Sign in See if your company qualifies

Questions

Questions about licensing business data to AI, answered

Quick answer

Each page gives a short answer first, then the explanation, examples, the rights and privacy points that apply, and sources.

100 pages

AI training data

Do AI companies buy private business data?Yes, some do. AI developers license non-public business data when it fills gaps that public web data cannot: specialist workflows, expert conversations, domain documents and labelled outcomes. Purchases are normally structured as licenses with defined permitted uses, and buyers generally require proof that the seller has the right to share the data and that personal information has been removed or lawfully handled.
What is RLHF data?RLHF data is the human input used in reinforcement learning from human feedback. It usually includes demonstrations (people writing ideal responses) and preference comparisons (people choosing the better of two model outputs). The preferences train a reward model, which then guides further model training. The method was described in research by Christiano et al. (2017) and applied to language models by Ouyang et al. (2022).
Structured vs unstructured AI training dataStructured data is organised in defined fields — tables, transactions, CRM records. Unstructured data has no fixed schema — emails, documents, chats, audio, images and video. Language and multimodal models rely heavily on unstructured data, which also carries more privacy risk because personal details can appear anywhere in the text. Structured data is easier to de-identify and validate but conveys less context.

Data licensing

Trust center

Can anonymized data be licensed?Yes. Data that is genuinely anonymised — so individuals can no longer be identified by reasonably likely means — generally falls outside data-protection law such as GDPR, which makes it far easier to license. The difficulty is achieving that standard. Free text, audio and rich records can often be re-identified, so many datasets are better described as de-identified or pseudonymised, which may still be regulated.

Data Value Index

How do AI companies value datasets?AI companies value a dataset by how much it is expected to improve a specific model or product, relative to what it costs to obtain and prepare. Key factors are relevance to the target task, scarcity, quality and labelling, volume and recency, clarity of rights, exclusivity, and the privacy and cleaning work required. There is no public price list; valuations are negotiated deal by deal.

Enterprise data

Sales & CRM data

Can CRM data be licensed?Parts of it can. CRM systems hold contact records, deal histories, notes, emails and call logs. The contact details themselves are personal data and are rarely licensable. The more valuable and more licensable parts are de-identified sales processes: how deals progress, what was said in notes and calls, and which actions led to outcomes — useful for sales assistants and forecasting models.

Human-generated data

What is human-feedback data?Human-feedback data is information in which people evaluate or correct AI or other outputs: ratings, pairwise preferences, rankings, edits and written critiques, plus demonstrations of the correct answer. AI developers use it to align models with human judgment, most notably through reinforcement learning from human feedback (RLHF). Businesses sometimes hold naturally occurring feedback, such as reviewer edits or QA scores.

Proprietary data

Public data vs proprietary data: what's the difference for AI?Public data is openly accessible — websites, published papers, government records — and is widely used to pre-train AI models. Proprietary data is held privately by an organisation, such as internal documents, customer interactions and operational records. For AI developers, proprietary data is scarcer and often more specialised, so it can improve domain performance in ways public data cannot. It must be licensed from the holder.

AI evaluation data

Training data vs evaluation data: what's the difference?Training data is used to teach a model — adjusting its parameters — while evaluation data is held back to measure how well the model performs. Training sets are usually larger; evaluation sets are smaller but must be high quality, representative and kept separate from training to give honest results. Both can be licensed, and evaluation licenses often restrict use to testing only.

How licensing works

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies Talk to SourceX

See if your company qualifies