Five AI data programmes compared: what they buy and how they pay
In 2026 at least five companies started publishing what they will pay businesses for their working records. The headline figures look similar. The terms underneath don't.
Each of the programmes below buys company operational data (email, chat, tickets, CRM history, documents, code) to train or test AI systems. The figures are what each company publishes on its own website. They are advertised ranges, not offers, and none of these programmes publishes the prices of completed deals.
micro1
- Its data partnerships page advertises $200k–$2m+ (its referral page says $100k–$2m+ for approved data packages).
- It looks for companies with 30+ employees, mature operations and documentation mostly in English. Demand is strongest for companies in the US, UK and Canada.
- Data is licensed only to labs and enterprises in the US and allied countries.
- After you transfer data, micro1 has 30 days to review it; payment follows within 30 days of your invoice.
Handshake AI
- Launched its employer data programme in April 2026. Partnerships typically range from $100k to $4m, depending on the data.
- It asks for 20+ full-time employees and 3+ years in operation.
- Everything is scoped upfront, covered by an NDA and de-identified before use.
Labelbox Prism
- Smaller, targeted datasets typically start around $50k; large partnerships can reach $1m+.
- You receive an upfront licence fee at signing plus a revenue share on downstream use.
- Partnerships are exclusive by default, though other structures are possible and priced by the buyer.
- Every licence has a kill clause forcing deletion within 30 days for material breach, a regulatory change or a change-of-control trigger you define.
- HR records and individual employee identifiers are excluded by default; Prism removes personal data when the data is exported.
Mercor
- Publishes no range. Its pitch is simple: the more tools you connect, the higher the payout.
- Connects to 100+ tools (Slack, Teams, Gmail, Salesforce, GitHub, Notion and others), then its pipeline extracts and de-identifies the data.
- Mercor covers extraction costs and pays by wire on delivery, typically 2–4 weeks after signing.
Turing (Project Lazarus)
- Buys codebases and operational data, mainly from start-ups, for a single payment of up to $1m.
- The figure is confirmed after sampling, usually within a week, and depends on how complete, rare and well-preserved the record is.
- Personal data and secrets are removed before anything reaches a model.
What the differences mean for a UK company
- Exclusivity. An exclusive licence can pay more, but it closes the door to every other buyer. See exclusive or non-exclusive?
- Payment shape. One payment, an upfront fee plus royalties, or payment on delivery: each spreads risk differently. We cover this in a separate guide.
- Who de-identifies, and where. Most programmes clean data on their side after export. The safer order is to de-identify inside your own systems first.
- Where the data ends up. micro1 limits buyers to the US and allied countries. Ask every buyer the same question.
- Referral fees. micro1 advertises up to $50k per referred company and Handshake $10k. If an adviser introduces you to a buyer, ask whether they're paid by that buyer.
Record Yield acts only for sellers. We compare terms across buyers and don't take money from them; anything a buyer pays on your deal is disclosed and passed to you. Our fee is 12% of what you receive (8% above £1m), and nothing if there's no deal. See pricing.
- These are advertised ranges, not offers.
- Compare exclusivity, payment shape and deletion terms, not just the headline.
- De-identify before anything leaves your systems.
- micro1, Company data partnerships (checked 30 Oct 2026)
- micro1, Refer a company (checked 30 Oct 2026)
- Handshake AI, Introducing employer data partnerships (April 2026)
- Handshake AI, License your data
- Labelbox Prism, FAQs (checked 30 Oct 2026)
- Mercor, Monetize your data (checked 30 Oct 2026)
- Turing, Project Lazarus (checked 30 Oct 2026)
What could your company's records earn?
A free, nine-question estimate. No files, no logins.
More in Market
Three months of the data market: what we've learned
A round-up of 40 posts: the prices, the rules and the patterns that matter most for sellers.
Who buys company data? The six types of buyer
From frontier labs to evaluation firms: who pays, what they want, and how they differ.
A $1.5bn settlement shows why licensed data is winning
The 2025 authors' settlement with Anthropic put a price on unlicensed training material. Buyers have noticed.