Data broker vs AI data licensing: how the two models differ for a company owner

What is the difference between a data broker and AI data licensing?

A data broker assembles and sells information about individuals, usually as lists or profiles. AI data licensing is a different transaction: a company licenses its own de-identified operational records, such as tickets, workflows and decision histories, to AI developers under a signed agreement, and keeps ownership.

The confusion is understandable because both involve "selling data." The legal treatment, the buyer, the product and the paperwork differ almost completely. This page separates the two so an owner can see which one a proposal actually is.

Side-by-side comparison

Does a company that licenses operational records need to register as a data broker?

Not automatically, and the answer depends on state law and on what is in the records. A few states, including California, run registries for businesses that knowingly sell personal information about people they have no direct relationship with. The usual test turns on whether the seller has a direct relationship with the individuals and whether personal information is being sold.

A company licensing its own de-identified process records is not building a list about strangers. But if a dataset still holds personal information about customers or consumers, privacy statutes such as the California Consumer Privacy Act text define "sell" and "share" broadly, and notice and contract obligations can follow. A first-party exemption in one state does not carry over to another.

Registry statutes and their definitions change, and this page does not cite a registry statute, so read the current text for each state where you operate. This is general information, not legal, tax or financial advice. Confirm with your own counsel which state registries and privacy laws reach your company before you sign anything.

Where the line sits: five questions

Use these to classify a proposal.

1. Whose records are they? Records the company generated while doing its own work lean toward licensing. Records about outside individuals lean toward brokering.
2. Is the value in the people or the process? If a buyer wants names and contact fields, it is a broker deal. If a buyer wants the sequence of steps, exceptions and outcomes, it is a workflow dataset.
3. Who decides scope? In licensing, the company approves what is included and what is excluded. In brokering, the seller is often asked to hand over everything it can.
4. What happens to identifiers? De-identification and redaction requirements should be agreed in writing before work begins.
5. Where does the data end up? A license names permitted uses and a term. A list sale usually has no meaningful limit on resale.

Illustrative example: two proposals on the same desk

Illustrative and fictional. The owner of a regional logistics software firm receives two offers in the same month.

Offer A is from a firm that wants the customer contact table, with email addresses and job titles, for resale to marketers. Offer B asks for a scoped license of dispatch exception tickets and resolution notes, with names and addresses removed, for AI agent training under an exclusive term.

Offer A is brokering and raises registration, consent and contract-duty questions. Offer B is the licensing model. The firm would still run a rights review first, because customer contracts may restrict use of ticket content.

When each model fits

A broker-style sale fits a business whose entire product is audience data and which already holds the registrations, consent records and downstream contracts that model needs. That is a different company from the mid-sized operating business this page is written for.

Licensing fits an operating company with several years of documented work across many systems, rights to the material and an owner willing to consider an exclusive license for an agreed term. Its value to buyers comes from the connected sequence of work and outcomes, not from who the individual people are. A company whose only asset is a customer list should not expect a licensing outcome and should get legal advice before any list sale.

What this means for a company considering a license

Three habits keep a licensing project out of broker territory.

• Inventory systems before anything else. The data inventory builder helps list each system, its history and what can be exported, without describing any confidential record.
• Exclude consumer personal data and protected health information from scope unless there is a clear licensing basis or authorization.
• Keep the buyer side clean. The comparison of selling directly or through an intermediary explains who handles buyer vetting, and the page on what lowers data value shows why personal-data-heavy sets are weaker products anyway.

For the wider trade-offs, see the pros and cons of licensing company data to AI developers and the ethics of licensing company data. The how it works page shows the seven-step process: introduction, qualification, inventory, price and terms, buyer review, close and delivery.

When neither model is the right fit

Do not proceed when the data mainly belongs to clients who have not consented, when it is mostly consumer personal data with no licensing basis, when it is mostly protected health information without authorization or de-identification, or when the archives no longer exist. A company under 50 full-time employees at peak (contractors excluded) is also outside the current baseline.

Next step

If you know a US company whose records are its own and well documented, register as a partner and make an introduction. Companies can also apply directly at sourcex.si/apply. Partners introduce and give basic fit information only; they never export, upload or describe confidential records.