-14 MB with image-only content is a multi-day job on a single laptop, not an overnight one. At a realistic few seconds per page plus I/O, expect days of continuous processing.
- Rotation. OCR accuracy collapses on pages rotated 90 degrees. Roughly 1,400 files would produce garbage text unless deskew/rotate detection is enabled before recognition.
- Originality. In-place OCR modifies the file. The audit requirement demands the originals stay untouched.
Correct approach: batch-process with Recognize Text in Acrobat against a copy of the tree, output to a new folder, leaving source files read-only. Enable automatic page rotation/deskew in the OCR settings. Also consider Optimize Scanned PDF to shrink file size before OCR to cut processing time. Spot-check a sample of 20 files, including known-rotated ones, before committing the full run. With 12 desktops, split the tree into 12 batches rather than one laptop.
2. The legal team also wants clause numbers in Excel. Does OCR alone deliver that?
Read the worked answer
No. OCR makes text searchable and selectable; it does not produce structured data. Copy-pasting clause numbers from 4,200 files into Excel is manual and error-prone.
The right tool is Export PDF to a spreadsheet format, but that only works reliably if the source has consistent layout. Given inconsistent scans, a better path is to OCR first, then use Acrobat's export or a templated extraction on a representative subset, and validate. Set expectations: OCR quality on 2011 scans with rotation issues means the Excel output will need human review.
Recommendation: deliver searchable PDFs as phase one (immediate value, low risk), and treat structured extraction as a separate, sampled, validated project.