How to sell or license your company's data to AI companies

Most companies don't sell their data outright. They license a copy of their records, usually de-identified, to a data company that uses it to train and test AI. The process runs from a short intake form to an estimate, a contract, an export, and payment once the buyer accepts the data. Depending on the buyer, that takes from a couple of weeks to about three months.

The harder part is choosing a buyer. More than a dozen programs now license company records, and each has its own criteria, process and terms. This guide covers who qualifies, how the process works, what to check first, and how to compare buyers.

What AI companies are buying

AI data buyers want the records of how your company actually works. micro1 lists "internal communications, SOPs, project histories, CRM records, support tickets, and other business documentation." Handshake AI's list of systems runs from Gmail, Slack and Google Drive to NetSuite, QuickBooks, ServiceTitan and AutoCAD.

AI labs want these records because they are training agents to do office work, and the public internet doesn't show how that work gets done. We explain why in Flight simulators for office work.

Buyers say variety matters more than volume. Nick Heiner of Surge AI told MIT Technology Review that one company's Slack history, if it captures conversations with variety and depth, can be worth more than another company's Slack, email and calendars combined.

Who buys company data

Most of this buying runs through data companies, not the AI labs themselves. Companies such as micro1, Mercor, Handshake AI and Scale AI license records from businesses and turn them into training and testing material for the labs. How the market for company data works explains who pays whom.

Our buyer index tracks 16 of these programs. They differ more than their websites suggest: in who they accept, which records they want, who handles your data, and how they pay. For an owner, comparing them is most of the work.

Does my company qualify?

Most programs that publish a size floor set it between 20 and 40 employees, though some go lower for certain industries. Most want several years of records spread across several systems. Several, including micro1, Nyne and Mode, favor US companies, and micro1 wants records mainly in English.

Sources: each program's own page (micro1, Handshake AI's program and license pages, Scale AI, Nyne, Mode, AfterQuery), checked October 31, 2026.

A closed company isn't necessarily out. AfterQuery welcomes closed companies, The General Data Company takes companies "that are no longer trading," and Turing's Project Lazarus acquires from teams "winding down or already closed." The question then is who has the authority to sign.

How the process works, step by step

Each buyer names and orders the steps its own way, but the sequence is similar.

1. List your systems. Write down every tool your company has used and for how long. Breadth raises the price: Mercor tells owners that "the more tools you share, the higher your payout will be." Include the systems people forget, such as accounting software, ticketing, project tools and code repositories.
2. Check your rights. Holding a record isn't the same as being free to license it (more in the next section). As Nyne puts it, "Your ability to export a record does not by itself establish permission to license it."
3. Apply. Intake ranges from a form to a call. micro1 starts referred companies with an AI interview. AfterQuery sends a mutual NDA and a short questionnaire, and replies within three business days. Nyne starts with a 15-minute scoping call.
4. Get an estimate. An estimate isn't a final price until the buyer has seen the data. Handshake AI says it can give an initial review and pricing within 12 to 24 hours once it has the details it needs. AfterQuery proposes terms only after reviewing a sample. Mode, in its October 13 announcement, describes "a 14-day, nonbinding letter of intent" that holds its estimate while it assesses the data.
5. Agree the terms in writing. The contract sets which systems and date ranges are included, how long the license runs, whether it's exclusive, who may receive the data, and how you're paid. Mode agrees "the systems and date ranges before any export." SuperAnnotate lets you "approve each" buyer.
6. Export and de-identify. This step decides who touches your data, and where. Mercor has you link your accounts, then extracts and anonymizes automatically. Scale AI anonymizes "everything inside your environment, before any data is extracted." micro1 receives the data through its APIs "or other approved methods" and de-identifies it itself. Turing works from a copy in an air-gapped environment.
7. Get paid. Most buyers pay on delivery or once they've accepted the data; some pay in milestones, and Labelbox Prism pays part at signing. micro1 has 30 days to review and approve the data, then pays within 30 days of your invoice. Mercor pays by wire on delivery, which it says typically comes two to four weeks after the signed agreement.

How long the whole process takes, as each buyer describes it:

What should I check before I apply?

Check your rights over the records before you share anything, including a sample. Daniel Goldberg of Frankfurt Kurnit, writing in Legal 500 in October 2026, sets out the main questions:

• Other companies' information. "Email and chat contain other companies' confidential information, often covered by NDAs and commercial agreements."
• Your own privacy promises. Companies "must comply with the privacy statements in effect when the data was collected."
• Your employees. Employee messages about pay, scheduling and working conditions may be protected under Section 7 of the National Labor Relations Act, and "a disciplinary thread or pay complaint remains sensitive without a name attached."
• Your software providers' terms. Goldberg advises reviewing "platform terms for the systems the data will be extracted from." Slack's API terms, for example, bar providers of apps offered outside their own organization from using data obtained through Slack's APIs to train a large language model. Ask each buyer how it will extract your data.
• What to leave out. He suggests excluding high-risk material, such as privileged documents, HR and personnel files, and payroll and tax records, rather than relying on de-identification. Some buyers already do: AfterQuery says payroll, privileged legal material and credentials "are never used."

He also warns against sharing sample data before this review, "since a sample itself could be a disclosure." This is a summary of one lawyer's published guidance, not legal advice. Your own counsel should review your situation.

What am I agreeing to?

You usually keep ownership and grant a license. micro1 says companies "retain ownership of their underlying data," and SuperAnnotate describes "licensing rights for a set period, not selling the data." Four terms decide what the deal is worth to you, and they differ from buyer to buyer:

• Exclusivity. Labelbox Prism structures licenses as "exclusive by default," and says it "can also support different exclusivity structures." Mode's agreement defines any exclusivity, so it's set deal by deal.
• Who receives the data. SuperAnnotate lets you approve each buyer and block competitors. The General Data Company has you decide in writing who may receive the results.
• How you're paid. Turing makes one payment. Labelbox Prism pays a fee at signing plus a share of later use. SuperAnnotate pays on delivery and again each time the data is relicensed.
• What happens afterwards. Many buyers delete the raw data after an agreed period, but training can't be undone. As Goldberg puts it, "once data is used to train a model, it cannot practically be pulled back out."

How do I choose a buyer?

Choose on two things: the value of the whole deal, and where your data goes.

Value. Start with the programs whose criteria you clear and which name the records you actually hold. Then compare the terms behind each estimate, not just the number: how long the license runs, whether it's exclusive, when you're paid, and whether you're paid again if the data is relicensed.

Where your data goes. Ask each buyer:

• Who extracts the data, and does anything leave your systems before it's de-identified?
• Who receives the data, and can you approve or block particular recipients?
• What is excluded, and what is deleted, and when?

Share a description of your records before you share any samples, and only after your rights review.

Our view: the terms and the handling decide more of a deal's value than the first number does. Two offers can look alike until you compare them side by side. Owners who compare several programs before committing to one are better placed than those who accept the first estimate.

Where we come in

This is the work we do. Company Data Advisors helps business owners find the right buyer for their records and get the most from the deal:

• We know who's buying what. We track every major program's criteria and terms, and we have direct contacts at the major data buyers and at AI labs. That tells us which buyers are likely to value your records most right now.
• We compare offers on the whole deal. Price, exclusivity, how long the license runs, payment timing and onward use, side by side, so you can see which offer is worth the most to you.
• We check where your data would go. Who handles it, where it's processed, who receives it, and what's deleted afterwards.
• You stay in control. Nothing about your company is shared with any buyer until you approve that buyer by name, and whether to proceed is always your decision.

Owners pay us nothing. Participating buyers may pay us referral fees, and we tell you whether a buyer pays us before any introduction. You can also apply to any program directly; our buyer index links to each one.

What to do next

1. Write down every system your company has used, and for how many years.
2. Note anything you may not be free to license: client material under NDA, privileged documents, HR and payroll records.
3. Answer seven quick questions to see which programs your company matches, or check the criteria above.
4. Talk to your counsel before you share a sample with anyone.