Business documents and forms for AI training
PDFs, scans and office files such as contracts, invoices, purchase orders, statements, applications and reports, ideally with extracted field values.
- Measured in
- Documents and pages, by document type and layout variety.
- Who buys
- Document AI vendors, labs training multimodal models, agent builders automating back-office work.
- Where it comes from
- Shared drives, document management systems, accounts payable archives, legal and insurance files.
- Rights to clear
- Counterparty confidentiality, NDAs, client ownership of deliverables and any regulatory record-keeping rules.
- Privacy
- Names, addresses, account numbers, signatures and IDs redacted or replaced. Financial account numbers removed entirely.
- Pricing signal
- Priced on document count, variety and whether ground-truth labels are included.
- Document AI models learn to read layouts, tables, signatures and fields from real business documents.
- The most useful datasets pair the original file with its extracted fields, which shows a model what the right answer is.
- Documents almost always contain counterparties' information, so confidentiality clauses and redaction come before licensing.
- Scanned and messy documents are often worth more than clean ones, because they are what models struggle with.
What makes document data valuable
Variety beats volume. A model that has seen 5,000 invoice layouts from 3,000 suppliers learns more than one that has seen a million invoices from one template. Buyers also value the hard cases: faded scans, handwriting, stamps, rotated pages, multi-page tables.
Labels multiply the value
If your systems already extracted fields from each document, such as invoice totals entered into an accounting system, that pairing is ground truth a model can learn from. Keep the link between file and fields when you prepare the data.
Rights and confidentiality
Business documents are full of other parties' information. Before licensing:
- Check NDAs and contract confidentiality clauses with counterparties
- Exclude documents that belong to clients, such as deliverables you created for them, unless the contract allows it
- Exclude regulated records with their own rules, such as health or credit records
- Remove or replace financial account numbers, tax IDs and signatures
Formats buyers accept
Original PDFs or images plus a JSON or CSV file of extracted fields per document, with a manifest listing document type, date range and source system. See how to build a data manifest.
Questions
Can we license invoices we received from suppliers?
Possibly, after checking supplier agreements for confidentiality terms and removing account numbers and personal details. Many companies license only the extracted, de-identified version.
Are templates worth anything?
Blank templates have little value. Completed, varied, real documents are what buyers want.
Do buyers need the original files?
For document AI, yes. Layout is part of what the model learns, so text-only extracts lose much of the value.
Related guides
Related data types
Sources
Need business documents and forms, or hold it?
Tell us what you need or what you hold. We check rights and privacy first, then introduce vetted counterparties. You license directly with each other.
- We reply to every enquiry within two working days.
- No upfront fees for sellers
- We never hold your raw data