Industries AI companies license data from, and what stays off limits

Which sectors do AI buyers target?

Nearly all of them. Buyers publish long lists of target industries, and several say any sector with a long operating history can qualify.

Mercor names eight target industries, from legal and healthcare to logistics and real estate¹. Scale AI lists 22, including property management, staffing, freight brokerage and field services such as plumbing and electrical work². Ventra lists construction and trades, wholesale and distribution, and staffing and call centres among the industries labs are buying from⁴. Polyshares buys from "any business with real operating history, in any sector"⁸.

Some buyers show stronger demand for particular sectors. Mode accepts law firms from 6 staff and accounting, finance and medical firms from 10, against 30 for other businesses⁶, and its press release says "AI labs are prioritizing legal and accounting data"⁷. Labelbox Prism marks software and internet data as "urgent" and financial services, healthcare and manufacturing as "high demand"¹⁰.

Sector is not what sets value, though. Polyshares says "the length and texture of the record matter more than headcount or revenue"⁸. What AI buyers want across sectors is covered in our page on what AI companies buy.

Sector matrix

This matrix shows, for each sector, the records buyers name, what UK rules usually keep out, the number of UK firms with 20 to 249 staff, and where buyers signal interest.

Counts are ONS figures for March 2026, rounded by the ONS to the nearest 5; the groupings and totals are ours (see the last section)²⁰.

Healthcare and care: why patient records are out of scope

Patient records are out of scope for Sleeping Asset. Some operational records may not be, if no patient can be identified from them.

UK GDPR prohibits processing "data concerning health" unless one of a short list of conditions applies¹⁴. In England, the Caldicott Principles cover confidential information wherever patients "can be identified and would expect that it will be kept private", and principle 8 asks for "no surprises" about how it is used¹⁶. Licensing patient information to train commercial AI fits none of that easily.

Buyers draw the same line. Endless Data lists care coordination, referral workflows and scheduling as in scope but excludes "protected health information, entirely"⁵. Handshake AI asks healthcare firms for "operating procedures, training materials, administrative workflows"³.

What might be licensable is a provider's own procedures, training material, rota and scheduling workflows, purchasing and IT tickets, once every patient and staff identifier is removed. Free text is the danger: an IT ticket or email can name a patient. If you deliver NHS-funded care, read your NHS contract and any data sharing agreements before offering any records for licensing, and take specialist health data advice.

Finance and insurance: client data and regulated firms

A financial firm can usually license how it works, not what it holds about clients. Client records carry personal data, confidentiality duties and, in insurance, some of the most sensitive data categories.

In 2020 the FCA, ICO and FSCS warned firms that "terms, conditions and clauses within a standard contract are highly unlikely to constitute sufficient legal consent" for passing client data on¹⁷. Insurance files can also hold health data, which needs an Article 9 condition¹⁴, and data about criminal convictions, which may be processed only under official authority or where the law authorises it¹⁵.

Buyers' own lists point to the licensable part: "financial models, process documentation, underwriting workflows"³, and for insurance agencies quoting, policy servicing and renewals, with "policyholder personal data" stripped⁵. Prism describes financial reasoning traces, the assumptions and rationale behind underwriting or valuation decisions, as urgently wanted¹⁰. Involve your compliance officer early, because regulated firms may have further obligations under their own rules.

Property management and estate agency: tenant data and maintenance tickets

Maintenance tickets, contractor correspondence and lease workflows are among the most licensable property records, once tenant details are removed.

Scale AI, Ventra and Endless Data all name property management^2,4,5. Endless lists "work orders, leasing and renewals, owner reporting, rent collection, vendor management, SOPs", with "tenant and owner personal data" stripped⁵. Mercor's real estate list adds leases and inspection reports¹.

Tenants and private landlords are individuals, and information that relates to an identifiable person is personal data¹⁸. A repair history attached to a flat number can identify a tenant even without a name, so addresses need the same care as names (see our anonymisation standard). Check your management agreements as well: some records may belong to the landlords you act for.

Professional services: client deliverables versus your own know-how

Your firm's own methods, templates and project management records can usually be licensed. Work you produced for clients, and what clients told you, usually cannot.

Several buyers blur this line. Mercor lists "client decks, SOWs, project files" as examples¹, while Endless Data excludes "client-confidential findings and client names" from consulting records⁵ and Polyshares' licence "does not cover what you hold on behalf of your own clients"⁸.

For accountants, ICAEW advises assuming "all unpublished information about a client's or employer's affairs, however gained, is confidential"¹², and warned in July 2026 that before client information is "uploaded, summarised, analysed or reused" firms must be sure they comply¹³. Solicitors must keep clients' affairs confidential unless the law permits disclosure or the client consents¹¹; see our law firms guide.

Copyright in templates and methods your staff wrote belongs to your firm, "subject to any agreement to the contrary"¹⁹, so check client contracts that assign rights in deliverables.

Construction and engineering

Construction is one of the few sectors with published deal evidence. Polyshares says it paid a general contractor USD 300,000 for a perpetual exclusive licence to its operating record (a company claim)⁹.

Ventra seeks "architectural plans, building designs" and other CAD libraries alongside operational records⁴, and Mode lists AutoCAD among the sources it buys⁶. Scale AI's interest in field services such as plumbing and electrical work covers many contractors².

The UK limit is ownership. Copyright in a drawing belongs first to its author, or to the author's employer if made in the course of employment¹⁹, so plans from architects, engineers or clients are not yours to license. Homeowners' names and addresses in job records are personal data¹⁸.

How many UK companies are in each sector?

About 50,930 UK firms with 20 to 249 staff are in the sectors AI data buyers name most often, out of 130,545 across the whole economy. Adding construction, wholesale, residential care and social work, and leaving out information services, gives about 77,860, the wider grouping in the table below. The method is on our UK market figures page²⁰.

Source: ONS Table 3, UK, March 2026; the economy-wide total is from Table 5; groupings and sums are ours²⁰. The ONS does not split the 20 to 49 band, and many buyers start higher: micro1 at 30 full-time staff²¹, Scale AI prefers 40² and SourceX 50 at peak²². So fewer firms than shown will qualify. Which buyers take UK firms, and at what size, is in our UK buyer comparison.

Choose your sector on the valuation form and we will tell you which records in it are worth licensing and which to leave out. We reply within one working day, and sellers pay nothing: see how we are paid.