Data type
License your business documents for AI training
Business documents your company wrote and owns, such as manuals, reports, templates and procedures, can often be licensed for AI training. Outrich Data sorts what you own from what you only hold, redacts names, signatures and client details, and pays you for a licensed copy. Privileged files and client-owned files stay out.
What counts as documents
Documents are the files your company has produced in the course of doing its work: operating procedures, technical manuals, internal reports, proposals, contract templates, training material, policies, forms and research. They live as PDFs, Word files, spreadsheets, slide decks and scanned paper, often spread across several storage systems and many years.
The line that matters is authorship. A procedure your team wrote is yours. A report a client sent you, a file you store for a customer, or a signed agreement that contains the other party's confidential terms is a different matter. Documents you hold on behalf of clients are excluded, as are records about minors.
Why AI labs want it
Much of the expert writing in the economy is never published. It sits in document management systems as inspection reports, underwriting memos, engineering specifications and standard operating procedures. AI labs want this material because it shows how professionals in a field write, reason and structure their work, in a depth public web pages do not match.
Documents with complex layouts are useful for a second reason. Tables, forms, stamps, handwriting and multi-column scans are hard for models to read. Labs use real business documents, especially those paired with the structured data extracted from them, to train and evaluate document understanding.
Removed before licensing
What comes out first.
- Names, signatures, initials and contact details of individuals
- Client and counterparty names, logos and letterhead
- Bank details, tax IDs and Social Security numbers
- Health information that could identify a patient, handled under a HIPAA de-identification method where HIPAA applies
- Attorney-client privileged material and work product, removed in full
- File metadata such as author, tracked changes, comments and revision history
- Deal terms that would identify a specific transaction or counterparty
Data title search
What gets checked.
- Authorship and ownership. Confirm the company wrote the document or holds the copyright. Documents written by consultants or agencies depend on the terms of that engagement.
- Confidentiality clauses. Signed agreements usually contain confidentiality terms that cover the agreement itself. Executed contracts often need the counterparty's consent or exclusion.
- Client files. Documents received from or prepared for a client are often the client's property or confidential information under your engagement terms.
- Privilege. Legal advice and litigation files should be screened out. Disclosing them can waive privilege, so review the screening approach with counsel.
- Third-party content. Reports, standards and licensed publications saved in your systems belong to their publishers.
- Privacy law. HR files, medical records and customer correspondence carry personal data subject to HIPAA, GDPR or CCPA, depending on the content and the people in it.
Who signsThe general counsel is usually the key approver, since ownership and confidentiality drive what qualifies. An owner or COO sponsors the deal and IT or records management handles the export.
Value
What moves the price of documents.
| Factor | Effect on value |
|---|---|
| Specialist content | Technical, clinical, legal, financial or engineering documents written by practitioners are generally worth more than general office files. |
| Paired structured data | Documents linked to the fields extracted from them, such as a form and its database record, are useful for document understanding and attract more interest. |
| Consistency over time | A long run of the same document type, such as years of inspection reports, is more useful than a mixed pile. |
| File quality | Native digital files and clean scans are easier to process than low-resolution faxes, though difficult scans have their own use. |
| Clear authorship | The share of documents your company clearly authored and owns sets the usable size of the dataset. |
Industries where this is common
Questions
Licensing documents: what companies ask.
Can we license signed contracts?
We have decades of scanned paper. Is that usable?
How do you separate our documents from client documents in a shared drive?
What about privileged legal documents?
Find out if your data qualifies.
Eleven questions, about five minutes, no call required. You get an answer on screen and a written follow-up.