How to Sell Data to AI Companies: A Guide for UK Businesses

AI developers have used up much of the easily available public text on the internet. To train and test systems that can do real work, they increasingly want records from real businesses: how work was quoted, planned, changed, delivered and checked. For UK businesses, especially those closing or winding down, this creates a new question. Can we sell data to AI companies, and should we? This guide explains how the market works and what to check before you start.

Why AI companies want business data

Modern AI systems are being built to act as assistants and agents inside companies. They are asked to answer questions from a firm's documents, draft quotes, follow up orders and handle exceptions. To learn and be tested on these tasks, developers need examples of the real thing: messy, linked records that show how decisions were made and what happened next. Real business data provides that context in a way generated examples struggle to match.

What AI companies look for

Linked workflows

An enquiry, its email thread, the quotation and revisions, the order, any problems and the final outcome. The connections matter as much as the documents.

Technical records

CAD and BIM files with revision histories, specifications, test results and quality records. See our guide to CAD data.

Communication archives

Email, Microsoft Teams and Slack, with personal data handled properly. See exporting Slack data.

Project-software histories

Tasks, tickets and status changes from tools such as Asana, Jira or Procore. See Jira export and project data.

Known outcomes

Records where the result is recorded: won or lost bids, final costs against estimates, passed or failed inspections.

How deals are usually structured

Most arrangements are licences rather than outright sales. The licence defines:

the data covered and any exclusions
the permitted use, such as training, evaluation or both
whether it is exclusive or non-exclusive
how personal data is removed or pseudonymised
security, transfer and deletion terms
payment, whether a fixed fee, staged payments or a share of later revenue

Buyers include AI labs, companies that build training and evaluation datasets, and intermediaries who assemble data from many sources.

What to check before you sell data to AI companies

Authority

Who controls the data? A trading company acts through its directors. In an insolvency, the office-holder decides. See insolvency data.

Personal data

UK GDPR applies. Most buyers expect names and identifiers to be removed or replaced. Read anonymising data before a sale.

Third-party rights

Client contracts, NDAs and customer-owned designs can restrict use.

Reputation

Consider how customers and former staff would view the use of the data.

Advice

Take independent legal advice before signing.

Realistic expectations

Not every archive has a buyer. Demand changes as developers' needs change, and published prices are rare. Treat any figure you see with caution, and judge your own archive on its linkage, depth, specialism and rights. Our data valuation guide sets out these factors.

Preparing your data

Keep systems running, or take full exports with metadata.
Write a non-confidential inventory of systems, date ranges and sizes.
Note how records link, for example by job numbers or client codes.
Identify restricted and personal material.
Agree internally who will make decisions.

How McKays helps

McKays helps UK owners, retiring founders and liquidators find out whether their digital records have a responsible AI licensing route. We start with a description only, never files. If we don't see a route, we say so. Browse the sector guides or see for advisers.

Types of buyer and what each wants

AI labs

Large developers who train foundation models. They tend to want substantial volumes, consistent formats and clear rights, and their legal requirements are often strict.

Dataset and evaluation builders

Companies that assemble training data, benchmarks and simulated work environments for AI developers. They may want smaller, specialist sets with known outcomes that can be used to grade a model's answers.

Data intermediaries and marketplaces

Businesses that license data from many owners and resell or relicense it. They may pay a share of later revenue rather than a fixed fee.

Vertical software companies

Firms building AI tools for a single industry, such as construction estimating or engineering quality control, that need sector-specific examples.

Each type has different requirements for format, volume, rights and price. A description of your records helps match them to the right kind of buyer.

The process, step by step

Description: you describe the systems, date ranges and kinds of records, with no files.
Initial view: an assessment of whether there is plausible demand and what the obvious rights issues are.
Confidential review: under an NDA, a closer look at structure and samples, usually with personal data removed.
Preparation: anonymisation, filtering and packaging into an agreed format.
Licence: negotiation of terms and independent legal review.
Transfer: secure delivery, followed by confirmation of deletion or retention as agreed.

At each stage it is reasonable to stop if the risks outweigh the benefits.

FAQ

Can a small business sell data to AI companies?

Possibly. Some buyers set minimum sizes, while others value specialist knowledge regardless of headcount. Suitability depends on the records.

Do I lose ownership of my data?

Under a licence you normally keep ownership and grant a defined use. An outright sale or assignment would transfer ownership.

Will my customers' details be shared?

They shouldn't be. Personal data is usually removed or pseudonymised before transfer, and the licence should prohibit re-identification.

Is it legal?

Licensing business data can be lawful if you have authority, respect contracts and comply with UK GDPR. Take advice.

What data is not suitable?

HR files, health information, payment card data and material you don't have the right to license.