Ask an operations lead which documents the company has and you get a list by folder: policies here, the handbook there, the wiki, the manuals, the sales playbook. Buyers read the same list differently. They are looking for text that shows decisions being made in order, by people who do the work, over time.
What do buyers say they want?
The clearest statements come from the buyers’ own pages. micro1 lists SOPs, process documentation and internal knowledge bases among the material it takes, alongside engineering, support, sales and legal records. Scale AI lists documents and files with metadata, and Jira and Confluence work and knowledge records. It says it looks for real operational workflows, domain depth and established data systems.
Neither publishes a per-document price. micro1’s lowest published tier begins at $100K a company. Scale AI shows $10K to $1M+ and is careful to call it illustrative rather than guaranteed. Both describe whole partnerships, so the question for a document set is less “what is it worth” than “does it add depth to what a buyer can license from us.”
How do the main document types compare?
| Type | What buyers can learn from it | Typical weakness |
|---|---|---|
| Operations and service manuals | Long, specific procedures in one domain | May include customer site details |
| SOPs and runbooks | Step-by-step decisions, with exceptions | Often short and scattered |
| Wiki or knowledge base | Linked pages, owners, years of edits | Comments and history full of names |
| Sales and support playbooks | How people handle real situations | Quote real customers and deals |
| Company policies | Rules and approvals | Short, generic, often from templates |
| Employee handbook | How the company describes itself | Benefit summaries belong to providers |
The top of the table is not there by accident. A manual or a well-kept wiki is long, internally consistent and specific to one kind of work. A travel policy is a page and a half that reads like every other travel policy.
Why do policies rank lower?
Because most of them are similar across companies. A security policy, an expense policy and an acceptable-use policy are often adapted from a template, an insurer’s model or a consultant’s pack. That makes them less distinctive, and it can make the rights less clear, since the template text may still belong to its publisher.
Policies earn their place in two cases. The first is when they are long and specific: a safety program for a particular kind of site, or a quality procedure tied to a regulated process. The second is when they sit inside a linked set, so a model can see the policy, the procedure that carries it out, and the checklist the team uses.
Why do manuals rank higher?
A field service manual written by your own engineers over several years is close to the opposite of a template. It is long, uses consistent terms, handles exceptions and is specific to your equipment and your customers’ problems. On micro1’s page, several years of records count in a company’s favor, and a manual’s revision trail is exactly that kind of history.
The caution with manuals is ownership. A manufacturer’s manual you keep on the shared drive is the manufacturer’s. Only what your team wrote counts.
Where do playbooks fit?
In the middle. A sales or support playbook shows how people handle real situations: objections, escalations, refunds, angry customers. That is valuable material. It is also full of real customers, real deals and real quotes from calls, and those have to come out. A playbook that survives redaction with its examples rewritten as roles (“a mid-size distributor,” “the account owner”) keeps most of its value.
What matters more than the document type?
Four things, in roughly this order:
- History. Versions and edit dates over several years, not a single snapshot.
- Structure. Links between pages, consistent templates, owners and labels. Scale AI’s interest in “documents and files with metadata” is a direct hint.
- Depth. Specialist work described in detail. Scale’s page asks for domain depth in close to those words.
- Clean rights. Written in-house, not for a client, not copied from anyone.
Language matters too. micro1 says data mainly in English is what it prioritizes.
Is there evidence that text licensing pays at all?
Yes, though mostly from publishers rather than operating companies. The publisher Wiley put its AI licensing income at $40 million for fiscal 2025, the year to April 30; in fiscal 2024 it was $23 million. That is published books and journals, a very different asset from a company’s SOPs, but it shows that buyers pay for long, edited, specialist text.
For operating companies, the useful benchmark is the buyers’ own published ranges, read as ceilings. The document value check shows which ones your size qualifies for. If you are deciding what to sell company documents from, start with the manuals and the wiki, not the policy folder.
Value hides in the document someone kept updating long after anyone asked them to.