Home / Resources

Owner’s resource

How to sell company data to AI developers

Begin with a description of your records and the rights behind them. A buyer conversation should come before any data transfer.

From existing records to a paid license

The first step is a description. Each later stage has its own decision and deliverable.

  1. 1

    Describe

    Record types, dates, formats and business context.

    A company summary
  2. 2

    Evaluate

    Match a buyer’s use case with records you can license.

    A defined scope
  3. 3

    Agree

    Set the rights, preparation, acceptance and payment terms.

    A signed license
  4. 4

    Deliver & get paid

    Deliver the agreed package and meet the payment milestone.

    An accepted delivery

What “selling data” actually means

In a company-data conversation, “selling” often means licensing defined uses of selected records. The commercial agreement determines what the buyer can do, for how long, and whether you can license the same material elsewhere. It should also distinguish the source records from any derived datasets, evaluations or trained models.

This resource is for authorized owners and company representatives. It is not a route to sell information from an employer without permission, consumer lists, passwords or records copied from another business. Possessing a file is not the same as having the right to license it.

What an AI buyer can learn from company records

A procedure describes the intended way to do a task. A project history can also show what happened when the task did not go to plan: the information available, the options considered, the correction and the result. Those are different kinds of material. Start by distinguishing instructions, examples of completed work and the history connecting the two.

For example, a final proposal contains an answer. The brief, rejected draft, reviewer comments and accepted revision may explain how that answer was produced. A buyer might be interested in training on examples, testing whether a model completes a task correctly, or constructing an environment in which a model uses tools. Ask which use is intended. The same folder is not automatically suitable for all three.

This also explains why the number of systems you use is a weak description on its own. Saying ‘we use email and a project tool’ tells a buyer little. Saying ‘a project reference connects the brief, decisions and approved deliverable’ describes an actual collection that can be evaluated.

Inventory record families before preparing an export

Choose one coherent record family first. Note where it lives, who knows the system, what time period is readable and which other records it depends on. If an export drops comments or attachments, record that limitation. Do not buy a cleaning tool or commission a full extraction before the buyer explains the required format and scope.

Record familyContext to look forFirst inventory question
Project historiesBriefs, changes, approval decisions and final outcomesCan one project be followed from start to finish?
Knowledge and proceduresVersions, effective dates, exceptions and reviewer rolesCan you distinguish the instruction in force at the time?
Issue and support historiesProblem description, attempted fixes, escalation and resolutionDoes the closed status explain how the issue was resolved?
Structured operational recordsRelated tables, keys, event times and status changesCan the relationships be explained without exposing actual records?
Expert work productsInputs, assumptions, revisions and review commentsWhich parts belong to the company, a client or a third party?

A useful first description, with an example

A first description should let someone ask the next sensible question. Include the business activity, record family, approximate coverage, available format, connections between records and known exclusions. Keep estimates labelled. A count should have a unit: cases, documents, events or projects are not interchangeable.

For example: ‘We retain completed project histories in a project system and document archive. The collection includes briefs, revisions and approval notes linked by project reference. Current records are readable; older attachments have gaps. We can describe the schema and approximate coverage. Client-owned material and personal information need a separate review.’ This is more useful than ‘we have a lot of valuable data’ without disclosing a single project.

You do not need every answer to make an initial enquiry. Identify which points the owner can answer and which need an operations lead, system administrator or contract reviewer. The goal is to establish a plausible scope, not to declare the collection ready before anyone has examined the requirements.

What to settle before a buyer evaluates a sample

Write down the answers before the sample leaves your controls. If a connector is proposed, have your technical owner review its access scope, logging, revocation and exclusions. A claim that a connector is read-only answers only one part of the question: it can still permit copying. This website does not receive samples or connect to your systems.

  • Purpose: what question will the sample answer, and who will evaluate it?
  • Scope: which record types, fields, periods and exclusions are agreed?
  • Access: is a limited export sufficient, or is a connector being proposed? What permissions would it request?
  • Use: is sample use limited to evaluation, or does the proposed agreement also permit training?
  • Acceptance: what constitutes an acceptable package, who decides and how are defects handled?
  • Cost: who does extraction, documentation, de-identification and any rework?
  • End of evaluation: what happens to the sample if neither side proceeds?

Look for records that explain real work

Start with a single repeatable workflow. A work order connected to an inspection result and corrective action can describe a process more clearly than a folder of unrelated PDFs. That does not establish buyer demand; it gives a buyer something specific to evaluate.

Describe the connections before counting files: which record identifies the job, where alternatives were discussed, which revision records a decision, and how the outcome is linked. Use record types and generic identifiers in your description, not customer names or copies of the discussion. If the links cannot be recovered, say so.

  • Process documentation: procedures, work instructions and version histories.
  • Decisions and outcomes: exception handling, inspection results and resolved quality issues.
  • Linked operational records: work orders, maintenance activity and production events.
  • Domain context: a data dictionary, units, timestamps and explanations of missing fields.

Look for the judgment already recorded

A final document may show what happened without explaining why. In your own inventory, look for records of alternatives, constraints, exceptions and the reasoning behind a decision. A rejected approach or an unsuccessful intervention can help explain an eventual outcome when the records genuinely connect them.

Start with familiar questions: why was a tool changed, how was a quality issue resolved, and what evidence informed the decision? Mark each answer as documented, partly documented or unknown. Experience that lives only in someone’s memory is different from an existing record. Do not reconstruct missing history and present it as contemporaneous evidence.

Historical records and new expert work are different

Licensing a historical collection concerns records that already exist. Commissioning an expert to demonstrate a task, judge model outputs or explain a new decision creates new work, with separate questions about time, instructions, permissions and payment. An interactive evaluation or reinforcement-learning environment is different again: it lets a system take actions and receive feedback rather than simply read a fixed collection.

Do not treat a request for one as an agreement to provide the others. Establish what is being requested before discussing scope. These resources focus on understanding existing company records; this website does not commission expert labor, build training datasets or operate evaluation environments.

Prepare a nonconfidential summary

Write down the record types, broad date range, approximate volume and export formats. Identify an internal decision-maker and anyone who must review customer contracts, privacy, security or intellectual property. Keep the initial summary at category level.

Useful in an initial inquiryKeep inside your company
“Five years of quality inspection logs”Actual inspection files or customer part numbers
“CSV exports and scanned PDFs”System logins, API tokens or live database access
“Rights review still required”Customer names, employee records or NDA-protected detail

Use the first conversation to test fit

Ask the prospective buyer which use cases it is actually scoping. Explain the exclusions you expect to need. A responsible next step may be a narrower inventory, a controlled sample under an appropriate agreement, or a decision to stop. Those steps are for you and the buyer to arrange directly.

  • Who is the contracting entity, and who will receive or use the licensed material?
  • What rights are requested for training, evaluation, sublicensing and derived work?
  • What must happen before payment becomes due?
  • Who performs and verifies any redaction, and what happens to rejected records?
  • What does exclusivity cover, and how does it affect future licensing?

Keep the commercial milestones separate

An inquiry starts a discussion; it is not an offer. These are separate decisions, and either side may stop before an agreement. The buyer’s written terms determine its own sequence.

StageWhat it establishes
Inquiry & initial fit reviewA category-level description for deciding whether further discussion is useful.
Consented introductionYour permission for one named recipient and specified contact details and metadata.
Buyer diligenceThe buyer evaluates the proposed scope; your reviewers assess the buyer, rights and safeguards.
Negotiation & agreementBoth parties resolve permitted uses, obligations and commercial conditions.
AcceptanceThe agreed criteria are met and recorded; this may differ from signing or delivery.
PaymentFunds are actually received under the agreed terms. A referral fee has its own conditions.

Where this site fits

We provide independent explanations and an initial fit inquiry. We do not buy data, operate a marketplace or hold raw datasets. We have no signed buyer partnerships. The public program resource is a starting point for your own review.

If an introduction becomes appropriate, you will be asked to approve a named recipient and the exact contact details and metadata to share. An inquiry here is not consent to send your details to every listed company.

Keep reading

Explore the earning potential of your existing data.

Tell us what your business already creates. Start with a description; keep the records in your own systems.

Explore your opportunity