What do the terms in a data licensing offer mean?

A data licensing offer grants a buyer the right to use a copy of your company’s records for stated purposes, usually training and testing AI models, in return for money. Its terms set the price, when it is paid, who else may license the same records, what the buyer may do with them and for how long.

What is data licensing, and how is it different from selling data?

Data licensing is permission to use a copy of records you keep, on written terms, while you go on using the originals. A sale transfers the records themselves.

• License. You keep your records and grant a buyer defined rights over a copy. Polyshares, for example, says you keep “the right to keep using it exactly as before”. Most programs for operating companies describe a license.
• Sale. Ownership moves to the buyer. Sales are more common when a company is closing: SimpleClosure’s terms say that when it or an affiliate acquires a company’s digital assets, “all right, title, and interest in and to the Digital Assets vests in the acquiring entity”.
• Licensor and licensee. The licensor is the company granting the license (you). The licensee is the party receiving it, which may be a data company that licenses the records onward to AI labs.

Am I selling my data or licensing it? covers what happens to the copy when a deal ends.

Who are the parties in a deal?

The company that pays you is often not the AI lab that ends up using the data. Each provider in the buyers table carries one of the first six labels below.

• Licenses directly. An AI data company that signs with you, pays you, and licenses or delivers the records to AI labs itself. Polyshares says, “We buy it. Polyshares is the counterparty on the deal.”
• Licensing agent. A company that licenses your data to its buyers on your behalf and passes on a share of each sale, as Troveo describes its business-data program (Troveo).
• Introducer. A company that matches you with a separate data partner, which signs with you and pays you. Avelence says the partner it selects “handles detailed eligibility and commercial terms directly with your company”, and its terms say Avelence “may receive compensation from a data partner for a successful introduction”.
• Listing marketplace. A site where you list a dataset or codebase for AI buyers to find. On HUD’s DataVendor, vendors set their own prices and buyers pay through its checkout, and the seller “receives 80% of the final payout, with 20% covering grading and packaging”. FileYield’s terms say it “is not a party to any transaction between sellers and buyers”, so the buyer pays you directly.
• Wind-down platform. A service built for companies that are closing, which can include selling or licensing their data. Sunset says it will buy a closing startup’s dataset and “assume responsibility for licensing it to downstream partners”.
• AI company, direct. An AI company that pays you itself, with no data company in between. Google’s Content Offer Pilot invites partners into an offer process where, it says, “we mutually agree on a fee”.
• End buyer. The AI lab or company that uses the data. Many providers do not name them; Polyshares, for one, says it “controls where the material goes and does not disclose which labs receive it”.

What do “indicative” and “firm” mean in an offer?

An indicative offer is a range worked out before anyone has looked at your records; a firm offer is a price for named records on written terms, made after review.

• Indicative offer. A figure from a calculator or a short profile. Nyne labels its estimate “indicative and not an offer”, and Avelence says its result is “not annual revenue, a single-license quote, a minimum payment, or a buyer offer”.
• Firm offer. A price tied to specific systems, date ranges, permitted uses and a payment trigger. Polyshares gives the order: “We do not pay for a record we have not seen, so the review comes first.”
• Due diligence. The buyer’s review between the two: a sample, a check that you have the right to license the records, and a look at how far back they go.

How do data licensing offers work? explains why two buyers price the same company differently.

How can you be paid?

Payment comes as a fixed amount, as a share of what the buyer earns from your data, or both, and the payment trigger decides when the money arrives.

• Upfront payment. A fixed amount for the license. License My Data, for one, says an “exclusive buyout pays once, upfront”.
• Revenue share. A percentage of each later sale of your data. Replay says its offers are “typically structured as an upfront cash payment plus a perpetual revenue share”.
• Payment trigger. The event that makes payment due: signing, delivery, approval or acceptance. Appen pays “on acceptance rather than submission”, so ask who decides acceptance and what happens to records that are not accepted.
• Net terms. The number of days after the trigger before payment. Polyshares pays net 30 to 60 days once the data has been approved, shared and anonymized.

Upfront payment or revenue share? weighs one against the other.

What do exclusive, non-exclusive and permitted uses mean?

Exclusivity decides who else may license your records, permitted uses decide what the buyer may do with them, and the term sets for how long. Together they can change what an offer is worth more than the headline price does.

• Exclusive license. Only this buyer gets the records, and you may not license them to anyone else while the exclusivity lasts. Polyshares’ data-handling page says its licenses are “exclusive by default unless we agree otherwise”, while one of its answers pages describes a “bounded, non-exclusive license”, so check which one your contract says.
• Non-exclusive license. You may license the same records to others. License My Data says a non-exclusive licence “can be placed with several buyers and paid on each”.
• Permitted uses. What the buyer may do with the data. Handshake AI limits use to “model training and evaluation only”, and Telegraph Lab says its agreement “identifies authorized recipients and permitted uses, including any access by frontier AI labs”.
• Scope. Which systems, date ranges and record types the license covers. Miro Advisory’s FAQ says “Scope, duration and exclusivity are negotiated, not assumed”.
• Term and perpetual license. The term is how long the license lasts; a perpetual license has no end date. Whatever the term, Frankfurt Kurnit notes that once records are in a trained model, “it cannot practically be pulled back out”.
• Exit terms. What happens to the data, and to payments, if either side ends the deal. Troveo says its agreement “includes clear exit terms”.

What do you hand over, and what do you sign for?

Before a firm offer you usually describe your records and show a small redacted slice of them; the contract then asks you to confirm they are yours to license.

• Data inventory. A short list of your systems, how far back each goes, roughly how much each holds and who can export it. See what a data inventory is.
• Data sample. A small, representative, de-identified slice of your records that lets a buyer check the inventory. It can also show you what would be delivered: Appen says “You review a representative sample pack before anything is licensed.” See what a data sample is, or build one with the sample tool.
• Retention period. How long a system keeps records before deleting them, which sets how far back your archive goes. The Slack, Google Workspace and Microsoft 365 pages show the defaults by plan.
• Representations. Statements you make in the contract that the buyer relies on. Replay says “You’ll be asked to represent that the data is yours to license”.
• Sign-off. Your approval of what is delivered. Scale AI lists “Your sign-off on every data package”.

What do anonymized and de-identified mean?

Both describe removing the link between records and the people in them; anonymized is the stronger claim, and the one to question.

• De-identification. NIST defines it as “any process of removing the association between a set of identifying data and the data subject”. Names, emails, phone numbers and account numbers are replaced or removed.
• Anonymized. The claim that no one can link the records back to a person. FTC technology staff wrote in July 2024 that data “is only anonymous when it can never be associated back to a person”.
• Re-identification. Linking de-identified records back to a person or company. Telegraph Lab says its process “reduces risk; it cannot guarantee that re-identification is impossible”, and License My Data says “re-identification is prohibited by contract”.
• Confidential content. Removing names does not make a pricing spreadsheet or a dispute less sensitive. Frankfurt Kurnit’s note that de-identification “addresses whether a record can be linked to an individual, not whether its content is confidential” is why exclusions matter as much as redaction.

What does “anonymized” mean, and could it be traced back to us? goes through each provider’s description of its process.