What makes business data licensable for AI?

Plenty of organisations hold data that AI developers would value: years of maintenance records, inspection reports, technical files, support conversations or expert decisions. But whether that data can actually be licensed for AI use depends on more than its usefulness. It depends on rights.

This guide sets out the questions that usually need answering before a licence can be agreed. It is general information, not legal advice: the answers depend on the specific data and the proposed use, and each party should take its own advice.

1. Who owns the data, and who can authorise its use?

The starting point is whether the organisation holding the data has the right to license it. Records created by a business in the course of its own operations are often, but not always, under its control. Data created jointly with customers, generated on a client's systems or received from partners may come with ownership or usage terms attached.

It also matters who inside the organisation can make the decision. A first conversation may be with an operations lead or engineer who knows what exists, but licensing will normally need sign-off from someone with authority over the organisation's data and contracts.

2. Does it contain personal data?

Business records frequently contain personal data: names in work orders, customer contact details, staff identifiers, free-text notes about individuals, or images showing people. Where it does, data protection law, including UK GDPR for UK organisations, applies to any further use, including providing it for AI training.

In practice this often means removing or transforming personal data before licensing, assessing whether there is a lawful basis for the proposed use, or excluding certain records altogether. Whether de-identification is sufficient depends on the data and the context. It shouldn't be assumed.

3. Is anything owed in confidence?

Contracts with customers, suppliers and partners commonly include confidentiality obligations. Records that describe a client's assets, processes or commercial terms may be covered even when they sit in the data owner's own systems.

Reviewing relevant contracts early avoids discovering late that a significant part of a dataset can't be shared, or can only be shared in aggregated or redacted form.

4. Does it include other people's intellectual property?

Technical document collections often mix material an organisation wrote itself with material it obtained elsewhere: manufacturer manuals, standards, licensed software output or third-party reports. The organisation may be entitled to use those internally without being entitled to license them onward.

5. What exactly is the permitted use?

Licensing data “for AI” is too vague to be useful to either side. Training a foundation model, fine-tuning a narrow model, building an evaluation benchmark and powering retrieval are different uses with different implications. A clear licence defines which uses are allowed and which are excluded, alongside restrictions on redistribution, retention and security.

• Which records, fields and time periods are included
• Which uses are permitted, and which are excluded
• How long the rights last, and where they apply
• Whether the licence is exclusive
• What must happen to the data at the end of the term

Answering the questions before the deal

None of these questions makes licensing impossible. They do make it important to work through them before delivery, not after. For data owners, that protects customers, staff and the business. For AI teams, it avoids investing in data that later turns out to be unusable for the intended purpose.

That is the part of the process Haselpin helps organise: establishing provenance, permitted use and restrictions alongside availability, so that both sides can agree terms with a clear view of what is being licensed.