Adding AI OCR to Your Due Diligence Workflow
Scanned PDFs used to be a black hole in due diligence. With OCR baked into every upload, that's no longer true. Here's how we use it.
Due diligence used to involve a lot of "printing, scanning, physically reviewing, and hoping you found the important clauses." With AI OCR baked into every upload, that workflow has collapsed from days to minutes. Here's how we use it.
The old workflow
A typical due diligence exercise for a mid-sized transaction involves reviewing 500-2000 documents. Of those, 30-40% are scanned PDFs — contracts signed years ago, physical filings, correspondence. Scanned PDFs used to be a black hole in the review process. You couldn't search them, couldn't copy text from them, and had to read them manually.
The result: either you spent extra time reading the scans, or you focused on the text-searchable documents and hoped nothing important was buried in the unsearched ones. Neither is great.
How AI OCR changes the workflow
In ShareAndGo, every uploaded document is processed through an OCR pipeline (using Gemini 2.0 Flash as the primary model, with fallbacks). The extracted text is indexed alongside any native text in the document. Search queries run across everything — scans and native text — uniformly.
For modern Gemini-based OCR on Australian English business documents, accuracy is well above 98% on most documents. The remaining errors are typically in the margins (page numbers, headers, footers) rather than the substantive content.
The practical benefits
1. Full-text search across everything. Type a search term, get hits across every document regardless of whether it started as a scan or a Word doc.
2. Natural-language Q&A. "Find all contracts that include a termination-for-convenience clause." The AI reads the extracted text and returns the matching documents, even if the exact phrase doesn't appear.
3. Automatic categorisation. Based on content, documents get auto-tagged as contracts, tax returns, financial statements, correspondence, etc. This happens on upload, so the data room is organised without manual intervention.
4. Redaction detection. We can flag documents where content has been visually obscured (black boxes, whitening) — useful when the seller has produced "redacted" versions but you need to know what's been hidden.
The honest limits
OCR is not magic. Handwritten documents still trip it up. Poor-quality scans (low DPI, skewed pages, faded ink) produce lower-accuracy text. Tables with complex layouts sometimes get the column associations wrong. Australian-specific formatting (certain tax form layouts, for example) can be tricky.
For high-stakes review, we still recommend having a human spot-check the OCR output on the most critical documents. The AI is a force multiplier, not a replacement.
The time savings
Typical feedback from firms that have switched: 2x to 5x faster on the document review phase of due diligence. For a 200-hour engagement, that's 80-150 hours saved — and those hours used to be the most tedious parts of the job.