AI data licensing glossary

Acceptance window

The period after delivery when the buyer checks the data against the agreed description before the final payment is due.

Aggregated data

Data combined into totals or statistics so that no single person or record can be picked out.

AI lab

A company that builds and trains large AI models. Labs license data directly or through data companies.

Anonymous summary

A one-page description of your data, covering types, years, volume, outcomes and rights, shared with buyers without your company’s name.

Consistent placeholder

A stand-in that replaces the same person or identifier everywhere it appears, like [CUSTOMER_41], so the record still makes sense.

Data company

A business that licenses data from many owners, prepares it and supplies it to AI labs.

De-identification

Removing or replacing details that identify people, and generalizing rare ones, so a person can’t reasonably be identified.

Deletion request

A person’s request, under privacy law, to have their information deleted. The license sets how these reach the buyer.

Evaluation data

Data used to test how well a model performs, rather than to train it. Often licensed separately.

Exclusion list

The written list of systems, folders, fields and record types that never leave your company.

Exclusivity

A promise not to license the same data to anyone else for a set period, often around 24 months.

Field of use

The specific purposes or industries a license covers. A narrow field of use leaves the rest open for other buyers.

Ground truth

A verified real-world result, like a final invoice next to an estimate, that lets a buyer check whether a model got it right.

Linked records

Records that reference each other across systems, such as a ticket tied to its chat thread, code change and CRM note.

Milestone payment

A payment triggered by a stage of the deal, such as signature, delivery or acceptance.

Non-exclusive license

A license that lets you license the same data to other buyers at the same time.

Outcome

What happened in the end: resolved, won, lost, approved, rejected or paid. Outcomes make records far more useful.

Perpetual use

Permission to keep using the data, or models trained on it, with no end date. Common, because training can’t be undone.

Personal information

Any information that identifies, or could reasonably be linked to, a person. Also called PII.

Provenance memo

A document delivered with the data recording where it came from, who owned it, what was removed and what rights apply.

Pseudonymization

Replacing identifiers with codes that could be reversed with a key. Weaker than de-identification on its own.

Re-identification

Working out who a person is from de-identified data. Licenses ban it by contract.

Referential integrity

Keeping the links between records intact after de-identification, so threads and histories still connect.

Refresh

A later delivery of new records under an existing license, usually paid separately.

Reinforcement learning environment

A simulated task or workplace where an AI agent practices and is scored. Real company records help labs build realistic ones.

Rights check

The review of ownership, contracts and privacy promises that decides what can be licensed.

Sample

A small, de-identified extract you approve so a buyer can test the data before signing.

Success fee

A broker’s fee paid only when a deal closes, as a percentage or fixed amount agreed in advance.

System of record

The main tool where a kind of work is recorded, such as your help desk for support or your CRM for sales.

Training data

Examples a model learns from. Company records teach models how real work is done.

Wind-down

The process of closing a company. Its records can often still be licensed, usually as a one-time archive.

Guides