How to sell company data to AI: licensing your records step by step

Sleeping Asset acts for you, the seller: we map your records, run each buyer's application with you and compare the outcomes.

The process at a glance

On our homepage these six steps are grouped into four: Assess (steps 1 and 2), Match (step 3), Clear (step 4) and Settle (steps 5 and 6).

What happens on the first call?

The first call is a free, confidential 30-minute conversation about whether licensing makes sense for your company. You leave knowing the likely fit, the buyers worth approaching and exactly how we are paid.

We ask about headcount, years of trading, your systems, the language of your records, and whether any records belong to clients or contain health data. These answers decide most of the fit: micro1 looks for 30 or more full-time staff and several years of records, mainly in English¹; Scale AI prefers 40 or more employees³. We also ask whether anyone has already registered with a buyer, because that changes how referrals work.

If the fit is weak, we say so. Not every company qualifies, and buyers decline some applications²; see what your data is worth. If your accountant or another adviser suggested licensing, they can read how accountants and advisers introduce a client.

How do we map your records without seeing them?

We map your records from metadata only: which systems exist, what they hold, the date ranges and rough volumes. You fill in a records inventory, often called a data manifest, and no content, sample or password is shared at this stage.

Buyers use the same idea: micro1 asks a referred company to complete a "data manifest"², and Cognor Labs describes its manifest as a 10 to 15 minute form with no files needed¹¹. Our inventory also flags the records that need care before any buyer sees them.

The records inventory we go through with you on the call

The example row is illustrative, and estimates are fine: we complete the rest with you on the call. What buyers look for in each system is explained in what AI companies buy.

Which buyers do we approach, and how do their rules shape the process?

We shortlist buyers whose published criteria fit you, run each buyer's own application with you and put the outcomes side by side. Buyers control parts of this process:

• micro1 requires the company to register itself through a referrer's unique link and complete its own profile and data manifest. Manual introductions and applications submitted on a company's behalf "do not replace the official application process", and an earlier registration cannot be re-attributed². micro1 then reviews the company and runs an initial AI interview¹. Its FAQ says demand is strongest in the United States, the United Kingdom and Canada, while also saying it prioritises US companies¹. We ask each buyer to confirm in writing that it accepts UK companies before we introduce you.
• Scale AI pays a referral only if it signs a partnership, and only one referral per deal, to whoever introduced it to the right stakeholder³.
• Mercor handles outreach, agreement and data pipeline itself once a business is referred⁴.
• Polyshares, Mode and SourceX buy the licence themselves and supply AI labs^6,7,10.

So register nowhere before we agree a shortlist. We say this partly for our own reasons: a buyer such as micro1 will not pay us a referral fee if your company has already registered with it². Registering directly costs you nothing, and the choice is always yours. Applications run in your company's name and buyers deal with you directly; we stay alongside you to prepare answers, compare offers and check terms. The current list, with checked dates, is on the buyers page.

How is personal and confidential data removed, and where?

Personal data is removed or transformed, and sensitive records are left out entirely. Where removal happens differs by buyer, and it matters: we prefer it to happen before records leave your systems. Where a buyer removes personal data after transfer, we tell you in writing first and your own solicitor confirms the transfer route before anything moves.

Removing names is not always enough. The ICO says pseudonymised data is still personal data, and that anonymisation is itself processing that needs a lawful basis¹². If raw personal data goes to a US recipient first, you need a valid transfer route, such as a recipient certified to the UK Extension of the Data Privacy Framework or an international data transfer agreement¹³. We prefer routes that remove personal data before records leave your control, either in your own environment or by a UK processor under contract to you. micro1 and Mercor de-identify after transfer^1,4; if you choose one of them, we set out that route in writing before you agree to it, and your solicitor confirms it.

We recommend excluding HR and grievance files, health information, payroll and client files before any buyer sees them. Our standard is on the anonymisation page; the legal background is in our UK law guide.

What does the licence say, and who approves it?

The licence defines exactly what the buyer gets and may do. Your own solicitor approves it before you sign; we flag commercial terms worth negotiating but do not give legal advice. Terms that vary between buyers:

• Scope: which systems and date ranges are included. Mode lists approved systems and date ranges before any export⁵.
• Rights: SourceX checks before signing whether you own the records, including limits in customer and supplier contracts¹⁰; Polyshares' licence excludes what you hold on behalf of clients⁷.
• Exclusivity: Labelbox Prism says its licences are exclusive by default and that extra exclusivity is reflected in the price the buyer pays⁹; Cognor Labs says most agreements include an exclusivity period, typically 24 months¹¹; SourceX can take exclusive AI training rights for a term set in the deal¹⁰.
• Deletion: Labelbox Prism describes a clause forcing deletion within 30 days on defined triggers, but models already trained "cannot be unlearned"⁹.
• Residual risk: SourceX's agreement records that re-identification may still be possible because no method is perfect¹⁰.

Our clause-by-clause checklist is in the data licence agreement guide.

When does the money arrive?

The buyer pays you under the licence, usually after it has reviewed and accepted the data. Published terms range from about two weeks after signing to 60 days after an invoice^4,10, and prices are quoted in US dollars, so agree the currency and who bears conversion costs in the licence.

Our introduction or referral fee is paid by the buyer, not deducted from your payment, and every amount is disclosed in writing before you sign. See how we are paid.

How long does the whole process take?

Plan for weeks to a few months. These are buyers' own claims, not guarantees, and rights checks, solicitor review and telling staff add time.

• Mode: "about three months from the first conversation through payment"⁵, including 15 to 30 days to collect and de-identify the data once you accept the final offer⁶.
• Labelbox Prism: claims "first call to first wire in under 21 days", with a written quote within 48 hours of scoping⁸.
• Polyshares: an offer within one to two weeks of intake⁷.
• Cognor Labs: an offer typically 48 to 72 hours after the data manifest¹¹.
• micro1: says there is no guaranteed timeframe between application and partnership².

What can go wrong, and what happens then?

Most problems surface before anything is exported. If a deal stops, you keep your records and pay nothing.

• No offer: we tell you why where we can, and whether another buyer fits better.
• A rejected sample or delivery: buyers review data before paying^1,7,10. The licence should say what happens to your data if it is rejected.
• A rights problem: client confidentiality or supplier contracts may take a system out of scope. Better found in step 2 than after signing.
• Staff concerns: workplace messages raise fair questions from employees. Tell them early; see Slack, Teams and email archives.
• A lost referral: if you registered with a buyer before working with us, that buyer may not pay us a fee². You still pay nothing.
• Changing terms: buyer programmes change often, so every buyer fact here carries a checked date. Terms are explained in the glossary.