OCR Data Room Documents Before Due Diligence Review

To OCR data room documents before a diligence review, download the room with its folder structure and index, test which files are image-only scans, convert those in priority order into a parallel folder tree, and keep a manifest that ties every converted file to its data room index number. Log anything that would not convert.

The aim is modest and practical: when the review team sits down, a search for "change of control" should reach every agreement in the room, and where it cannot, the team should know exactly which files it missed. The sections below take the work in the order you meet it.

Why are so many data room documents image-only scans?

Data room documents end up as image-only scans when the seller uploads its records as they sit in its files, and those records are paper that was scanned, or PDFs that were printed, signed in ink and scanned back in. Test your own room before assuming how much of it is searchable.

Several routes lead to the same result:

• Executed copies. The Word draft had text. The copy that went into the contract file is the one with signatures, and that copy may have come off a copier.
• Old paper. Leases, title documents, permits and board minutes from before a company kept electronic files exist only as whatever scan was later made of them.
• Documents from third parties. A document that reached the company on paper or by fax is kept as it was received.
• Flattening. A document printed to an image PDF before upload, for instance to make a redaction permanent, is legible and unsearchable.
• Compilations. A single PDF named "Customer contracts 2019" may hold thirty agreements scanned in one pass.

A data room may have a search box of its own. Whether it reaches scanned pages depends on the platform and how the room was set up, so test it instead of assuming either way. Open a scanned document, pick an unusual word you can see on the page, and search the room for it. If the document does not come back, the room's search is covering only files that already had text, and every negative search result so far is unreliable.

Expect a mixture. Files exported straight from software, such as spreadsheets and reports, carry text already. Signed contracts, property documents, corporate records and anything old are the places to test first.

Start with the folders a buyer will read in due diligence: customer contracts, financing, and any merger agreement. A capitalization table and a list of each subsidiary belong in that first pass too. Convert those scans before the reports that already have text, and write down which files stayed as images.

Which data room folders should be converted first?

Convert first the folders where a missed clause changes the deal: material customer and supplier contracts, financing documents, real property, intellectual property assignments and employment agreements. Folders of natively generated reports can wait, or be skipped if they are searchable already.

Start from the questions the review has to answer, then triage by sampling.

A ten-minute triage

1. Open the room's index and list the top-level folders.
2. In each folder, open five documents spread through the list. For each, try to select a word.
3. Note the share that are scans, and whether the scans are clean or poor.
4. Note the rough page count for the folder from the index, if the index gives one.
5. Rank the folders by two things: how much the review depends on them, and how much of the folder is unsearchable.

A ranking might look like this. The folder names and conditions are illustrative, and your own sampling decides.

Employment folders have their own difficulties, because a personnel file mixes offer letters, standard-form agreements and signed acknowledgments in one scan. The post on OCR for employment contracts deals with them.

Priority also depends on the timetable. If the first diligence call is in three days, convert the first-priority folders completely before starting the next group. A room that is half-converted in every folder helps nobody, because no search of any folder can be trusted.

Downloading and converting a data room without breaking its index

Download the room in one pass with its folder structure, export the index alongside it, and treat that download as a dated snapshot you never edit. Convert into a second folder tree with the same shape. The originals keep their names and places, so the index still describes them.

Before any of this, read the terms you accepted to enter the room and the non-disclosure agreement behind them. They may limit downloading, copying or passing documents to an outside service. If downloads are disabled or the terms are unclear, ask the seller or deal counsel, and do not work around the restriction. The post on uploading client documents to an OCR tool lists what to ask a conversion vendor.

Then:

1. Take the snapshot. Use the room's bulk download if it has one, so the folders arrive as the seller arranged them. Record the date and time.
2. Export the index. If the room can produce its index as a spreadsheet or a PDF, save it with the same date. This is the list your manifest will be built from.
3. Lock the originals. Set the snapshot folder to read-only. Nobody renames, merges, splits or annotates anything inside it.
4. Reconcile. Count the files you received against the index. Files that are listed and missing are your first questions for the seller.
5. Sort scans from text. Test each file, or each file in the priority folders, and mark the result in the manifest. Only the scans need converting.
6. Convert into a mirror. Create a second tree named for the snapshot date with the same folder names. Each output goes in the folder that matches its source.
7. Choose outputs by use. A searchable PDF lets a reviewer read the signed page and search it in one file. Plain text or Word suits searching across many files and quoting.
8. Check a sample of each folder against the page images before telling anyone the folder is ready.

Size limits shape the batch. Schema Law takes PDF, PNG or JPEG files of up to 20 MB, and the pricing page gives each plan's monthly pages and its page limit per document. Add up the page counts of the folders you ranked first before choosing, and note any file above the upload limit, since it will have to be split and the parts recorded.

Compilations need a decision. A 300-page PDF holding thirty contracts can be converted whole, which keeps it matched to its one index entry, or split into thirty files, which makes each contract separately searchable. If you split, the splitting happens in the mirror, never in the snapshot, and each part records the page range it came from.

Can you ask the seller for searchable data room files?

Yes, you can ask, and it is worth asking early. Whether the seller agrees is a matter of negotiation and of what the seller has. A request made in the first days, through the room's question function or through deal counsel, can save conversion work if it is granted.

Be specific about what would help. A general complaint that the room is unsearchable gets a general answer.

How a seller answers depends on the deal. Ask for what you need, keep the list short, and record the reply.

A request might read:

Folders 3 (Material Contracts) and 5 (Real Property) consist largely of image-only scans that cannot be searched. To allow an efficient review, please (a) provide text-searchable versions of the documents in those folders if available, (b) provide a legible copy of documents 3.2.14 and 5.1.7, which are unreadable in part, and (c) provide the data room index in spreadsheet form.

Two points about whatever comes back. Signed copies are still the documents under review, so a native Word file is a reading aid and cannot replace the executed scan. And a seller's searchable PDFs were made by someone's OCR with someone's settings. Test them the way you would test your own: search for a word you can see on a poor page, and compare a few amounts with the image.

If the seller declines, you have lost nothing, and the request is on the record as the reason the review needed extra time.

Keeping each converted file tied to its data room index number

Tie each converted file to its index number with a manifest: one row per data room document, holding the index number, the path and file name as downloaded, and the name of every output made from it. The manifest is what lets a finding in the diligence report cite a document the seller can locate.

The post on extraction across a due diligence data room covers recording contract terms once the files are readable. The manifest comes before that and tracks files, not terms.

Three situations test the manifest.

Nested numbers. In a hierarchical index, 3.2.14 is the fourteenth document in the second subfolder of folder 3. Sorted as text, 3.2.14 comes before 3.2.2. Add helper columns for each level if you need the list in room order, and leave the original number untouched in its own column.

The room is renumbered. A room can change during a deal. If documents are added and the index is renumbered, last week's 3.2.14 may become 3.2.15. This is why the snapshot date sits in every row. When you take a fresh download, export the new index and match old rows to new ones by file name and page count. Record both numbers. A report that cites "3.2.14 (index of 9 October)" stays accurate after the room moves on.

Duplicates. The same supply agreement may appear in the contracts folder, again as an exhibit to a financing document and a third time under a customer's name. Convert it once if the files are identical, and point the other rows at the first. If the copies differ, for example one has the signature page and one does not, that is a finding, and both stay in.

For later additions, compare the new index with the old one and convert only the rows that are new or changed. Add them to the same manifest with their own snapshot date.

Logging data room documents that would not convert

Keep an exceptions log of every document that was not converted or converted badly, with the reason and what was done about it. Without it, a search that finds nothing in a file you never managed to read gets reported as a clean result.

Reasons, and what each calls for:

Each entry should carry the index number, the reason, the date, who will review the document by eye, and the outcome. Do not predict from this table how any tool will cope with handwriting or poor scans. Run the file, look at the output next to the page, and log what you see.

The log earns its place when searching begins. Suppose the team searches 431 contract documents for an anti-assignment clause, and 19 of them sit in the exceptions log. The honest statement is that the search covered 412 documents and the other 19 were read by a person or remain unreviewed. The post on how to search scanned contracts for a clause explains how to run those searches and how to report a clause as absent.

Close the log out before the report is issued. Every entry should end in one of three states: converted after a better copy arrived, reviewed by eye by a named person, or not reviewed, with the reason stated in the report's scope section.