The work matters more than the industry
AI developers are training and testing models on multi-step office work: closing the books, marking up a contract, resolving a ticket, dispatching a truck. A January 2026 Epoch AI post, based on interviews with 18 people at companies that build and use AI training environments, found coding is still a major source of demand and enterprise workflows are a major growth area. Some environments are built as copies of real apps, such as team chat or a spreadsheet.
That's why we sort records by business function first and by industry second. Most operating companies hold several of these, whatever they sell:
- Finance and accounting: month-end close, reconciliations, payables and receivables, approvals, forecasts.
- Legal, contracts and compliance: contract requests, redlines, signed agreements, policies.
- Software engineering and IT: code history, tickets, incidents, service-desk records.
- Customer support: tickets, escalations and how each case was resolved.
- Sales and CRM: pipeline history, proposals, notes tied to won and lost deals.
- Operations, dispatch and field service: work orders, schedules, exceptions, job photos.
- Procurement and supply chain: purchase orders, supplier threads, inventory changes.
- HR and recruiting: process records only, such as onboarding steps. Personnel files stay out.
- Email, chat and project tools: the layer that ties the rest together.
That last layer matters, because it's often where the reasons behind the other records were written down.
Industries where these records pile up
Any company that has run these functions in business software for years can hold a useful record. Among the sectors where that's common: software and internet companies, managed IT services, marketing and other agencies, e-commerce and retail, insurance brokerage, law firms and in-house legal teams, accounting firms and company finance teams, manufacturing and distribution, energy and utilities, equipment rental, freight brokerage and 3PL, field services such as HVAC, plumbing and electrical, professional services, property management and hospitality. Business Insider reported that law and accounting, among the fields AI developers most want to automate, "see especially high demand."
To be plain about it: we also take trades and property companies, such as roofers, solar installers and builders, because their records have the shape described above, with work orders, dispatch, calls and photos tied to outcomes. That's our reasoning, not a buyer's stated demand, and the same goes for any industry on this list. Regulated sectors such as healthcare, financial services, telecom, staffing and education are covered in the next section. The industries page lists the industries we feature and the others we accept.
Industries that need special handling
These hold records buyers want, under rules that come first. We accept them, flag them at intake and set scope with your counsel:
- Healthcare and life sciences. Patient information must be de-identified to HIPAA's standard or covered by patient authorizations before it leaves, and a vendor holding it for a provider may use it only as its business associate agreement allows.
- Banking, lending and investment firms. Under Regulation P, information that doesn't identify a consumer, such as aggregate or blind data without names, addresses or account numbers, isn't personally identifiable financial information. Anything that identifies a customer stays out.
- Telecom carriers. Individually identifiable customer proprietary network information (CPNI) needs the customer's approval for uses beyond providing the service. Aggregate customer information is treated differently.
- Education. For a use like this, student records can be released without consent only once identifying information is removed and the holder reasonably determines no student can be identified, counting other available information. Small classes make that hard.
- Staffing and recruiting. Candidate and employee files are personal data almost line by line, so process records with people removed are the realistic start.
- Contact centers and BPOs. Recordings made for clients usually belong to the client under the service contract, and call-recording consent still applies.
- Injury and personal-lines insurance claims. Bodily-injury and workers' compensation files carry medical details, so they're handled like health records.
- Law and accounting firms' client work. See can law firms and accounting firms license their records.
What isn't a fit
- Income tax return information. A tax preparer may not sell even anonymous statistics built from returns, so we leave it out entirely.
- Defense and government contractors' controlled data. Controlled unclassified information may be shared only in furtherance of a lawful government purpose, and releasing export-controlled technical data to a foreign person, even inside the U.S., is itself an export that can need a license.
- Public-sector agencies. We don't broker government bodies' own records.
- User content from consumer apps and social platforms. Posts, chats and profiles are people's content, not a company's operating record.
- Media archives. Film, music, image and publishing catalogs are content licensing, a separate market.
Where to start
If your company runs on business software and has several years of linked history, the record matters more than the industry. The intake asks for your industry, systems and counts, and flags special handling up front. Buyer interest varies by industry and by record, and we can't promise any bid.
This is general information, not legal advice. The rules above are summaries; the sources below have the full text.
Follow-up questions
Sources
- Epoch AI, Gradient Updates, "An FAQ on reinforcement learning environments" (Jan. 12, 2026): coding as a major source of demand, enterprise workflows as a major growth area, environments built as copies of apps (opens in a new tab)
- Business Insider, "AI labs are fueling a wild market for your emails and Slack DMs" (Oct. 13, 2026): law and accounting see especially high demand (opens in a new tab)
- 45 CFR 164.514, HIPAA de-identification standard (eCFR) (opens in a new tab)
- 45 CFR 164.508(a)(4), authorization required for a sale of protected health information (eCFR) (opens in a new tab)
- 45 CFR 164.504(e), business associate contracts (eCFR) (opens in a new tab)
- 12 CFR 1016.3, Regulation P definitions, including information that doesn't identify a consumer (Cornell LII) (opens in a new tab)
- 47 U.S.C. 222(c), customer proprietary network information (Cornell LII) (opens in a new tab)
- 34 CFR 99.31(b)(1), FERPA release of de-identified records (Cornell LII) (opens in a new tab)
- 26 CFR 301.7216-2(o), tax return preparers may not sell statistical compilations (Cornell LII) (opens in a new tab)
- 32 CFR 2002.16(a)(1), dissemination of controlled unclassified information (Cornell LII) (opens in a new tab)
- 15 CFR 734.13(b), release of technology or source code to a foreign person in the U.S. is a deemed export (Cornell LII) (opens in a new tab)
- 22 CFR 120.50(a)(2), release of technical data to a foreign person in the U.S. under the ITAR (Cornell LII) (opens in a new tab)
This is general information about how data licensing works. It is not legal advice, and we are not your lawyer. Have your own counsel review any license before you sign.